Shroom speak

Monitoring moods
in the classroom
What we learned from
learning to code

Volume 126 May/June

Number 3 2023

AI
is coming for
the classroom.
Meet the teachers
who believe it could
improve
education.
 S P ON SOR ED CON TEN T

Building the backbone

for innovation, speed,
and thriving humanity
“New IT” is an evolution built on smart devices,
edge and cloud computing, 5G networks, and AI.

F
.+)Ƶ ġ,+3!.! ,(0"+.)/0$0 Ƶ Č.!2+(10%+*%6!$+3!*0!.,.%/!/1/! 0Č remotely in real time using telepresence
can detect abnormal activities support business agility, and confront and teleoperation and learns tasks as it
in supermarkets, to edge servers climate change with sustainable goes. The robot is operated by a stream-
helping preserve biodiversity in remote solutions. ing augmented reality headset with 3D
locations, today’s technologies drive video to give the user a realistic view of
innovation in ways never before imag- THE FIVE ELEMENTS OF NEW IT the work being done. The user’s head
%*(!ċė **+20%+*/!.2!/0$!,1.,+/! Although new technology and powerful position controls the robot arm, and a
of making our life better, our work more applications are constantly emerging, handheld device controls movements.
productive, and our planet more sus- !*+2+% !*0%ü!/ü2!'!5+),+*!*0/
tainable,” says Yuanqing Yang, CEO +""101.!ġ.! 5 !*2%.+*)!*0č/).0 Edge computing helps data eliminate
and chairman of Lenovo. devices, edge computing, cloud com- boundaries: Processing volumes of
puting, high-speed networks such as 5G, 0*(! 0+,!."+.)*!%//1!/ċ *
Technology leaders are reimagining * Ƶ ċ$%/ !ü*%0%+*.!/+*0!/3%0$ .!/,+*/!Č)*5+.#*%60%+*/.!01.*%*#
an infrastructure where multiple technical leadership too, says Yang, citing to edge computing, which processes
technologies join to spur innovation in a 2022 Lenovo global research study of data close to the source to enable fast
a secure, compliant, and user-friendly ĆĀĀ$%!"0!$*+(+#5+þ!./%*3$%$ and real-time analysis and response,
environment. Long gone are the days "+1.+10+"ü2!ƫ/#.!!%0ė,01.!/ while maintaining privacy and security
+"ė0. %0%+*( * %0/(%!*0 !2%!/Č and describes the future of information requirements. “Edge computing allows
servers, data centers, and on-premises +))1*%0%+*/0!$*+(+#5Ĩ ƫĩ data to be treated closer to where data
applications,” says Yang. He says tra- ‘extremely’ or ‘very well.’” is generated—directly at the edge site,
%0%+*( Č/$+.0$* "+.ė%*"+.)0%+* lowering latency for faster response
technology,” is being replaced by what Smart devices connect AI to human times, increased agility, and greater
!*+2+((/ė*!3 ČĘ+.ė%*0!((%#!*0 problems: Ƶ+. %*#0+00%/0Č0$! resilience,” says Yang.
transformation.” Yang explains that *1)!.+"%*0!.*!0+"0$%*#/Ĩ +ĩ
“The new IT enables digital transfor- devices worldwide will reach 29 billion For example, Kroger, one of the larg-
)0%+*/! +*ü2!'!5!(!)!*0/č 5ĂĀăĀċ +Ě/!4,+*!*0%(#.+30$ģ !/0#.+!.5$%*/%*0$!*%0! 00!/Č
smart devices, edge computing, cloud smart devices empowered by advanced 0!)! 3%0$!*+2+* 2%/1(Ƶ 
computing, high-speed networks, sensors—provides a wide range of technology provider Everseen to build
* .0%ü%(%*0!((%#!*!ċ$%/*!3 industries with competitive advantages. a system of secure self-checkout kiosks.
IT architecture can create countless Ƶ /!.2!./,01.!1*/0.101.!  00
opportunities.” Manufacturers can use smart devices like each checkout from 20 high-resolution
robots to stand in for workers in danger- cameras. The system detects if an item is
This technology paradigm promises ous or remote workspaces, and accel- not scanned, and prompts the customer
to support innovation and boost em- erate and automate assembly lines. For 0+.!/*ċ 0*(/+,%*#*//+%0!Ě/
ployee productivity, and also to power example, Lenovo’s Daystar Robot works )+%(! !2%!ċ%*!0$%/.!-1%.!/
 S P ON SOR ED CON TEN T

enormous computing power, an edge dents optimistic about the future of !*+2+!/0%)0!/0$!Ƶ /+(10%+*
solution processes the data near the hybrid cloud. %),.+2! +. !."1(ü(()!*05ĂĀŌ
source. “Over 75% of checkout errors and productivity by 18%.
can be corrected without employee 5G networks enable innovation and
intervention,” says Yang. flexibility: Connecting the essential com- Consider that a single PC order will
,+*!*0/+"*!3 .$%0!01.!.!-1%.!/ launch a series of complex tasks across
Ƶ* #(+(%+ %2!./%05*+*,.+ü0 /(*  "/0Č!þ%!*0Č* 1/0+)%6(!*!03+.'- multiple production lines, and requires
Conservation uses edge computing to ing. The answer: 5G—the next generation alignment of thousands of parameters,
.% #!ąĀĀ)%(!/+"%ü!*ċƵ0 of mobile wireless voice and data com- such as employee schedules, materials,
+%*/+*ƫ.1/+! /(* Č+*!+"0$!)+/0 munication technology. The 2022 Lenovo production processes, and equipment
remote places on Earth, it uses camera study also found that 72% of CTOs see statuses. Lenovo’s largest manufacturing
traps to document endangered and opportunities for their companies to use base for PCs, LCFC Electronics, pro-
invasive species. Camera data used to be 5G multiaccess edge computing (MEC) cesses up to 690,000 orders per year.
stored on a hard drive and periodically even more with the demand for hybrid While accounting for these large-scale
ý+3*0+*0%#+Čƫ$%(!Č0+,.+!//Č0'- options dominating the workplace. “The calculations is a challenge for people, an
ing as long as three months. Today, edge popular hybrid work model that many Ƶ !*#%*!*!/%(5..50$!)+10Č* 
computing data centers process data on *ý!4%(5)'!.!(ġ0%)! &1/0)!*0/
the island—time-savings that can save "+..+ +.#.*1(.+&!0%2!/ċ$!Ƶ 
(%2!/ċė$! /(* ƫ+*/!.20%+*0!)* solution’s autonomous learning ability
process six months’ worth of visual data also means the more it operates, the
within just one week, enabling them to smarter it becomes. “This smart
draw analytical insights within min- solution has also improved
utes instead of weeks,” Yang says. !*!.#5!þ%!*5* .! 1! 

87 %
greenhouse gas emissions
Cloud computing provides con- by thousands of tons a year,”
nection: "0$!,* !)%01#$0 says Yang.
technology leaders anything, it’s
that public, private, hybrid, and A LOOK TO THE FUTURE
multicloud computing is imper- of CTOs agree these Technologies such as smart
5 elements capture and
ative for fast and agile services devices, edge computing,
describe the future
and development. of ICT architecture cloud computing, 5G, and
Source: Lenovo Ƶ .!"%(%00%*#/$%"0".+)
“Normally, we wouldn’t think of information technology to intel-
tablets as life-saving equipment, but ligent transformation. “New IT is
when emergency hospitals needed shaping the future in many innova-
to be built during the covid-19 out- tive ways,” says Yang. “In the future,
break, these devices and innovative the objects you work on, the colleagues
infrastructure played a critical role,” you work with, the environment you
says Yang. companies have adopted over the last work in, and the outcome you deliver
three years is only possible with a high- might all be real or virtual, ranging from
“In tough times, like the pandemic, it was speed network,” says Yang. AI assistants and digital twins to the
new IT that kept us connected, produc- metaverse.”
tive, and engaged.” He continues, “The AI tools mimic human intelligence to
public cloud became more popular by solve problems: By combining data, As always, while change surges ahead,
,.+2% %*#0$!ý!4%%(%05Č/(%(%05Č*  computing power, and sophisticated technology executives must careful-
on-demand accessibility that we needed (#+.%0$)/ČƵ *$* (!)1$)+.! ly consider the real-life outcomes of
at the time. But, many enterprise appli- data much faster than a human worker, !,(+5%*#*!3 %*"./0.101.!ċ!1.%-
cations and data are still running and can be adjusted by users to accommo- ty, compliance, and usability standards
stored in private cloud or on-prem data date change, can help users learn better must still be upheld. “Environmental,
centers. In fact, we will continue to see processes, and can help anticipate risks /+%(Č* #+2!.**!Ĩĩ#+(/)1/0
0$!+ġ!4%/0!*!+",.%20!Č,1(%Č*  such as cost overruns, accidents, and be a major consideration,” says Yang.
hybrid cloud for compute, storage, and maintenance needs. Using multiple ė *0$!"101.!Č!2!.5!(!)!*0+"*!3
network needs.” Ƶ 0!$*+(+#%!/* +,0%)%6! (#+- .$%0!01.!)1/0%*+.,+.0!
rithms, Lenovo Research created new ċ$!*5+1//!//0$!.!01.*/
The same Lenovo study found that processes for its manufacturing facility +*%**+20%+*Č%0Ě/*+0&1/0ü**%(
cloud, software, and computing are key that dramatically improved production payback but also social impact.”
components for the future of a hybrid planning processes, with some six-
work environment, with 84% of respon- hour processes cut to 90 seconds.
02 From the editor

Cheat codex
Welcome to the Education Issue. I cheated on my editor’s letter.
This one that you are reading right now. I’m very sorry. 
Look, I didn’t set out to do this, but the thing about maga-
zines is that they have very hard deadlines, and if you miss them,
you’re left with blank pages. So when I realized I only had a few
hours left to finalize this, well, I freaked out. 
And then I did what an increasing number of us are doing:
I turned to ChatGPT, OpenAI’s massively mind-blowing gen-
erative AI software, to help me out. After training it on some
of my previous work, I asked about the use of AI in education.

AI is already doing big things in education. By crunching Mat Honan

massive amounts of data on student performance, AI algo- is editor in
chief of
rithms can tailor instruction to fit the needs of individual
MIT Technology
learners, which can mean big improvements in student out- Review
comes. Chatbots and virtual assistants can provide students
with on-the-spot assistance and feedback. Who needs a tutor
when you have an intelligent assistant? 
But here’s where things get really exciting: language mod-
ChatGPT will have all sorts of genuinely useful and transfor-
els like ChatGPT can actually generate human-like text, which
mative applications in the classroom. Yes, they will almost cer-
makes them perfect for a whole bunch of educational applica-
tainly also be used for cheating. But banishing these kinds of
tions. These models can create interactive learning materials,
technologies from the classroom, rather than trying to harness
like chatbots that answer students’ questions or create person-
them, is shortsighted. Rohan Mehta, a 17-year-old high school
alized quizzes. They can even generate summaries of complex
student in Pennsylvania, makes a similar argument on page 20,
texts, reports on data sets, or entire essays and research papers.
suggesting that the path forward starts with a show of faith by
There are definitely some challenges as well. One of the
letting students experiment with the tool.
biggest concerns is the risk of bias in the algorithms. We need
Meanwhile, on page 26, Arian Khameneh takes us inside
to make sure these algorithms are designed and tested in a way
a classroom in Denmark where students are using mood-
that ensures all students get a fair shake. Another, of course,
monitoring apps as the country struggles with a huge increase
is the potential for cheating. 
in depression among young people. You’ll also find a story from
But the bottom line is that AI and language models like
Moira Donovan on page 34 about how AI is being used to help
ChatGPT are going to change the way we think about educa-
further our analysis and understanding of centuries-old texts,
tion, and we need to make sure we’re using these tools in ways
transforming humanities research in the process. On page 48,
that are ethical, equitable, and effective.
Joy Lisi Rankin dives deep into the long history of the learn-to-
So are the preceding four paragraphs, which were generated code movement and its evolution toward diversity and inclusion.
by ChatGPT and then lightly edited by me, ethical? If they were And please do not miss Susie Cagle’s story on page 64 about a
presented as my own work without an explicit disclosure (like California school that, rather than having students try to flee
this one), I would argue that the answer is no. And even with from wildfire, hardened its facilities to ride out the flames, and
such a disclosure, we’re still in a bit of a gray area—there are all what we can learn from that experience.  
sorts of questions about everything from plagiarism to accuracy Of course, we have a lot more for you to read, and hopefully
to the data these models were trained on.     think about, as well. And as always, I would love to hear your feed-
The reality is that we are in an entirely new place when it back. You can even use ChatGPT to generate it—I won’t mind.
comes to the use of AI in education, and it is far from clear what
that is going to mean. The world has changed, and there’s no Thank you,
ROBYN KESSLER

going back. 
As William Douglas Heaven, our senior editor for AI, makes Mat
clear in this issue’s cover story on page 42, technologies like @mat/[email protected]
04

“They don’t want it

to be vilified.
They want to be
taught how to use it.”
–p. 42

Front The education issue Back

2 From the editor 26 Calibrating the classroom 64 Living with wildfire

Denmark’s enthusiasm for educational technology Should we build places that
THE DOWNLOAD are easy to escape from,
is taking on a new frontier: children’s well-being.
8 The latest on CRISPR thera- or places that are easy
BY ARIAN KHAMENEH
pies; the UN’s urgent climate to defend? By Susie Cagle
to-do list; Maillardet’s automa-
ton; VR for incarcerated peo- 34 AI is transforming humanities research 70 Rocky journeys to the
land of mathematics
ple; tech that reads your mind; Historians are using neural networks to draw
There is no shortage of
super plants; a high school- new connections in the analysis of history. popular books and lectures
er’s defense of ChatGPT. Plus,
BY MOIRA DONOVAN on math—but they can take
this month’s job of the future:
us only so far.
embryo predictions marketer.
42 The education of ChatGPT By Pradeep Niroula
EXPLAINED The panicked reaction around cheating in school 76 Shift happens
22 How do fungi communicate? doesn’t tell the whole story. Meet the teachers who A designer’s obsession turns
Turns out they have a lot think AI could actually make learning better. into a fascinating history
to say. By Michael Hathaway of keyboards, covering 150
BY WILL DOUGLAS HEAVEN
and Willoughby Arévalo years from early typewriters
to the pixelated ones in our
PROFILE 48 What we learned from “learning to code” pockets. By Allison Arieff
24 The Nigerian university dropout Historically, learn-to-code efforts have provided
who builds EVs opportunities for the few; new efforts aim for greater 82 The lasting impact
Mustapha Gajibo electrifies of an online salon
inclusivity. BY JOY LISI RANKIN
communal buses to make In the late 1980s, Stacy Horn
them more accessible. launched ECHO as a place
By Valentine Benjamin
58 Teaching the biliterate brain for quirky, tech-savvy New
to read Yorkers to congregate online.
What’s the best way to teach kids who grow up tog- A few users never left.
By Nika Simovich Fisher
gling from book to screen and back again?
BY HOLLY KORBEY ARCHIVE

88 Screen time
Figuring out the best way
Cover
to integrate tech into the
illustration
classroom continues to be
by Selman
a challenge.
Design
06 Masthead

Editorial Corporate Consumer marketing MIT Technology Review Insights

Editor in chief Chief executive officer and publisher Vice president, marketing and and international
Mat Honan Elizabeth Bramson-Boudreau consumer revenue Vice president, Insights and international
Alison Papalia Nicola Crepaldi
Executive editor, operations Director of retention marketing
Finance and Operations Global director of custom content
Amy Nordrum Taylor Puskaric Laurel Ruma
Chief financial officer, head of operations
Executive editor, newsroom Director of acquisition marketing
Enejda Xheblati Senior manager of licensing
Niall Firth Alliya Samhat Ted Hu
General ledger manager
Editorial director, print Director of event marketing
Olivia Male Senior editor, custom content
Allison Arieff Nina Mehta Michelle Brosnahan
Accountant
Editorial director, audio and live journalism Email marketing manager
Anduela Tabaku Senior editor, custom content
Jennifer Strong Tuong-Chau Cai Kwee Chuan Yeo
Human resources director
Editor at large Circulation and print production manager
Alyssa Rousseau Editor, custom content
David Rotman Tim Borton Teresa Elsey
Manager of information technology
Science editor
Colby Wheeler Senior project manager
Mary Beth Griggs Advertising sales Martha Leibs
Data analytics manager
News editor Senior vice president, sales and
Christopher Doumas Project manager
Charlotte Jee brand partnerships
Natasha Conteh
Office manager Andrew Hendler
Features and investigations editor
Linda Cardinal [email protected] Director of partnerships, Europe
Amanda Silverman 201-993-8794 Emily Kutchinsky
Managing editor
Associate vice president, integrated Director of partnerships, Asia
Timothy Maher Technology
marketing and brand Marcus Ulvne
Chief technology officer Caitlin Bergmann
Commissioning editor
Drake Martinet [email protected]
Rachel Courtland Board of directors
Vice president, product Executive director, brand partnerships
Senior editor, MIT News Cynthia Barnhart, Cochair
Mariya Sitnova Marii Sebahar
Alice Dragoon Alan Spoon, Cochair
Senior software engineer [email protected] Lara Boro
Senior editor, biomedicine 415-416-9140
Molly Frey Peter J. Caruso II, Esq.
Antonio Regalado
Executive director, brand partnerships Whitney Espich
Associate product manager
Senior editor, climate and energy Kristin Ingram Sanjay E. Sarma
Allison Chase
James Temple [email protected] David Schmittlein
Digital brand designer 415-509-1910 Glen Shor
Senior editor, AI
Vichhika Tep
Will Douglas Heaven Executive director, brand partnerships
Podcast producer Stephanie Clement Customer service and
Anthony Green
Events stephanie.clement@ subscription inquiries
Senior vice president, technologyreview.com National
Senior reporters 214-339-6115
events and strategic partnerships 877-479-6505
Tanya Basu (humans and technology)
Amy Lammers Executive director, sales and brand
Eileen Guo (features and investigations) International
Jessica Hamzelou (biomedicine) Director of event content and experiences partnerships 847-559-7313
Melissa Heikkilä (AI) Brian Bryson Debbie Hanley
[email protected] Email
Tate Ryan-Mosley (tech policy)
Head of international and custom events 214-282-2727 [email protected]
Reporters Marcy Rizzo
Senior director, brand partnerships Web
Casey Crownhart (climate and energy)
Senior event content producer Ian Keller www.technologyreview.com/
Rhiannon Williams (news)
Erin Underwood [email protected] customerservice
Zeyi Yang (China and East Asia)
Director of events 203-858-3396 Reprints
Copy chief
Nicole Silva Senior director, brand partnerships [email protected]
Linda Lowenthal
Miles Weiner 877-652-5295
Event operations manager
Senior audience engagement editor [email protected]
Elana Wilner Licensing and permissions
Abby Ivory-Ganja 617-475-8078 [email protected]
Manager of strategic partnerships
Audience engagement editor Senior director, digital strategy, planning,
Madeleine Frasca Williams
Juliet Beauchamp and ad ops
Event coordinator
MIT Technology Review
Creative director, print Katie Payne
Bo Richardson [email protected] 196 Broadway, 3rd Floor
Eric Mongeon
Cambridge, MA 02139
Digital visuals editor Digital sales strategy manager 617-475-8000
Stephanie Arnett Casey Sullivan
[email protected] Our in-depth reporting reveals what’s
going on now to prepare you for what’s
Media kit coming next.
www.technologyreview.com/media
Technology Review, Inc., is an indepen-
dent nonprofit 501(c)(3) corporation wholly
owned by MIT; the views expressed in
our publications and at our events are not
always shared by the Institute.
08

studies of sickle-cell and a related disease, beta thalassemia, and

that the therapy could be approved for sale in the US within a

The year. It is widely expected to be the first treatment using CRISPR

to go on sale. Vertex hasn’t said what it could cost, but you can
expect a price tag in the millions.

Download
To scientists, CRISPR is a revelation because of how it can
snip the genome at specific locations. It’s made up of a cutting
protein paired with a short gene sequence that acts like GPS,
zipping to a predetermined spot in a person’s chromosomes.
Along with Vertex, a wave of biotech companies, like Intellia,
Beam Therapeutics, and Editas Medicine, are hoping they can
use this technology to develop successful treatments. Many of
them are running the trials on Liu’s list.
At the summit, Liu highlighted the case of Alyssa, a teenager
More than 200 people in the UK with a form of leukemia that affects a type of white
blood cell called T cells. Chemotherapy didn’t work, and neither
have been treated with did a bone marrow transplant. So doctors at Great Ormond Street
Hospital in London tried a CRISPR-based approach. It involved
experimental CRISPR taking healthy T cells from a donor and using CRISPR to mod-
ify them. The treated cells were altered so that they wouldn’t be
therapies rejected by Alyssa’s immune system, but they would be able to
track down and attack Alyssa’s own cancerous T cells. The edited
But exciting trial results are tempered by safety cells were then given to Alyssa as a treatment. It seems to have
and ethical concerns. worked. “As of now, approximately 10 months after treatment,
her cancer remains undetectable,” Liu said.
By Antonio Regalado But not all these trials will be successful. For instance, in
January the San Francisco biotech Graphite Bio had to stop its
There are now more than 50 experimental studies underway that own tests of a gene-editing treatment for sickle-cell after its first
use gene editing in human volunteers to treat everything from patient’s blood cell counts dropped dangerously. The problem
cancer to HIV and blood diseases, according to a tally shared was caused by the treatment itself. Graphite’s stock plunged and
with MIT Technology Review by David Liu, a gene-editing spe- now the firm’s future is in question.
cialist at Harvard University. The trick facing all these efforts remains getting CRISPR
Most of these studies—about 40 of them—involve CRISPR, where it needs to go in the body. That’s not easy. In Gray’s case,
the most versatile of the gene-editing methods, which was devel- doctors removed bone marrow cells and edited them in the lab.
oped only 10 years ago. One of the first patients treated using But before they were put back in her body, she underwent pun-
a CRISPR procedure, in 2019, was Victoria Gray. At the Third ishing chemotherapy to kill off her remaining bone marrow in
International Summit on Human Genome Editing, held in London order to make room for the new cells.
in March, her story left the room in tears. “I stand here before
you today as proof miracles still happen,” Gray said of her bat-
tle with sickle-cell disease, in which misshapen blood cells that
don’t carry enough oxygen can cause severe pain and anemia.
She described to the audience episodes that left her hospital-
ized for months at a time. Her children were worried she might
die. But then she underwent a treatment that involved editing
the genes in cells from her bone marrow. Her new “super cells,”
PHOTO COURTESY OF VICTORIA GRAY

as she calls them, have transformed her life. Within minutes of

receiving her transfusion of edited cells, she shed tears of joy,
she told us. It took seven to eight months for her to feel better,
but after that point, “I really began to enjoy the life that I once
felt was just passing me by,” she said.
The company developing Gray’s treatment, Vertex
Pharmaceuticals, says it’s treated more than 75 people in its
10 The Download

In essence, the Vertex treatment

requires a bone marrow transplant, an
ordeal in itself. Vertex thinks the treat- Book reviews
ment will be suitable for “severe” cases,
a market it estimates at 32,000 people in
Europe and the US. Even then, patients
Not Too Late: Changing the Climate Story from
won’t get the treatments if insurers and
Despair to Possibility 
governments balk at paying. It’s a real risk.
The first-generation CRISPR treat- Edited by Rebecca Solnit and Thelma Young
ments are limited in another way. Most Lutunatabua (Haymarket Books, 2023)
use the tool to damage DNA, essentially Hope isn’t a lottery ticket, says activist and writer
shutting off genes—a process that the Rebecca Solnit. Rather, it’s a tool, and an essential
Harvard biologist George Church has one for anyone living through climate change today.
famously described as “genome vandalism.” This essay collection from scientists, policymakers,
Treatments that attempt to break genes and activists balances laying out the very real stakes of
include one designed to zap HIV. Another the climate crisis with making the case for possibility. 
is the one Gray got. By breaking a specific
bit of DNA, her treatment unlocks a sec-
Saving Time: Discovering a Life Beyond
ond version of the hemoglobin gene. Since
the Clock
hemoglobin is the errant protein in sick-
le-cell, booting up another copy solves the By Jenny Odell (Random House, 2023)
problem. According to Liu’s analysis, two- “What is time?” we’ve all asked ourselves and each
thirds of current studies aim at “disrupting” other throughout the pandemic. For most of us the
genes in this way. Liu’s lab is working on question was rhetorical, but not for Odell, who urges
next-generation gene-editing approaches. us to seriously reconsider the ways we spend it. Not
These tools employ the CRISPR protein bad advice, but her observations on, say, the impor-
to deftly swap individual genetic letters tance of rest or the choke hold capitalism has on the
or make larger edits. These are known as hours in our collective days can feel more like a blend-
“base editors.” ing of other people’s ideas than a new way forward.
Lluís Montoliu, a gene scientist at
Spain’s National Center for Biotechnology,
Science Fiction: Voyage to the Edge
says these new versions of CRISPR have
of Imagination
“lower risk and better performance,”
although delivering them to “the right Edited by Glyn Morgan (Thames & Hudson, 2022)
target cell in the body” remains difficult. Science fiction, writes Nalo Hopkinson in this catalogue
Commercial viability remains one of for the London Science Museum’s 2022 exhibition,
the biggest limits on CRISPR’s impact “is the literature of social and technological change …
now and in the foreseeable future. Nearly it tests in order to discover the truth.” It upends our
all CRISPR trials underway aim at either brains visually, too, as the images herein will attest.
cancer or sickle-cell disease, with mul-
tiple companies chasing the exact same
Meganets: How Digital Forces Beyond
problems. According to Fyodor Urnov, a
Our Control Commandeer Our Daily Lives
researcher at the University of California,
and Inner Realities
Berkeley, this means thousands of other
PHOTOS COURTESY OF THE PUBLISHERS

inherited diseases that could be treated By David B. Auerbach (Public Affairs, 2023)
with CRISPR are just being ignored: “This Auerbach defines meganets as autonomous digital
is near-entirely due to the fact that most of forces that have brought an unprecedented level
them are too rare to be a viable commercial of chaos to our politics, economy, and social lives.
opportunity.” Q We might think that Microsoft, Google, et al. are in
control of these networks; Auerbach suggests that
Antonio Regalado is senior editor for bio- what chatbots might really herald are new modes of
medicine for MIT Technology Review. mass manipulation and deception. Q
12 The Download

But in the near term, there’s a clear path to the IEA, an annual investment of $11
forward for the emissions cuts needed to billion would be needed to clean up the
The UN’s put the planet on the right track. Here sector, but the value of the captured meth-
are some of the tasks with the lowest cost ane could be more than enough to cover
urgent climate and highest potential to address climate the cost.
change during this decade, according to
to-do list the new IPCC report.  Protect natural ecosystems
that trap carbon. 
Cheap and available tech- Deploy wind and solar power, The impacts of human-caused
nologies can help us meet and a lot of it. 
Cutting emissions in the near term 3  climate change “threaten our life
support system, nature itself,” said

1 
climate goals this decade.
will require shifting away from pol- Lee. Conserving and restoring natural eco-
Here’s how, according to the
luting fossil fuels for energy pro- systems will not only be key for preserving
new UN climate report. duction and toward renewable energy biodiversity—it’ll have emissions bene-
By Casey Crownhart sources like wind and solar power. The fits too. Natural ecosystems can trap and
scale of wind and solar deployment already store carbon, and tropical rain forests are
Time is running short to address climate underway is staggering: the world is set to among the planet’s largest carbon sinks.
change, but there are feasible and effec- build as much wind and solar capacity in Preserving these and other ecosystems
tive solutions on the table, according to a the five years between 2022 and 2027 as could be a low-cost, high-value way to
new UN climate report released in March. it did in the past two decades, according slow climate change. 
Only swift, dramatic, and sustained to the International Energy Agency.  Policies around the world are already
emissions cuts will be enough to meet the Plummeting costs have helped this helping to cut deforestation, according to
world’s climate goals, according to the new growth: between 2010 and 2019, the cost of the IPCC report. And in December 2022,
report from the Intergovernmental Panel solar energy fell by about 85%, the report over 190 nations signed a UN biodiver-
on Climate Change (IPCC), a body of cli- says. Wind energy costs dropped by about sity pledge to protect 30% of the natural
mate experts that regularly summarizes half during the same time frame. Now, world by 2030.
the state of this issue. wind and solar are among the cheapest
“We are walking when we should be energy sources available—deploying new Use energy efficiently in vehicles,
sprinting,” said Hoesung Lee, IPCC chair, solar and wind farms can be even cheaper homes, and industry. 
in a press conference announcing the than just maintaining existing coal power Shifting to public transporta-
report in March. To limit warming to 1.5 °C
(2.7 °F) above preindustrial levels, the
target set by international climate agree-
plants in the US. But as inexpensive as
wind and solar are, they can still repre-
sent a significant financial investment.
4  tion and biking for some travel
needs is an inexpensive way to
limit near-term emissions. And boosting
ments, annual greenhouse-gas emissions That’s why the new report emphasizes efficiency in everything from vehicles to
will need to be cut by nearly half between that improved access to financing, espe- appliances could shave off emissions too.
now and 2030, according to the report. It cially for developing nations, would help Public policies have already been effective
calculates that the results from actions speed climate action. at boosting efficiency measures in partic-
taken now will be clear in global tempera- ular, according to the report. Efficiency
ture trends within two decades.  Cut methane emissions from gains can also help make climate prog-
“We already have the technology fossil-fuel production and waste.  ress in sectors like aviation and ship-
and the know-how to get the job done,” Cutting methane emissions this ping, which will be much more difficult
said Inger Andersen, executive director
of UN Environment Programme, during
the press conference. 
2  decade will be key to reaching
climate goals and limiting peak
warming levels: hitting the 1.5 °C target
to clean up in the long term. Many of
these solutions are the same ones that
the IPCC and others have been talking
Stopping climate change will still be will require methane emissions to fall by about for decades.   
complicated and expensive, and long-term a third between 2019 and 2030, accord- Now, there’s only one clear path for-
emissions cuts may rely on technologies, ing to the IPCC report. Some of the top ward. “We must move from climate pro-
like carbon dioxide removal, that are still targets for emissions cuts include oil and crastination to climate action,” Andersen
unproven at scale. In addition to tech- gas production and food waste. said, “and we must begin this today.” Q
nological advances, cutting emissions in Investments in new infrastructure to
industries that are difficult to transform cut methane emissions from oil and gas Casey Crownhart is a climate reporter at
will take time, funding, and political action.  could end up breaking even: according MIT Technology Review.
14 The Download

Exquisite
bullshitters
Not for the first time, we’re
experiencing what it’s like to be
both seduced by and skeptical
of artificial intelligence.
By Ariel Aberg-Riger

GUTTER CREDIT HERE

Artist Ariel Aberg-Riger is author of

America Redux: Visual Stories From
Our Dynamic History.
GUTTER CREDIT HERE
The Download
15
16 The Download

virtual reality as an answer. The goals vary from helping reduce

aggressive behavior to facilitating empathy with victims to, as in
Can VR help Colorado’s case, reducing recidivism. Though the state’s prison
budget sits close to $1 billion, Colorado has one of the worst
inmates successfully return-to-prison rates in the country, at around 50%. Nationally,
as many as two-thirds of the 600,000 people released from state
reenter the and federal prisons each year will be rearrested within three years.
Is VR the long-missing piece in an unwieldy puzzle of
outside world? resources and programs meant to help reverse these statistics?
Or is it yet another experiment that will fail to adequately pre-
Corrections systems are using simulators to pare incarcerated individuals for life beyond lockup? “It’s not
provide incarcerated people with more lifelike going to be the silver bullet, but it is a tool that I think is very
instruction, but research into the effectiveness powerful for a lot of people, because they never really get a
chance to practice what we’re trying to teach them,” says Bobbie
of VR in these settings is scant.
Ticknor, an associate professor of criminal justice at Valdosta
By Daliah Singer State University. “I think we should use everything we can find
and see what works the best.”
Atorrus Rainer, age 41, is standing in the center of a stuffy, Proponents like Ticknor say VR can immerse incarcerated
fluorescent-lit room. A virtual-reality headset covers his eyes like people in the sights and sounds of modern life and help them
oversize goggles. Every so often, he extends his arm, using the develop digital literacy in a secure corrections environment.
VR controller to pick up garbage bags, a toothbrush, and toilet “When you’re role-playing, when you’re learning a new skill, the
paper during a simulated trip to the supermarket. The experi- closer you can bring them to doing what they’re actually going to
ence is limited—Rainer has to follow a pre-written shopping have to do out in the real world, the better,” says Ethan Moeller,
list and can only travel to specific locations within the empty founder and managing director of Virtual Training Partners,
store—but the sheer number of products available, even in this which helps organizations successfully implement virtual-reality
digital world, still overwhelms him. So does the self-checkout tools. “VR does that better than any other training medium.”
station: those didn’t exist in 2001, when Rainer, then a teenager, Others are more skeptical. Like Dr. Cyndi Rickards, an asso-
was sentenced to more than 100 years in prison. His first expe- ciate teaching professor at Drexel University who leads weekly
rience with one is this virtual interaction taking place inside criminology courses inside Philadelphia prisons. People who
Fremont Correctional Facility, a medium-security prison about are incarcerated wear the “label of inmate on their back. It’s
two hours south of Denver. a dehumanizing system,” she says, “so to suggest that VR is
Rainer is practicing in the hopes of stepping into a real store going to reintegrate them into society after being in a punitive
in the near future through an initiative launched in Colorado in system...just further objectifies folks, it continues a pattern of
2017 in response to US Supreme Court rulings that deemed juve- dehumanizing folks, and I’ve not read any compelling evidence
nile life without parole sentences unconstitutional. People who that this is the route we should use to integrate people to be
meet certain requirements—for example, if they were under 21 members of a healthy and contributing society.”
when they committed felony crimes and have been incarcerated Rainer believes the grocery store simulation was beneficial
for a minimum of 20 to 30 years—can apply to work through but is aware that the real world, should he step back into it, will
the three-year Juveniles and Young Adults Convicted as Adults be very different from the video-game-like version he’s inter-
Program (JYACAP) in an effort to earn early parole. acting with at Fremont. “Going back to society, I don’t want
The premise of JYACAP is that learning the basic skills they to freeze up while I’m in a grocery store or something, not
missed the chance to acquire while incarcerated will provide figuring out what I need to buy because [there are] too many
these juvenile lifers with their best chances for success upon options,” he says. “I don’t really like working on a computer,
release. That’s a formidable challenge. Because of safety concerns, but I know I got to.”
they have had limited access to the internet. Though they’re As VR technology grows more affordable, the programming
now adults, many have never used, or even seen, a smartphone becomes an increasingly budget-friendly option for states that
or a laptop. Or had a credit card. “We had to figure out a way of are already dealing with persistent workforce shortages. “If we
giving them these opportunities in a restricted environment,” reduce recidivism rates, it actually helps the community and
says Melissa Smith, interim director of prisons for the Colorado reduces crime,” explains Sarah Rimel, the former technology
Department of Corrections. research program manager at Colorado’s National Mental Health
Though its use is not yet widespread, a handful of state cor- Innovation Center. “It reduces the amount of money that’s put
rections departments, from Ohio to New Mexico, have turned to into the prison systems.”
 The Download 17

VR has proved a beneficial therapeutic tool, helping to lower times greater odds of getting a job. “Above just the employment
depression rates, reduce anxiety, conquer phobias, promote rate, those that interviewed with Molly [the virtual hiring man-
emotional empathy, and address post-traumatic stress. VR ager] had stronger interview skills over time, greater reductions
exposure therapy has been successfully used to help vulnera- in interview anxiety over time, and greater increase in motivation
ble populations such as veterans and sexual-assault survivors to interview over time,” says Matthew Smith, a professor of social
confront, and better cope with, their triggers and trauma. All work at the University of Michigan, who led the effort. He and
that research is based on interventions done with people who his team are now enrolling a larger group in a validation study.
are not incarcerated, however. Colorado doesn’t have any data sets to point to. Only one
The currently available evidence in correctional settings is of the 16 people who’ve been released through JYACAP over
limited and mostly anecdotal. But there have been some posi- the course of almost three years have been rearrested. Two of
tive findings. For example, a short-term pilot initiative in Alaska those 16 were paroled before completing the full curriculum.
that incorporated mindfulness techniques through VR resulted “If the right scenarios are used,” says Cheryl Armstrong, one
in decreased reports of depressive or anxious feelings and fewer of the first JYACAP graduates, “it [VR] is helpful, to a certain
disciplinary write-ups. In Michigan, a virtual-reality tool for job extent, to give you an idea of what you’re going to be facing.”
interview training, originally developed for people with serious While Valdosta State’s Ticknor estimates that fewer than 10%
mental illness, was piloted with 44 men involved with the justice of corrections facilities are currently using VR simulators with
system. The findings, published in March 2022, showed that incarcerated individuals, she expects that to change soon. “I
82% of those who used the tool landed a job within six months of would be very surprised within five years if this is not a very reg-
being released, compared with 69% of other program participants. ular treatment modality for this particular population,” she says. Q
When variables like age, race, and time served were taken into
account, the data suggested that those who used the tool had 7.4 Daliah Singer is a freelance journalist based in Denver.

future. It works by measuring thousands

Jobs of the future: of individual genetic differences between
them, resulting in what’s known as a poly-
Marketer of embryo predictions genic score. Testing an embryo costs about
$1,000. “This was not even in the realm
By Antonio Regalado of possibility when I was born. And so it’s
really exciting,” says Carr.
Elizabeth Carr is head of commercial devel- always had reporters asking me questions,
opment at Genomic Prediction, a New and I was like, ‘I could do a better job.’” Downside of being America’s first IVF baby:
Jersey genetic testing startup that assesses She later moved into ghostwriting and “There’s no lying about my age for sure,”
IVF embryos for their future risk of a dozen marketing for IVF clinics. she says. “And yes, it is slightly awkward
common diseases so parents can pick the to still be referred to as a baby.” Q
“best” one. It’s a controversial area that Handling controversy: Embryo scores are
has critics anguishing over the dawn of hotly debated, with some geneticists call-
consumer eugenics. ing them unproven and even unethical.
Carr skillfully dodges this debate, noting
Wow, that résumé: Carr, who is in charge that IVF itself once attracted similar con-
of sales and marketing, is a natural fit for cerns. “Not to draw a very, very, obvious
the job. That’s because she is “America’s connection to my own life, but it’s really
PHOTO COURTESY OF ELIZABETH CARR

first test-tube baby,” as the headlines pro- no different, right?” she says. “If you have
claimed in 1981, when she became the moral objections to the test or you don’t
first person born through IVF in the US. agree with it, don’t use it. It’s the same
thing with IVF.”
Career path: Carr started out as a health
journalist, spending 15 years at the Boston The product: Genomic Prediction says
Globe. “I had my first press conference its scoring method can help parents pick
when I was three days old,” she says. “I which IVF embryo has the brightest health
18 The Download

Tech that aims to

read your mind and
probe your memories
is already here
We need new rules to protect our cognitive liberty,
says futurist and legal ethicist Nita Farahany.

By Jessica Hamzelou

Nita Farahany, a futurist and legal ethicist at Duke

University in Durham, North Carolina, has spent much
of her career exploring the impacts of new technolo-
gies—in particular, those that attempt to understand
or modify our brains.
In recent years, we’ve seen neurotechnologies
move from research labs to real-world use. Schools wandering, engagement, boredom, interest, happy,
have used some devices to monitor the brain activ- sad. You could work out how they are thinking or
ity of children to tell when they are paying attention. feeling, whether they are hungry, whether they are
Police forces are using others to work out whether a Democrat or Republican.
someone is guilty of a crime. And employers use them to keep You can also pick up a person’s reactions, and try to probe
workers awake and productive. the brain for information and figure out what’s in their mem-
These technologies hold the remarkable promise of giving ory or their thought patterns. You could show them numbers
us all-new insight into our own minds. But our brain data is pre- to try to figure out their PIN number, or images of political
cious, and letting it fall into the wrong hands could be danger- candidates to find out if they have more positive or negative
ous, Farahany argues in her new book, The Battle for Your Brain. reactions. You can probe for biases, but also for substantive
I chatted with her about some of her concerns. knowledge that a person holds, such as recognition of a crime
scene or a password.
Your book describes how technologies that collect and probe
our brain data might be used—for better or for worse. What can Until now, most people will only have learned about their brain
you tell from a person’s brain data? data through medical exams. Those records are protected.
When I talk about brain data, I’m referring to the use of EEG, What about brain data collected by consumer products?
fNIRS [functional near-infrared spectroscopy], fMRI [func- I feel like we’re at an inflection point. [A lot of] consumer
tional magnetic resonance imaging], EMG, and other modal- devices are hitting the market this year and in the next two
ities that collect biological, electrophysiological, and other years. There have been huge advances in AI that allows us to
functions from the human brain. These devices tend to collect decode brain activity, and in the miniaturization of electrodes,
data from across the brain, and you can then use software to which [allows manufacturers] to put them into earbuds and
try to pick out a particular signal. headphones. And there has been significant investment from
big tech companies. It is, I believe, about to become ubiqui-
tous. The only person who has access to your brain data right
PHOTO COURTESY OF NITA FARAHANY

Neuroscientists listened in on people’s brains for a week.

They found order and chaos. now is you, and it is only analyzed in the internal software of
The study shows that our brains exist between chaos and your mind. But once you put a device on your head … you’re
stability—a finding that could be used to help tweak them immediately sharing that data with whoever the device manu-
either way. facturer is, and whoever is offering the platform. Q
Brain data is not thought. But you can use it to make infer-
ences about what’s happening in a person’s mind. There are Jessica Hamzelou is a senior reporter at MIT Technology
brain states you can decode: tired, paying attention, mind Review. Read the full Q&A at www.technologyreview.com.
 The Download 19

These pothos plants are

engineered to remediate
air pollution

of Exposure Science & Environmental Epidemiology provided

some sobering context: it would take 680 plants to clean the air in
The air-cleaning a 1,500-square-foot room—highly unrealistic for most plant par-
ents. If the French biotech startup Neoplants has its way, though,
qualities of plants are you might need only one. 
Neoplants’s marquee product, announced late last year, is the
getting a genetically Neo P1, the first houseplant genetically engineered to remediate
indoor air pollution. At first blush, this high-tech pothos—a tropical
modified boost vine native to the Solomon Islands, also known as “Devil’s Ivy”—is
indistinguishable from the real thing. It’s photogenic, fast-growing,
Indoor plants aren’t as good for air quality as you and hard to kill. But unlike typical nursery stock, it also metabo-
might think, but a French startup aims to fix that. lizes indoor air pollutants missed by traditional air purifiers, which
filter particulate matter: the volatile organic compounds (VOCs)
By Claire L. Evans produced by paint, gas stoves, and building materials.
“It’s actually a two-pronged approach,” explains Neoplants’s
In the late 1980s, NASA conducted a study to determine how well chief technology officer and cofounder, Patrick Thorbey. The first
indoor plants like aloe vera, Chinese ivy, and potted chrysanthe- prong is the genetic engineering of the plant’s metabolism. By
mums abate air pollution. The results were a boon to nursery own- introducing additional genes into the plant, Thorbey’s Paris-based
ANTOINE GUILLOTEAU

ers everywhere: the research showed that houseplants can capably team coaxed the pothos to produce enzymes allowing it to use the
dispatch harmful pollutants including benzene and formaldehyde.  VOCs it absorbs as carbon sources in its normal cellular metabo-
But NASA’s study was conducted in sealed chambers mimick- lism. In a virtuous cycle, more air pollution only creates more plant
ing future long-term space habitats. A 2020 analysis in the Journal matter and greater pollution-fighting capacity. 
20 The Download

The second prong is bacterial. In a Neoplant, as in nature,

microbes do the heavy lifting; two strains of symbiotic bacteria
inserted into the Neo P1’s soil turn formaldehyde and the class
of pollutants known as BTEX—benzene, toluene, ethylbenzene,
and xylene—into harmless sugars and amino acids. 
“Bacteria are really important parts of most nutrient cycles,”
Op-ed
explains Jenn Brophy, a Stanford researcher whose lab develops
genetically engineered plants with greater resilience to climate
change. “But microbiomes are very difficult to maintain. As
soon as you ship a product to somebody, the viability of these
bacteria declines.” This vulnerability seems to be Neoplants’s
business model: the company will offer concentrated doses of
proprietary microorganisms it calls “power drops” to maintain A ban on ChatGPT
the plant’s air-cleaning efficiency. These will need to be applied
monthly, much like replacing the filter in an air purifier. “Dyson, does more harm
they sell their filters,” says cofounder and CEO Lionel Mora.
“We sell microbiome.” than good 
For now, the pothos itself is responsible for only about 30%
of the Neo P1’s air-cleaning capacity—the microbiome handles Most of my classmates still haven’t heard
the rest—but Mora and Thorbey expect that to change soon. about OpenAI or any of its AI models.
It’s faster to improve on microbes than plants, they explain, so That’s a problem.
“the limits of what we can do with the plant are still far in front
of us,” Mora says. “We are at the frontier of what is doable right By Rohan Mehta
now, but we see tremendous potential.”
The Neo P1 is the company’s first volley. “Air-filtering plants The release of ChatGPT has sent shock waves through
may get people to think about GMOs in a new way,” says Brophy. the halls of higher education. Universities have rushed
“Having something that you can touch and feel that is nonthreat- to release guidelines on how it can be used in the class-
ening, but tangible, is a great way to get people introduced to room. Professors have taken to social media to share a
the concept of genetically modified organisms.” spectrum of AI policies. And students—whether or not
they’ll admit it—have cautiously experimented with the
“I’ll be disappointed if idea of allowing it to play a part in their academic work. 
But the notion of a measured response to the emer-
there’s a plant on the moon gence of this powerful chatbot seems to have barely
and it’s not a Neoplant.” penetrated the world of K–12 education. Instead of
transparent, well-defined expectations, high school-
The timing is fortuitous. Pothos plants have become famil- ers across the country have been confronted with a
iar companions in the indoor landscape of remote work just as silent coup of blocked AI websites.1 
the political debate about gas stoves has raised our awareness That’s a shame. If educators actively engage with
of once-unfamiliar domestic hazards. According to the EPA, students about the technology’s capabilities and lim-
Americans spend around 90% of their lives inside, where con- itations—and work with them to define new academic
centrations of some pollutants can be anywhere from two to standards—ChatGPT, and generative AI more broadly,
five times higher than they are outdoors. “Usually we feel safe could both democratize and revitalize K–12 education
indoors,” says Mora. “But covid has shown us that even indoors, on an unprecedented scale.
invisible things can be very harmful.” A bold claim, I know. But after a few months of
It’s clear that Mora and Thorbey are ultimately looking beyond putting generative AI to the test (a nerdy case of
indoor air cleaning and toward climate applications. “It’s easier senioritis, if you will), I’m optimistic. Exhibit A?
to have an impact in the bedroom than to start with the atmo- College applications. 
sphere,” Thorbey says. “But I’ll be disappointed if there’s a plant Few things are as mentally draining as applying
on the moon and it’s not a Neoplant.” Q to college these days, and as I slaved away at
my supplemental essays, the promise of using ChatGPT
Claire L. Evans is a writer and musician exploring ecology, as a real-time editor was attractive—partly as a poten-
technology, and culture. tial productivity boost, but mostly as a distraction. 
 The Download 21

I had ChatGPT carefully review my cloying use individual educational outcomes. English teachers
of semicolons, grade my writing on a 0–10 scale (the could use it to rephrase the notoriously confusing
results were erratic and maddening)2, and even role- answer keys to AP test questions, to help students pre-
play as an admissions counselor. Its advice was fun- pare more effectively. They could provide each student
damentally incompatible with the creative demands with an essay antithetical to the one they turned in,
of the modern college essay, and I mostly ignored it. and have them pick apart these contrary arguments
But the very act of discussing my writing “out loud,” in a future draft. No human teacher could spend the
albeit with a machine, helped me figure out what I time or energy needed to explain pages upon pages
wanted to say next. Using ChatGPT to verbalize the of lengthy reading comprehension questions or com- Rohan Mehta
space of possibilities—from the scale of words to pose hundreds of five-page essays, but a chatbot can. 
paragraphs—strengthened my own thinking. And Educators can even lean into ChatGPT’s tendency
I’ve experienced something similar across every to falsify, misattribute, and straight-out lie as a way
domain I’ve applied it to, from generating fifth- of teaching students about disinformation. Imagine
grader-level explanations of the French pluperfect using ChatGPT to pen essays that conceal subtle logi-
to deciphering the Latin names of human muscles. cal fallacies or propose scientific explanations that are
All this adds up to a simple but profound fact: almost, but not quite, correct. Learning to discriminate
anyone with an internet connection now has a per- between these convincing mistakes and the correct
sonal tutor, without the costs associated with private answer is the very pinnacle of critical thinking, and
tutoring. Sure, an easily hoodwinked, slightly delu- this new breed of academic assignment will prepare
sional tutor, but a tutor nonetheless. The impact of students for a world fraught with everything from
this is hard to overstate, and it is as relevant in large politically correct censorship to deepfakes. 
public school classrooms where students struggle to There are certainly less optimistic visions for the
receive individual attention as it is in underserved future. But the only way we avoid them—the only
and impoverished communities without sufficient way this technology gets normalized and regulated
educational infrastructure. As the psychologist alongside its similarly disruptive forebears—is with
Benjamin Bloom demonstrated in the early 1980s, more discussion, more guidance, and more under-
one-on-one instruction until mastery allowed almost standing. And it’s not as if there’s no time to catch up.
all students to outperform the class average by two ChatGPT won’t be acing AP English classes anytime
standard deviations (“about 90% … attained the level soon, and with the recent release of GPT-4, we are Footnotes
… reached by only the highest 20%”).   already seeing an explosion of ed-tech companies that
1 At least in my case,
ChatGPT certainly can’t replicate human inter- reduce the effort and expertise needed for teachers the entirety of
action, but even its staunchest critics have to admit and students to operate the bot.  openai.com has
it’s a step in the right direction on this front. Maybe So here’s my pitch to those in power. Regardless been blocked, not
only 1% of students will use it in this way, and maybe of the specific policy you choose to employ at your just chat.openai.com.
Kind of annoying if I
it’s only half as effective as a human tutor, but even school, unblock and unban. The path forward starts
want to access the
with these lowball numbers, its potential for democ- by trusting students to experiment with the tool, and fine-tuning docs. 
ratizing educational access is enormous. I would guiding them through how, when, and where it can
even go so far as to say that if ChatGPT had existed be used. You don’t need to restructure your whole 2 The most impres-
during the pandemic, many fewer students would curriculum around it, but blocking it will only send sive thing I have
seen ChatGPT do
have fallen behind.  it underground. That will lead to confusion and mis-
is revise one of my
Of course, those decrying ChatGPT as the end of interpretation in the best of cases, and misuse and essays. In it, I dis-
critical thinking would likely protest that the bot will abuse in the worst.  cussed two global
only exacerbate the lazy academic habits students ChatGPT is the only beginning. There are simply political figures,
might have formed over the course of the pandemic. too many generative AI tools to try to block them all, but concealed
their identities
I have enough experience with the tips and tricks we and doing so sends the wrong message. What we
through personifi-
PHOTO COURTESY OF ROHAN META

high schoolers employ on a regular basis to know that need is a direct discourse between students, teach- cation. To “make my
this is a valid concern—one that shouldn’t be brushed ers, and administrators. I’m lucky enough to be at a essay a 10/10” and
off by casting ChatGPT as just the latest in a long line school that has taken the first steps in this direction, “increase clarity,”
of technological revolutions in the classroom, from and it’s my hope that many more will follow suit. Q ChatGPT filled their
names in. The fact
the calculator to the internet.
that it has emergent
That said, ChatGPT has just as much poten- Rohan Mehta is a high school senior at Moravian abilities like this
tial in the classroom as it does for improving Academy in Bethlehem, Pennsylvania. blew my mind!
22 Explained

How do fungi
Turns out they have a lot to say.
By Michael Hathaway
and Willoughby Arévalo

communicate?
Illustration by Kate Dehler

Within a fungus  With other organisms 

A
lthough most of us think of
fungi as “mushrooms,” these Each growing tip has both autonomy from Fungi “talk” and respond to many other
spore-producing bodies are and accountability to the whole organism, beings.
just the reproductive organs akin to the relationship of social insects to Through mycorrhizal mutualisms,
of mycelium—decentralized, the hive. Between the cells within every they may share water and food with
weblike bodies of branching tubes. Though mycelium flows a stream of chemicals, plant partners. Parasitic fungi produce a
usually microscopic, these structures can nutrients, and electrical impulses. Their myriad of plant growth regulators, mod-
be enormous; the largest known example movements act to keep the whole informed ifying plants to suit their needs. Some
is a honey mushroom (Armillaria) that about happenings and coordinate actions fungi, such as truffles, mimic animal sex
covers almost 10 square kilometers (3.7 across the network. Research by Andrew pheromones to attract mammals and
square miles) and has lived for millennia.  Adamatzky, a professor of unconventional insects that act as “sporinators,” the fun-
As organisms living in complex rela- computing at the University of the West gal equivalent to pollinators. Other fungi
tions to other life forms, fungi could not of England in Bristol, suggests that they are prey to roundworms (also known as
exist without communicating. And while influence the mycelium’s internal bio- nematodes). When they detect a nema-
they’ve traditionally been viewed as ses- electrical signals, which may form a sort tode nearby, they can produce defensive
sile, or permanently fixed in place, mycelia of “language.” While a mycelium neither compounds to ward it off. Other fungi
move by extending the tips of their tubes is nor contains a nervous system, myce- hunt nematodes by detecting their chem-
through a substrate, which could be a patch lia share much in common with these ical presence. 
of soil or a fallen log. systems. Both have branched structures, Mycorrhizal fungi are central in
As fungi grow, they are constantly sens- reinforce or prune pathways as needed, current debates about the “wood-
ing, learning, and making decisions. Fungi and use some of the same amino acids to wide web,” but many representations
are like polyglots: they both “speak” and transmit information. unfairly present fungi as living fiber-
understand a wide range of chemical sig- optic cables that allow trees to “talk”
nals. They release and respond to chemicals Between fungi of the same species  to each other. Fungi are more than just
that float through the air and flow through Many fungi are sexual and must mate to passive wires; they are, in fact, actively
water. Fascinatingly, fungi not only per- reproduce. They send out pheromones perceiving, interpreting, and signaling
ceive but actively interpret a chemical’s and “sniff” out those of others, and then themselves. They do this constantly,
meaning depending on the context and they grow toward those that seem attrac- with a wide range of beings. How mush-
in relation to other chemicals.   tive (based on whatever it is fungi are rooms create and interpret these sig-
Studies of how fungi communicate lag attracted to). Whenever two mycelia meet, nals in a cacophony of chemical and
way behind research on communication they communicate to negotiate their rela- electrical noise remains a fascinating
of plants and especially of animals. Most tionship, which can range from fusion (to mystery.
are based on several “lab rat” species, so form a reproductive or nonreproductive
knowledge about other types is limited, but partnership) to indifference to physical Michael Hathaway is the author
of What a Mushroom Lives For:
here we summarize what’s known about exclusion and even chemical antagonism. Matsutake and the Worlds They
three realms of communication: within a Each mated mycelium negotiates the Make. Willoughby Arévalo is
the author of DIY Mushroom
fungus, between fungi of the same species, physical dynamics of fusion, and of life Cultivation: Growing Mushrooms at
and with other organisms.  in partnership thereafter. Home for Food, Medicine, and Soil.
24 Profile

The Nigerian
Mustapha Gajibo aims to bring
electrification to communal buses.
By Valentine Benjamin

university dropout
Photographs by Fati Abubakar

who builds EVs

Nigerians have become accustomed to long The first, introduced in 2021, is a 12-seat major things we put into consideration,”
lines for gasoline and wild fluctuations in bus constructed from a number of locally he says. “The only way to achieve that is
bus fares. Though the country is Africa’s sourced materials. It has a range of 212 kilo- by fully designing and building these vehi-
largest producer of oil, its residents don’t meters and can be charged in 35 minutes cles locally.”
benefit from a steady supply. via a solar-powered system integrated into State and local governments are now tak-
Mustapha Gajibo, 30, is doing what he the back. In a recent test run funded by the ing notice. In early 2022, for example, the
can to alleviate the problem: his startup,
Phoenix Renewables Limited, is launching
a homegrown electric-vehicle industry in
the northeastern city of Maiduguri.
Gajibo dropped out of university in his
third year to run it. His first project was
converting the internal-combustion engines
of commonly used vehicles in the city to
electric versions. He focused on two types
of vehicles that residents often pay to ride:
seven-seat minibuses and the motorized
tricycles known as kekes. Phoenix Renewables maintains a fleet of a dozen retrofitted electric
He faced skepticism at first: limited minibuses capable of covering a distance of 150 kilometers on a charge.

power charging infrastructure has con-

strained the adoption of electric vehicles company, the buses transported 35,000 governor of Borno State, where Maiduguri
in the region. “Many people don’t believe passengers in Maiduguri in just one month. is situated, commended Gajibo’s work
that electric mobility is possible and com- Deborah Maidawa, an electrical building and awarded him 20 million naira (about
mercially viable in the city of Maiduguri,” services engineer who lives in Maiduguri, $45,000) for research and development,
Gajibo says. But his electrification scheme believes Gajibo’s EVs are a good way to meet as well as 15,000 square meters of land for
has been gaining traction. The company local needs. “Incorporating solar gives the a factory. The Nigerian government has
now maintains a fleet of a dozen electric vehicles an edge over other EVs that are expressed interest in having his company
minibuses that can cover a distance of 150 springing up, and I believe they will flood build electric patrol vehicles for the police
kilometers on a charge and cost about $1.50 the Nigerian market,” she says. and armed forces.
to power to full capacity. A brand-new gas-powered passenger Gajibo’s ultimate goal is to compete with
Building the necessary infrastructure is minibus with automatic transmission can Tesla and other bigger brands. “We want
crucial to the success of the project. Gajibo cost nearly 5 million naira (about $10,000). to have our vehicles driven in New York,
and his cofounder Sadiq Abubakar Issa Gajibo says it will cost around the same to London, Munich, and other big cities across
designed a 60-kilowatt-hour solar-powered buy one of his solar-powered 12-seaters. the world,” he says.
charging station in the city and are looking He plans to roll out 500 units across eight
at creating more. Nigerian cities in the coming months and Valentine Benjamin is a Nigerian
Now, Gajibo has moved on from retrofit- hopes this time he’ll be able to sell them. travel journalist and photographer
who reports on global health, social
ting internal-combustion vehicles to building “Our products are quite affordable, justice, politics, and development
electric vehicles from scratch. and the cost of the vehicle is one of the in Nigeria and sub-Saharan Africa.
26

Denmark’s
enthusiasm
for educational tech
is taking on
a new frontier:
children’s well-being.

By
Arian Khameneh

Calibrating
the
classroom
Illustrations by
Nicole Rifkin
GUTTER CREDIT HERE
28

taken immediately, they “see no hope for

turning the negative trend around.”
To help address the problem, some
Danish schools are moving to address chil-

In
dren’s well-being through platforms like
Woof, the one used in the fifth-grade class-
room. Built by a Denmark-based startup,
it frequently surveys schoolchildren on a
variety of well-being indicators and uses
an algorithm to suggest particular issues
a Copenhagen suburb, a fifth-grade classroom for the class to focus on.
These platforms are quickly gaining
is having its weekly cake-eating session, a common tradition in
ground. Woof, for example, has been imple-
Danish public schools. While the children are eating chocolate mented in classrooms in more than 600
cake, the teacher pulls up an infographic on a whiteboard: a bar schools across Denmark, with more on
the way. Its founders believe Woof fills
chart generated by a digital platform that collects data on how an important niche: they say teachers
they’ve been feeling. Organized to display the classroom’s weekly have expressed widespread dissatisfaction
“mood landscape,” the data shows that the class averaged a mood with existing tools, in particular a govern-
ment-run well-being survey. That survey
of 4.4 out of 5, and the children rated their family life highly. “That’s audits schools once a year and delivers
great!” the teacher exclaims, raising two thumbs up in the air. results on a delay; it might provide a snap-
shot for policymakers but is hardly useful
She then moves to an infographic on sleep pencil and drawing pad. It’s a tool that for teachers, who need regular feedback
hygiene. Here the data shows the students is bound to the child’s opportunity to to adjust their work.
struggling, and the teacher invites them express themselves,” Mari-Ann Letnes, “There is simply a need for tools to
to think of ways to improve their sleeping an education scientist in Norway, said in check in [with the children] where you
habits. After briefly talking among them- a 2018 interview. In a 2019 status report don’t need to be active,” says Mathias
selves, the children suggest “less screen on the use of technology in schools, the Probst, a cofounder of Woof. “Where you
time at night,” “meditation before sleep,” Danish Ministry of Education stated that don’t need to talk to all 24 children before
and “having a hot bath.” They collectively “creativity and self-expression with digital starting a class, because before you know
make a commitment to implement these technologies are a part of building students’ it, 15 minutes of class time has already
strategies. At next week’s cake time, they motivation and versatile development.” passed.” And teachers could benefit, he
will be asked whether or not they followed Now, some teachers and administrators are suggests, from “something that can bring
through. hoping technology can be used to tackle a data structure into all of this.”
These sorts of data-driven well-being mental health as well. Woof is not alone in its attempt to quan-
audits are becoming more and more com- Danish schoolchildren are in the midst tify children’s moods. A handful of other
mon in Denmark’s classrooms. The country of a mental-health crisis that one of the platforms have been adopted by Danish
has long been a leader in online services country’s biggest political parties has schools, and schools in Finland and the
and infrastructure, ranking as the most called a challenge “equal to inflation, the UK are using mood-monitoring software
digitally developed nation in the UN’s environmental crisis, and national secu- as well. In the US, the tech can extend
e-government survey. In recent years its rity.” No one knows why, but in just a few beyond collecting self-reports to hunting
schools, too, have received big investments decades, the number of Danish children for hints of concerning behavior by sur-
in this type of technology: it is estimated and youth with depression has more than veilling students’ emails, chat messages,
that the Danish government allocated $4 sextupled. One-quarter of ninth grad- and searches on school-issued devices.
to $8 million, a fourth of the high school ers report that they have attempted self- A number of people say mood-
budget for teaching aids, to procuring dig- harm. (The problem isn’t exclusive to monitoring tech has great potential. “We
ital platforms in 2018. In 2021, it invested Denmark: depressive episodes among can use digital tools to evaluate well-being
some $7 million more. US teens increased by some 60% between on a 24-hour basis. How is the sleep?
These investments are rooted in a 2007 and 2017, and teen suicide rates have How is the physical activity, how is the
Nordic tradition of education that cen- also leaped by around 60% over the same interaction with others? ... How does [the
ters the child’s experience and encourages period.) A recent open letter signed by child’s] screen time compare to physical
interactive learning; some Scandinavian more than 1,000 Danish school psychol- time? That’s central to understanding what
education researchers think technology ogists expressed “serious concerns” over well-being actually is,” the late Carsten
can help draw children in as playful, active the mental state of the children they see in Obel, who was a professor of public health
participants. “Technology is an extended their work and warned that if action isn’t at Aarhus University and a leader in the
 29

development of another student-surveying realized that schools in low-income neigh-

tool called Moods, said in a 2019 video. borhoods face a vicious cycle. Difficult cir-
But some experts are heavily skeptical cumstances at home can make students in
of the approach. They say there is little evi- these schools more challenging to teach.
dence that quantification of this sort can Staff turnover rates are high because of
be used to solve social problems, and that stress and burnout, with some teachers keen
fostering a habit of self-surveillance from to switch to “easier” schools. Parents with
an early age could fundamentally alter chil- resources often take their children else-
dren’s relationship to themselves and each where, so kids with more problems make
other in a way that makes them feel worse up an even greater proportion of those who
Teachers using Woof see rather than better. “We can hardly go to a remain, exacerbating the stress teachers
classroom averages instead restaurant or to the theater without being face and the likelihood that they’ll leave.
of data on individual children. asked how we feel about it afterwards and All this compounds the well-being crisis
ticking boxes here and there,” says Karen that children are experiencing elsewhere.
Vallgårda, an associate professor at the “I saw so many children ending up in
University of Copenhagen who studies fam- difficult situations, which could have been
ily and childhood history. “There is a quan- prevented if action had been taken earlier,”
tification of emotions and experiences that says Danckert, who before her stint as a
is growing, and it’s important that we ask teacher worked as an analyst in the chil-
ourselves whether that’s the ideal approach dren and youth section of Copenhagen’s
when it comes to children’s well-being.” Social Services Administration.
Others are asking how much children Danckert and Probst, who has a back-
and their parents actually know about ground in consulting, set out to build a way
what data is being collected—and how it to help schools manage such situations
is being used. While some platforms say before they spiral into serious mental-health
they are collecting minimal or no person- problems—problems that schools’ thinly
ally identifiable data, others mine deep stretched counseling systems may not catch
into individual children’s mental states, until it’s too late.
physical activity, and even friend groups. Woof, the solution they devised, is a
“Their practice is very Silicon Valley– web app that children can access on com-
like. They preach data transparency but puters or phones (a 2019 study found that
have none themselves,” says Jesper Balslev, 98% of Danish children between 10 and
a research consultant at the Copenhagen 15 have access to a smartphone). Its user
School of Design and Technology, of some interface primarily features a cartoon dog,
of these platforms. Balslev says he is con- which asks the children various questions
Bloomsights takes a more cerned that Woof and other platforms are about their life. The tool is designed to be
data-intensive approach, track-
being swiftly and naively rolled out without used on a weekly basis, generating a “mood
ing schoolchildren individually.
The platform even generates
adequate regulation, testing, or efforts to landscape” for the class by prompting kids
sociograms that display their relationships make sure that the school culture allows to rate their mood and other aspects of their
with one another. children to abstain from participating in lives on a 1–5 scale. The result is supposed
them. “Our regulatory technologies to deal to add up to a comprehensive image of
with this are terrible,” he says. It’s possible child welfare in that classroom over time. 
that will change, he adds, “but right now, all Teachers and administrative staff can
the hobs are turned on at the same time.” read weekly reports on a class’s overall
self-reported mood and how factors like
oof is run from a basement office their sleep hygiene, social activity, aca-

W on the outskirts of Copenhagen,

with a small team of three full-
time staffers. The founders, Mathias Probst
demic performance, and physical activity
affect that mood. Classrooms are profiled,
and interventions are recommended to
and Amalie Danckert, got the idea for the improve the scores in categories where they
company after working as public school are doing less well. Finally, the teacher and
Klassetrivsel also
generates results on individuals,
teachers through Teach First Denmark, an the children look at the data together and
including sociograms. organization similar to Teach for America help each other with tools and strategies
in the United States. to improve these sticking points.
When Probst and Danckert entered the Woof’s data is anonymized; the app
public school system, they say, they quickly reports on classroom averages instead of
30

individual children. Danckert says that’s

because the company was unwilling to walk
right up to the edge of what was legally
and ethically feasible under data privacy
laws. Probst also describes feeling uneasy
that collecting data on individual children
might create a narrative and lock them into
it, rather than helping them break negative
patterns. “It’s worrying that there is so much
personally attributable data on platforms
working with children,” he says.
The startup fully launched Woof
less than a year ago, in the fall of 2022.
According to beta test data collected on
30 schools before its full launch, 80% of
classes that use Woof see mood improve
by, on average, 0.35 points on the 1–5
scale within one month. Woof maintains
that the platform isn’t meant to replace
teacher-student contact. It should rather
be understood as a support tool for teach-
ers that provides structured action plans
and feedback.
But some experts have doubts about
whether Woof’s methods are effective.
They are particularly skeptical about the
self-reported nature of the platform’s data.
According to Balslev, education apps
have not proved that they perform any better
than analog interventions, such as having
teachers advise children to turn off their
computers and ask them how they slept last
night. He points to historical lessons, such
as a 2015 OECD study finding that digita-
lization in schools in a variety of countries
had exacerbated a range of problems it was
supposed to improve, with a net negative
effect on learning outcomes.
“We intuitively trust data or the quanti-
tative regime more than we trust humans,”
he says. “I have found no, or very few, stud-
ies that examine the use of ed tech in con-
trolled environments.”
And there is good reason to take
self-reported well-being data with cau-
tion: children may not be providing honest
information. Balslev claims that when tech-
nology is introduced into a social context, it
can’t be assumed that students will demon-
strate ideal behavior and cooperate with
its intentions. For example, in interviews
he has done with high school students,
he says they have reported gaming digital
systems to do things like get more time for
an assignment or make a writing exercise
look longer than it actually is.
 31

Though dishonest answers are of course Citing his experiences as a teacher in computer.” Nord is concerned about how
possible, Probst and Danckert argue that inner-city schools in Los Angeles for six many teachers who don’t work directly
Woof’s anonymous approach makes authen- years, Rockenbach says it can be a chal- with the children still have access to their
tic responses more likely than they might lenge to know what is really going on with data. She believes the app straddles ethical
be otherwise. “Many students from low- children who struggle in an environment boundaries given how much it impinges
income areas are very aware of whether that might be marked by gang violence on students’ private lives.
they are anonymous or not. And they are and poverty. He says Bloomsights can “They have no chance of understand-
very aware of what is disclosed about their help in situations where the signals are ing what is going on. It’s not like we give
family life,” says Danckert. “The students not so clear. them a long presentation explaining how
don’t want to talk about what is happening Rockenbach believes that anonymous it’s used and who has access [to the data],”
at home, because they are worried that it will data only makes early intervention more Nord says. “And if we did, we would get no
start a case [with a social services agency],” difficult, since it creates more work for honest answers. If they actually understood
Probst adds. He and Danckert believe that teachers and educators in trying to iden- the amount of data I can see about them
the anonymous approach builds trust and tify who has problems and needs help. For and how many others can see it as well,
promotes honest disclosure, as students can this reason, he thinks collecting individual I believe they would answer differently.”
be sure that it won’t trigger the teacher’s data is a necessity. According to the data policies of
legal obligation to report red flags further The program, which operates through a Klassetrivsel, one of the platforms that
up in the system. web app, takes self-reporting measurements collect non-anonymized data, consent is
similar to Woof’s: monthly surveys of stu- not required from either parents or children
oof isn’t the only well-being dents, measuring various indicators of men- before the app is used in the classroom. The

W platform making inroads in

Danish schools. Platforms like
Bloomsights, Moods, and Klassetrivsel
tal and physical well-being and students’
evaluation of their learning environment.
But Bloomsights stands out in its use of
company claims that since the app is an inte-
grated tool used for “well-being purposes”
at a public institution, it falls under a Danish
(Danish for “classroom well-being”) are also sociograms, which are constructed from the legal clause that exempts public authorities
getting traction. Each takes a more data-
intensive and less anonymous approach
than Woof, tracking and identifying school- “It’s worrying that there is so much
children individually. Bloomsights and personally attributable data on platforms
Klassetrivsel even go as far as generating
“sociograms”—network diagrams that dis-
working with children.”
play the children’s relationships with each
other in detail. students’ reports of who their friends are from requirements about obtaining consent
Bloomsights turns self-reported data and who they connect and spend time with. for data collection. And since the platforms
from the same individuals over time into Rockenbach says these sociograms are aren’t classified as “information society ser-
indicators including “signs of loneliness,” crucial tools to detect social isolation and vices” like Facebook or Google, there is no
“academic mindset,” and “signs of bully- might even help identify children who are parental consent required under the General
ing.” Bloomsights is also used in the US, vulnerable to bullying. He points to testimo- Data Protection Regulation, the European
where some school districts are including nial reports from schools as an indicator that Union’s sweeping data privacy law.
it as part of an “early warning system” for the platform helps improve well-being. But, Legal precedents seem to back up
identifying potential school shooters. he adds, “we haven’t conducted a full-on Klassetrivsel’s claims about how the data
The company’s US operations are based research project that might compare, for law applies to its work. In 2019, a parent
in Colorado. Cofounder Adam Rockenbach example, a school that uses Bloomsights submitted a complaint to the Danish Data
says the hope in bringing Bloomsights to versus a school that doesn’t. That’s some- Protection Agency, claiming that a data-
the US was to spread the Scandinavian thing that we’re looking to do.” driven well-being platform at her child’s
values of well-being and community. He Indeed, some teachers wonder how use- school was engaging in forced monitoring
asserts that the app is not meant to be a ful—or even ethical—the app is. “It’s some of the child. The parent further argued that
dystopian “Big Brother” but an extension very intimate things that are asked, and “measuring and monitoring well-being is
of what teachers already do. they [the children] don’t necessarily know not the same as improving well-being.”
“You notice the student is coming into who is going to see it,” says Naya Marie The agency ruled in favor of the school’s
class, and maybe they’re coming to class late Nord, a teacher at a suburban Copenhagen municipality: the app was deemed a tool
more frequently than before, and they look school that uses Bloomsights. “Of course, for maintaining tasks of “crucial social
a little disheveled,” he says. “A good teacher I as a teacher should have insight into interest” that fall under the responsibil-
is going to go find two or three minutes to how my students are feeling. But that’s ity of schools.
connect with that student: ‘Hey, it seems something that I prefer to have conveyed “Usually, the legal authority that these
like there’s something off here. Is there any in the confidentiality between me and third-party apps operate under is that they
way I can help you?’” the student, rather than it being told to a are offering a service on behalf of the public
32

authorities,” says Allan Frank, an IT law- or Balslev, the embrace of slick data- excessive focus on whether children are
yer at the agency. But they must still store
data correctly and not collect more than is
necessary. They must also operate under
F driven solutions is due partly to their
political appeal. In Denmark, tech-
nology sometimes tends to be presented
happy can cause them to pathologize nor-
mal fluctuations in life. New studies also
indicate that declining well-being is largely
the aegis of governmental authorization, as the solution to everything connected to attributed to environmental and social
he says: “If there is a random teacher or a teaching and education. The simple info- pressures rather than individual factors.
school that has been convinced to suddenly graphics that ed-tech companies offer, he Vallgårda believes that rather than pour-
set it up without the supervision of the says, have an allure for government offi- ing resources into tools that further a quan-
municipality or the Ministry of Education, cials faced with thorny social and peda- titative agenda, schools should instead
then that would be a problem.” gogical issues. be prioritizing efforts to hire and train
In Denmark, parents can opt out if they “What is fantastic about the digital professionals like teachers and school
don’t want data collected on their chil- [initiatives] is that they are good at mak- psychologists.
dren through these apps. According to ing politicians look actionable—as if they But digital platforms are significantly
Bloomsights, this is also the case in the US: have made some decisions,” Balslev says. cheaper than hiring or training more peo-
although practices vary, Rockenbach says But efficacy is not as much of a priority, ple. Viskum, the fifth-grade teacher, points
that parents typically sign a paper once a he says: “It’s quick and easy to produce out that budgets are tight and waiting
year that lists all the different services the some metrics that appear rhetorically con- lists for appointments with the school
school uses. vincing. The infographic might provide a psychologist are miles long. Given the
But because the apps are used in an very thin sliver of truth about reality, but material reality, the appeal of ed tech is
educational context and are framed as it doesn’t touch the core of the situation.” understandable, even when there are few
altruistic, both parents and policymakers In fact, the technology risks actually results to back it up.
tend to have their guard down. “There making the situation worse, says Karen While the quantification of children’s
are a lot of other apps where I limit my Vallgårda, the University of Copenhagen lives might make academics balk, the
son’s use, but I’m not concerned about researcher. She is concerned that the children I met told me that they enjoyed
apps used in the school the same way I am
about TikTok and YouTube, for example,”
says Janni Hindborg Christiansen, mother “The infographic might provide a very
of one of the children in the fifth-grade thin sliver of truth about reality, but it doesn’t
classroom that uses Woof. “At least Woof
is used in a controlled environment and
touch the core of the situation.”
has a good purpose. I trust it more than
so many other apps that I’d be more crit- “surveillance paradigm” could have using Woof and especially liked how the
ical toward.” unintended consequences for children’s app helped them talk more nicely to each
And for parents who don’t want their self-understanding. other. At a school I visited in a low-income
children using such platforms, opting out “If we are asked to monitor ourselves neighborhood (the class scored 3.4 on the
is not always straightforward. according to a quantitative logic, emo- mood scale), a teacher said she was just
Henriette Viskum, the teacher of the tions such as indignation and sorrow happy to have a tool that might give her
fifth-grade class, describes Woof lessons can appear as problematic emotional a general idea of what was going on with
as a part of her class’s core programming, reactions, despite the fact that they are the children.
just like math, and says parents need to completely natural in certain scenarios When I asked Woof’s Probst about
talk with the teacher to pull their child out of life. The children can feel that what Vallgårda’s criticisms, he said that unlike
of the program. “If it’s a huge problem, they are feeling is wrong or undesirable, researchers studying children academi-
we’ll find a solution and then the child which is likely to propel greater well-being cally, those who work with children every
doesn’t have to participate,” Viskum says. issues rather than ameliorating them,” day in the classroom can’t afford to think
“But then I would, as a teacher, put a big Vallgårda says. in abstract terms.
question mark around why the parents “When we instill a measure of self- “It’s all well and good to be a theorist
are so strongly opposed to working with surveillance with children based on a and have the opinion that you shouldn’t
well-being. I would be a bit concerned and clearly communicated ideal of how to be doing certain things, but there is also
curious about that.” structure one’s everyday life, one’s eat- a reality out there in the classrooms,” he
The closeness between teachers and ing habits, and how to feel in certain says. “There is a practical situation where
students can also make the degree of ano- contexts, there is a risk that children teachers face children who are struggling
nymity blurry. Viskum told me that if almost develop ‘double unhappiness’ due to not so much that they break down in tears
an entire class reports high scores on family just being unhappy but also failing to live during class. You have to do something
life, for example, but one child does not, up to these ideals.” there.”
she can usually intuit who that person is Vallgårda’s concerns are echoed by Arian Khameneh is a freelance
and might casually try to take steps to help. other researchers, who argue that an journalist based in Copenhagen.
 35

IS TRANSFORMING
HUMANITIES RESEARCH

Historians are using neural networks By Moira Donovan

to draw new connections in the analysis of history. Illustrations by Beth Hoeckel
36

It’s an evening in 1531,

in the city of Venice. In a printer’s workshop, an appren-
tice labors over the layout of a page that’s destined for an
astronomy textbook—a dense line of type and a wood-
block illustration of a cherubic head observing shapes
moving through the cosmos, representing a lunar eclipse.
Like all aspects of book pro- history one document at a time.
duction in the 16th century, it’s a But it introduces distortions of its
time-consuming process, but one own, including the risk that machine
that allows knowledge to spread learning will slip bias or outright
with unprecedented speed. falsifications into the historical
Five hundred years later, the record. All this adds up to a ques-
production of information is a dif- tion for historians and others who,
ferent beast entirely: terabytes it’s often argued, understand the
of images, video, and text in tor- present by examining history: With
rents of digital data that circulate machines set to play a greater role
almost instantly and have to be in the future, how much should we
analyzed nearly as quickly, allow- cede to them of the past?
ing—and requiring—the training
of machine-learning models to Parsing complexity
sort through the flow. This shift ig data has come to the
in the production of information
has implications for the future of
everything from art creation to drug
B humanities through initia-
tives to digitize increasing
numbers of historical documents,
development. like the Library of Congress’s col-
But those advances are also mak- lection of millions of newspaper
ing it possible to look differently pages and the Finnish Archives’
at data from the past. Historians court records dating back to the 19th
have started using machine learn- century. For researchers, this is at
ing—deep neural networks in once a problem and an opportunity:
particular—to examine historical there is much more information, and
documents, including astronom- often there has been no existing way
ical tables like those produced in to sift through it.
Venice and other early modern cit- That challenge has been met
ies, smudged by centuries spent in with the development of compu-
mildewed archives or distorted by tational tools that help scholars
the slip of a printer’s hand. parse complexity. In 2009, Johannes
Historians say the application Preiser-Kapeller, a professor at the
of modern computer science to the Austrian Academy of Sciences, was
distant past helps draw connec- examining a registry of decisions
tions across a broader swath of the from the 14th-century Byzantine
historical record than would oth- Church. Realizing that making
erwise be possible, correcting dis- sense of hundreds of documents
tortions that come from analyzing would require a systematic digital
 37

survey of bishops’ relationships,

Preiser-Kapeller built a database
All this adds up to a question for historians:
of individuals and used network With machines set to play a greater role in
analysis software to reconstruct the future, how much should we cede to them
their connections.
This reconstruction revealed
of the past?
hidden patterns of influence, lead-
ing Preiser-Kapeller to argue that
the bishops who spoke the most in
meetings weren’t the most influen-
tial; he’s since applied the technique The treatise is also the star (NLP) to read the text would need
to other networks, including the player in a digitized collection of to be retrained for each book.
14th-century Byzantian elite, uncov- 359 astronomy textbooks published The language also posed a prob-
ering ways in which its social fabric between 1472 and 1650—76,000 lem. Many texts were written in
was sustained through the hidden pages, including tens of thousands regionally specific Latin dialects
contributions of women. “We were of scientific illustrations and astro- often unrecognizable to machines
able to identify, to a certain extent, nomical tables. In that comprehen- that haven’t been trained on historical
what was going on outside the offi- sive data set, Matteo Valleriani, languages. “This is a big limitation in
cial narrative,” he says. a professor with the Max Planck general for natural-language process-
Preiser-Kapeller’s work is Institute for the History of Science, ing, when you don’t have the vocabu-
but one example of this trend in saw an opportunity to trace the lary to train in the background,” says
scholarship. But until recently, evolution of European knowledge Valleriani. This is part of the reason
machine learning has often been toward a shared scientific world- NLP works well for dominant lan-
unable to draw conclusions from view. But he realized that discern- guages like English but is less effec-
ever larger collections of text—not ing the pattern required more than tive on, say, ancient Hebrew.
least because certain aspects of human capabilities. So Valleriani Instead, researchers manually
historical documents (in Preiser- and a team of researchers at the extracted the text from the source
Kapeller’s case, poorly handwritten Berlin Institute for the Foundations materials and identified single links
COURTESY OF THE LIBRARY OF THE MAX PLANCK INSTITUTE FOR THE HISTORY OF SCIENCE, BERLIN

Greek) made them indecipherable of Learning and Data (BIFOLD) between sets of documents—for
to machines. Now advances in deep turned to machine learning. instance, when a text was imitated
learning have begun to address This required dividing the collec- or translated in another book. This
these limitations, using networks tion into three categories: text parts data was placed in a graph, which
that mimic the human brain to pick (sections of writing on a specific automatically embedded those sin-
out patterns in large and compli- subject, with a clear beginning and gle links in a network containing
cated data sets. end); scientific illustrations, which all the records (researchers then
Nearly 800 years ago, the helped illuminate concepts such used a graph to train a machine-
13th-century astronomer Johannes as a lunar eclipse; and numerical learning method that can suggest
de Sacrobosco published tables, which were used to teach connections between texts). That
the Tractatus de sphaera, an intro- mathematical aspects of astronomy. left the visual elements of the texts:
ductory treatise on the geocen- At the outset, Valleriani says, 20,000 illustrations and 10,000
tric cosmos. That treatise became the text defied algorithmic inter- tables, which researchers used neu-
required reading for early modern pretation. For one thing, typefaces ral networks to study.
university students. It was the most varied widely; early modern print
widely distributed textbook on geo- shops developed unique ones for Present tense
centric cosmology, enduring even their books and often had their own omputer vision for histor-
after the Copernican revolution
upended the geocentric view of
the cosmos in the 16th century.
metallurgic workshops to cast their
letters. This meant that a model
using natural-language processing
C ical images faces similar
challenges to NLP; it has
what Lauren Tilton, an associate

Top: A page from a 1531 published com- Bottom: A table of values of oblique
mentary of Prosdocimo di Beldomando ascension calculated for the elevation of
on Johannes de Sacrobosco’s Tractatus 48 degrees and 40 minutes to the celes-
de sphaera. The page shows portions of tial North Pole. The values were calcu-
the original and the commentary texts lated by the French royal mathematician
where the mechanics of solar and lunar Oronce Finé.
eclipses are discussed.
38

professor of digital humanities at be printed 1,000 different ways,” are painstaking tasks, especially
the University of Richmond, calls Valleriani explains. So researchers when inscribed objects have been
a “present-ist” bias. Many AI mod- developed a neural network archi- moved or are missing contextual
els are trained on data sets from tecture that detects and clusters cues. Specialized historians need
the last 15 years, says Tilton, and similar tables on the basis of the to make educated guesses. To help,
the objects they’ve learned to list numbers they contain, ignoring Yannis Assael, a research scien-
and identify tend to be features of their layout. tist with DeepMind, and Thea
contemporary life, like cell phones So far, the project has yielded Sommerschield, a postdoctoral
or cars. Computers often recog- some surprising results. One pattern fellow at Ca’ Foscari University of
nize only contemporary iterations of found in the data allowed research- Venice, developed a neural network
objects that have a longer history— ers to see that while Europe was called Ithaca, which can reconstruct
think iPhones and Teslas, rather fracturing along religious lines after missing portions of inscriptions
than switchboards and Model Ts. the Protestant Reformation, sci- and attribute dates and locations
To top it off, models are typically entific knowledge was coalescing. to the texts. Researchers say the
trained on high-resolution color The scientific texts being printed deep-learning approach—which
images rather than the grainy black- in places such as the Protestant city involved training on a data set of
and-white photographs of the past of Wittenberg, which had become more than 78,000 inscriptions—is
(or early modern depictions of the a center for scholarly innovation the first to address restoration and
cosmos, inconsistent in appear- thanks to the work of Reformed attribution jointly, through learning
ance and degraded by the passage scholars, were being imitated in from large amounts of data.
of time). This all makes computer hubs like Paris and Venice before So far, Assael and Sommerschield
vision less accurate when applied spreading across the continent. The say, the approach is shedding light
to historical images. Protestant Reformation isn’t exactly on inscriptions of decrees from an
“We’ll talk to computer science an understudied subject, Valleriani important period in classical Athens,
folks, and they’ll say, ‘Well, we says, but a machine-mediated per- which have long been attributed to
solved object detection,’” she says. spective allowed researchers to see 446 and 445 BCE—a date that some
“And we’ll say, actually, if you take a something new: “This was abso- historians have disputed. As a test,
set of photos from the 1930s, you’re lutely not clear before.” Models researchers trained the model on
going to see it hasn’t quite been as applied to the tables and images a data set that did not contain the
solved as we think.” Deep-learning have started to return similar inscription in question, and then
models, which can identify patterns patterns. asked it to analyze the text of the
in large quantities of data, can help These tools offer possibili- decrees. This produced a different
because they’re capable of greater ties more significant than simply date. “Ithaca’s average predicted
abstraction. keeping track of 10,000 tables, date for the decrees is 421 BCE,
In the case of the Sphaera proj- says Valleriani. Instead, they allow aligning with the most recent dat-
ect, BIFOLD researchers trained a researchers to draw inferences ing breakthroughs and showing how
neural network to detect, classify, about the evolution of knowledge machine learning can contribute to
and cluster (according to similar- from patterns in clusters of records debates around one of the most sig-
ity) illustrations from early modern even if they’ve actually examined nificant moments in Greek history,”
texts; that model is now accessible only a handful of documents. “By they said by email.
to other historians via a public web looking at two tables, I can already
service called CorDeep. They also make a huge conclusion about 200 Time machines
took a novel approach to analyz- years,” he says. ther projects propose to
ing other data. For example, var-
ious tables found throughout the
hundreds of books in the collec-
Deep neural networks are
also playing a role in examining
even older history. Deciphering
O use machine learning to
draw even broader infer-
ences about the past. This was
tion couldn’t be compared visu- inscriptions (known as epigraphy) the motivation behind the Venice
ally because “the same table can and restoring damaged examples Time Machine, one of several local

Computers often recognize only

contemporary iterations of objects that have
a longer history—think iPhones and Teslas,
rather than switchboards and Model Ts.
40

“time machines” across Europe rather than having an expert extract Is it real?
that have now been established to information to feed into the net- n YouTube, viewers can
reconstruct local history from dig-
itized records. The Venetian state
archives cover 1,000 years of his-
work as in his work on the bish-
ops—and says it produces a lot of
“artificial complexity” but nothing
O now watch Richard Nixon
make a speech that had
been written in case the 1969 moon
tory spread across 80 kilometers that serves in historical interpreta- landing ended in disaster but fortu-
of shelves; the researchers’ aim was tion. The algorithm was unable to nately never needed to be delivered.
to digitize these records, many of distinguish instances where two Researchers created the deepfake
which had never been examined by people’s names appeared on the to show how AI could affect our
modern historians. They would use same roll of taxpayers from cases shared sense of history. In seconds,
deep-learning networks to extract where they were on a marriage cer- one can generate false images of
information and, by tracing names tificate, so as Preiser-Kapeller says, major historical events like the
that appear in the same document “What you really get has no explan- D-Day landings, as Northeastern
across other documents, reconstruct atory value.” It’s a limitation histori- history professor Dan Cohen dis-
the ties that once bound Venetians. ans have highlighted with machine cussed recently with students in
Frédéric Kaplan, president of learning, similar to the point people a class dedicated to exploring the
the Time Machine Organization, have made about large language way digital media and technology
says the project has now digitized models like ChatGPT: because mod- are shaping historical study. “[The
enough of the city’s administrative els ultimately don’t understand what photos are] entirely convincing,” he
documents to capture the texture of they’re reading, they can arrive at says. “You can stick a whole bunch
the city in centuries past, making it absurd conclusions. of people on a beach and with a tank
possible to go building by building It’s true that with the sources that and a machine gun, and it looks
and identify the families who lived are currently available, human inter- perfect.”
there at different points in time. pretation is needed to provide con- False history is nothing new—
“These are hundreds of thousands text, says Kaplan, though he thinks Cohen points to the way Joseph
of documents that need to be digi- this could change once a sufficient Stalin ordered enemies to be
tized to reach this form of flexibility,” number of historical documents are erased from history books, as an
says Kaplan. “This has never been made machine readable. example—but the scale and speed
done before.” But he imagines an application of with which fakes can be created is
Still, when it comes to the proj- machine learning that’s more trans- breathtaking, and the problem goes
ect’s ultimate promise—no less formational—and potentially more beyond images. Generative AI can
than a digital simulation of medi- problematic. Generative AI could create texts that read plausibly like
eval Venice down to the neigh- be used to make predictions that a parliamentary speech from the
borhood level, through networks flesh out blank spots in the histori- Victorian era, as Cohen has done
reconstructed by artificial intelli- cal record—for instance, about the with his students. By generating
gence—historians like Johannes number of apprentices in a Venetian historical handwriting or typefaces,
Preiser-Kapeller, the Austrian artisan’s workshop—based not on it could also create what looks con-
Academy of Sciences professor who individual records, which could be vincingly like a written historical
ran the study of Byzantine bishops, inaccurate or incomplete, but on record.
say the project hasn’t been able to aggregated data. This may bring Meanwhile, AI chatbots like
deliver because the model can’t more non-elite perspectives into Character.ai and Historical Figures
understand which connections are the picture but runs counter to stan- Chat allow users to simulate inter-
meaningful. dard historical practice, in which actions with historical figures.
Preiser-Kapeller has done his conclusions are based on available Historians have raised concerns
own experiment using automatic evidence. about these chatbots, which may,
detection to develop networks from Still, a more immediate concern for example, make some individuals
documents—extracting network is posed by neural networks that seem less racist and more remorse-
information with an algorithm, create false records. ful than they actually were.
 41

False history is nothing new— This “black box” problem is not

unique to history: even developers
but the scale and speed with which fakes of machine-learning systems some-
can be created is breathtaking. times struggle to understand how
they function. Fortunately, some
methods designed with histori-
ans in mind are structured to pro-
vide greater transparency. Ithaca
produces a range of hypotheses
ranked by probability, and BIFOLD
In other words, there’s a risk that researchers are working on the
artificial intelligence, from histor- interpretation of their models with
Days of future past ical chatbots to models that make explainable AI, which is meant to
Three key projects underway predictions based on historical reveal which inputs contribute most
in the digital humanities records, will get things very wrong. to predictions. Historians say they
Some of these mistakes are benign themselves promote transparency by
anachronisms: a query to Aristotle encouraging people to view machine
CorDeep on the chatbot Character.ai about learning with critical detachment: as
Who: Max Planck Institute for the History his views on women (whom he saw a useful tool, but one that’s fallible,
of Science as inferior) returned an answer just like people.
What: A web-based application for classifying that they should “have no social
content from historical documents that media.” But others could be more The historians
include numerical and alphanumerical consequential—especially when of tomorrow
tables. Software can locate, extract, they’re mixed into a collection of hile skepticism toward
and classify visual elements desig-
nated “content illustrations,” “initials,”
“decorations,” and “printer’s marks.”
documents too large for a historian
to be checking individually, or if
they’re circulated by someone with
W such new technology per-
sists, the field is gradually
embracing it, and Valleriani thinks
an interest in a particular interpre- that in time, the number of histori-
tation of history. ans who reject computational meth-
Ithaca Even if there’s no deliberate ods will dwindle. Scholars’ concerns
Who: DeepMind deception, some scholars have about the ethics of AI are less a rea-
What: A deep neural network trained to concerns that historians may use son not to use machine learning,
simultaneously perform the tasks of tools they’re not trained to under- he says, than an opportunity for
textual restoration, geographic attri- stand. “I think there’s great risk in the humanities to contribute to its
bution, and chronological attribution, it, because we as humanists or his- development.
previously performed by epigraphers. torians are effectively outsourcing A s t h e Fr e n c h h i s t o r i a n
analysis to another field, or perhaps Emmanuel Le Roy Ladurie wrote
a machine,” says Abraham Gibson, in 1968, in response to the work of
Venice Time Machine Project a history professor at the University historians who had started experi-
Who: École Polytechnique Fédérale de of Texas at San Antonio. Gibson menting with computational history
Lausanne, Ca’ Foscari, and the State says until very recently, fellow his- to investigate questions such as
Archives of Venice torians he spoke to didn’t see the voting patterns of the British par-
What: A digitized collection of the Venetian relevance of artificial intelligence to liament in the 1840s, “the historian
state archives, which cover 1,000 their work, but they’re increasingly of tomorrow will be a programmer,
years of history. Once it’s completed, waking up to the possibility that or he will not exist.”
researchers will use deep learning to they could eventually yield some
reconstruct historical social networks. of the interpretation of history to
Moira Donovan is an independent
science journalist based in
a black box. Halifax, Nova Scotia.
42

The panicked reaction around cheating in school

doesn’t tell the whole story. Meet the teachers
who think AI could actually make learning better.

The education of
ChatGPT
BY Will Douglas Heaven ILLUSTRATIONS Selman Design

he response from schools and Virginia in the United States to This initial panic from the edu-
T and universities was swift
and decisive.
Queensland and New South Wales
in Australia.
cation sector was understandable.
ChatGPT, available to the public
Just days after OpenAI dropped Several leading universities via a web app, can answer questions
ChatGPT in late November 2022, in the UK, including Imperial and generate slick, well-structured
the chatbot was widely denounced College London and the University blocks of text several thousand
as a free essay-writing, test-taking of Cambridge, issued statements words long on almost any topic it
tool that made it laughably easy to that warned students against using is asked about, from string theory
cheat on assignments. ChatGPT to cheat. to Shakespeare. Each essay it pro-
Los Angeles Unified, the second- “While the tool may be able to duces is unique, even when it is
largest school district in the US, provide quick and easy answers to given the same prompt again, and
immediately blocked access to questions, it does not build critical- its authorship is (practically) impos-
OpenAI’s website from its schools’ thinking and problem-solving skills, sible to spot. It looked as if ChatGPT
network. Others soon joined. By which are essential for academic would undermine the way we test
January, school districts across and lifelong success,” Jenna Lyle, a what students have learned, a cor-
the English-speaking world had spokeswoman for the New York City nerstone of education.
started banning the software, from Department of Education, told the But three months on, the out-
Washington, New York, Alabama, Washington Post in early January. look is a lot less bleak. I spoke to a
44

number of teachers and other educa- It is far too soon to say what the thinking, ‘Okay, it’s here. How can
tors who are now reevaluating what lasting impact of ChatGPT will be— we use it?’”
chatbots like ChatGPT mean for how it hasn’t even been around for a full “It was a storm in a teacup,” says
we teach our kids. Far from being semester. What’s certain is that David Smith, a professor of biosci-
just a dream machine for cheat- essay-writing chatbots are here to ence education at Sheffield Hallam
ers, many teachers now believe, stay. And they will only get better at University in the UK. Far from
ChatGPT could actually help make standing in for a student on dead- using the chatbot to cheat, Smith
education better. line—more accurate and harder says, many of his students hadn’t
Advanced chatbots could be to detect. Banning them is futile, yet heard of the technology until
used as powerful classroom aids possibly even counterproductive. he mentioned it to them: “When I
that make lessons more interac- “We need to be asking what we started asking my students about it,
tive, teach students media liter- need to do to prepare young peo- they were like, ‘Sorry, what?’”
acy, generate personalized lesson ple—learners—for a future world Even so, teachers are right
plans, save teachers time on admin, that’s not that far in the future,” to see the technology as a game
and more. says Richard Culatta, CEO of the changer. Large language models like
Educational-tech companies International Society for Technology OpenAI’s ChatGPT and its succes-
including Duolingo and Quizlet, in Education (ISTE), a nonprofit that sor GPT-4, as well as Google’s Bard
which makes digital flash cards advocates for the use of technology and Microsoft’s Bing Chat, are set
and practice assessments used by in teaching. to have a massive impact on the
half of all high school students in Tech’s ability to revolutionize world. The technology is already
the US, have already integrated schools has been overhyped in the being rolled out into consumer and
OpenAI’s chatbot into their apps. past, and it’s easy to get caught up business software. If nothing else,
And OpenAI has worked with edu- in the excitement around ChatGPT’s many teachers now recognize that
cators to put together a fact sheet transformative potential. But this they have an obligation to teach their
about ChatGPT’s potential impact feels bigger: AI will be in the class- students about how this new tech-
in schools. The company says it also room one way or another. It’s vital nology works and what it can make
consulted educators when it devel- that we get it right. possible. “They don’t want it to be
oped a free tool to spot text written In February, vilified,” says Smith. “They want to
by a chatbot (though its accuracy is From ABC to GPT a survey of K–12 be taught how to use it.”
teachers and
limited). uch of the early hype around Change can be hard. “There’s still
teenage stu-
“We believe that educational
policy experts should decide what
M ChatGPT was based on how
good it is at test taking. In
dents in the US
some fear,” says Stansbury. “But we
do our students a disservice if we
found that
works best for their districts and fact, this was a key point OpenAI MORE THAN get stuck on that fear.”
schools when it comes to the use touted when it rolled out GPT-4, Stansbury has helped organize
of new technology,” says Niko the latest version of the large lan-
1/2 workshops at her university to allow
OF THE
Felix, a spokesperson for OpenAI. guage model that powers the chat- TEACHERS faculty and other teaching staff to
“We are engaging with educators bot, in March. It could pass the bar had used share their experiences and voice
across the country to inform them exam! It scored a 1410 on the SAT! ChatGPT their concerns. She says that some
of ChatGPT’s capabilities. This is an It aced the AP tests for biology, art compared with of her colleagues turned up wor-
important conversation to have so history, environmental science, mac- ONLY ried about cheating, others about
that they are aware of the potential
benefits and misuse of AI, and so
roeconomics, psychology, US his-
tory, and more. Whew!
1/3 losing their jobs. But talking it out
helped. “I think some of the fear
OF THE
they understand how they might It’s little wonder that some school STUDENTS that faculty had was because of the
apply it to their classrooms.” districts totally freaked out. media,” she says. “It’s not because
But it will take time and Yet in hindsight, the immediate of the students.”
resources for educators to inno- calls to ban ChatGPT in schools In fact, a US survey of 1,002
vate in this way. Many are too were a dumb reaction to some very K–12 teachers and 1,000 students
overworked, under-resourced, and smart software. “People panicked,” between 12 and 17, commissioned
beholden to strict performance says Jessica Stansbury, director of by the Walton Family Foundation
metrics to take advantage of any teaching and learning excellence at in February, found that more
opportunities that chatbots may the University of Baltimore. “We had than half the teachers had used
present. the wrong conversations instead of ChatGPT—10% of them reported
 45

education,” she says. “We’ve been use ChatGPT to generate an argu-

talking about it for years.” ment and then had them annotate
Take cheating. In Crompton’s it according to how effective they
view, if ChatGPT makes it easy to thought the argument was for a spe-
cheat on an assignment, teachers cific audience. Then they turned in
should throw out the assignment a rewrite based on their criticism.
rather than ban the chatbot. Breaking down the assignment
We need to change how we in this way also helps students focus
assess learning, says Culatta: “Did on specific skills without getting

Cheating is not a new ChatGPT kill assessments? They

were probably already dead, and
sidetracked. Donahoe found, for
example, that using ChatGPT to
problem: schools have they’ve been in zombie mode for a
long time. What ChatGPT did was
generate a first draft helped some
students stop worrying about the
survived calculators, call us out on that.” blank page and instead focus on the
critical phase of the assignment. “It
Google, Wikipedia, Critical thinking can help you move beyond partic-

essays-for-pay E
mily Donahoe, a writing
tutor and educational devel-
ular pain points when those pain
points aren’t necessarily part of the
websites, and more. oper at the University of
Mississippi, has noticed classroom
learning goals of the assignment,”
she says.
discussions starting to change in Smith, the bioscience professor,
the months since ChatGPT’s release. is also experimenting with ChatGPT
Although she first started to talk to assignments. The hand-wringing
her undergraduate students about around it reminds him of the anx-
the technology out of a sense of duty, iety many teachers experienced
she now thinks that ChatGPT could a couple of years ago during the
using it every day—but only a third help teachers shift away from an pandemic. With students stuck at
of the students. Nearly all those excessive focus on final results. home, teachers had to find ways to
who had used it (88% of teachers Getting a class to engage with AI set assignments where solutions
and 79% of students) said it had a and think critically about what it were not too easy to Google. But
positive impact. generates could make teaching feel what he found was that Googling—
A majority of teachers and stu- more human, she says, “rather than what to ask for and what to make of
dents surveyed also agreed with this asking students to write and perform the results—was itself a skill worth
statement: “ChatGPT is just another like robots.” teaching.
example of why we can’t keep doing This idea isn’t new. Generations Smith thinks chatbots could be
things the old way for schools in the of teachers have subscribed to a the same way. If his undergraduate
modern world.” framework known as Bloom’s taxon- students want to use ChatGPT in
Helen Crompton, an associate omy, introduced by the educational their written assignments, he will
professor of instructional technol- psychologist Benjamin Bloom in the assess the prompt as well as—or
ogy at Old Dominion University in 1950s, in which basic knowledge of even rather than—the essay itself.
Norfolk, Virginia, hopes that chat- facts is just the bedrock on which “Knowing the words to use in a
bots like ChatGPT will make school other forms of learning, such as prompt and then understanding the
better. analysis and evaluation, sit. Teachers output that comes back is import-
Many educators think that like Donahoe and Crompton think ant,” he says. “We need to teach
schools are stuck in a groove, that chatbots could help teach those how to do that.”
says Crompton, who was a K–12 other skills.
teacher for 16 years before becom- In the past, Donahoe would set The new education
ing a researcher. In a system with her students to writing assignments hese changing attitudes
too much focus on grading and not
enough on learning, ChatGPT is
in which they had to make an argu-
ment for something—and grade
T reflect a wider shift in the
role that teachers play, says
forcing a debate that is overdue. them on the text they turned in. This Stansbury. Information that was
“We’ve long wanted to transform semester, she asked her students to once dispensed in the classroom
46

is now everywhere: first online,

then in chatbots. What educators
must now do is show students not
only how to find it, but what infor-
mation to trust and what not to,
and how to tell the difference.
“Teachers are no longer gatekeep-
ers of information, but facilitators,”
she says.
In fact, teachers are finding
opportunities in the misinformation
and bias that large language models
often produce. These shortcomings
can kick off productive discussions,
says Crompton: “The fact that it’s
not perfect is great.”
Teachers are asking students to
use ChatGPT to generate text on a
topic and then getting them to point
out the flaws. In one example that a
colleague of Stansbury’s shared at
her workshop, students used the
bot to generate an essay about the
history of the printing press. When
its US-centric response included
no information about the origins of
print in Europe or China, the teacher
used that as the starting point for a
conversation about bias. “It’s a great
way to focus on media literacy,” says
Stansbury.
Crompton is working on a
study of ways that chatbots can
improve teaching. She runs off a
list of potential applications she’s
excited about, from generating test
questions to summarizing infor-
mation for students with differ-
ent reading levels to helping with
time- consuming administrative
tasks such as drafting emails to
colleagues and parents.
One of her favorite uses of the
technology is to bring more interac-
tivity into the classroom. Teaching
methods that get students to be
creative, to role-play, or to think
critically lead to a deeper kind of
learning than rote memorization,
she says. ChatGPT can play the role
of a debate opponent and generate
counterarguments to a student’s
 47

learning preferences. Teachers be a jumping-off point rather than

might prepare a few different ver- a crutch.”
sions of their teaching materials to And, of course, some students
cover a range of students’ needs. will still use ChatGPT to cheat. In
Culatta thinks that chatbots could fact, it makes it easier than ever.
generate personalized material for With a deadline looming, who
50 or 100 students and make wouldn’t be tempted to get that
bespoke tutors the norm. “I think assignment written at the push of
in five years the idea of a tool that a button? “It equalizes cheating for

Teachers are asking gives us information that was writ-

ten for somebody else is going to
everyone,” says Crompton. “You
don’t have to pay. You don’t have to
students to use feel really strange,” he says.
Some ed-tech companies are
hack into a school computer.”
Some types of assignments will be
ChatGPT to generate already doing this. In March, harder hit than others, too. ChatGPT
Quizlet updated its app with a is really good at summarizing infor-
text on a topic and then feature called Q-Chat, built using mation. When that is the goal of an

getting them to point ChatGPT, that tailors material to

each user’s needs. The app adjusts
assignment, cheating is a legitimate
concern, says Donahoe: “It would
out the flaws. the difficulty of the questions
according to how well students
be virtually indistinguishable from
an A answer in that context. It is
know the material they’re study- something we should take seriously.”
ing and how they prefer to learn. None of the educators I spoke to
“Q-Chat provides our students with have a fix for that. And not all other
an experience similar to a one- fears will be easily allayed. (Donahoe
on-one tutor,” says Quizlet’s CEO, recalls a recent workshop at her uni-
Lex Bayer. versity in which faculty were asked
positions, for example. By expos- In fact, some educators think what they were planning to do differ-
ing students to an endless supply of future textbooks could be bundled ently after learning about ChatGPT.
opposing viewpoints, chatbots could with chatbots trained on their con- One faculty member responded: “I
help them look for weak points in tents. Students would have a conver- think I’ll retire.”)
their own thinking. sation with the bot about the book’s But nor are teachers as wor-
Crompton also notes that if contents as well as (or instead of) ried as initial reports suggested.
English is not a student’s first lan- reading it. The chatbot could gen- Cheating is not a new problem:
guage, chatbots can be a big help erate personalized quizzes to coach schools have survived calculators,
in drafting text or paraphrasing students on topics they understand Google, Wikipedia, essays-for-pay
existing documents, doing a lot to less well. websites, and more.
level the playing field. Chatbots also Not all these approaches will For now, teachers have been
serve students who have specific be instantly successful, of course. thrown into a radical new experi-
learning needs, too. Ask ChatGPT Donahoe and her students came up ment. They need support to figure
to explain Newton’s laws of motion with guidelines for using ChatGPT it out—perhaps even government
to a student who learns better with together, but “it may be that we support in the form of money, train-
images rather than words, for exam- get to the end of this class and I ing, and regulation. But this is not
ple, and it will generate an expla- think this absolutely did not work,” the end of education. It’s a new
nation that features balls rolling she says. “This is still an ongoing beginning.
on a table. experiment.” “We have to withhold some of
She has also found that students our quick judgment,” says Culatta.
Made-to-measure need considerable support to make “That’s not helpful right now. We
learning sure ChatGPT promotes learning need to get comfortable kicking the
ll students can benefit from rather than getting in the way of it. tires on this thing.”
A personalized teaching
materials, says Culatta,
Some students find it harder to move
beyond the tool’s output and make
Will Douglas Heaven is a senior
editor for AI at MIT Technology
because everybody has different it their own, she says: “It needs to Review.
48

A
decade ago, tech powerhouses the likes of Microsoft,
Google, and Amazon helped boost the nonprofit
Code.org, a learn-to-code program with a vision:
“That every student in every school has the oppor-
tunity to learn computer science as part of their core
K–12 education.” It was followed by a wave of non-
profits and for-profits alike dedicated to coding and
learning computer science; some of the many others
include Codecademy, Treehouse, Girl Develop It, and
Hackbright Academy (not to mention Girls Who Code,
founded the year before Code.org and promising
participants, “Learn to code and change the world”).
Parents can now consider top-10 lists of coding sum-
mer camps for kids. Some may choose to start their
children even younger, with the Baby Code! series
of board books—because “it’s never too early to get
little ones interested in computer coding.” Riding this
wave of enthusiasm, in 2016 President Barack Obama
launched an initiative called Computer Science for
All, proposing billions of dollars in funding to arm
students with the “computational thinking skills they
need” to “thrive in a digital economy.”

The
long
history
of
“learn
to
code”
MIT MUSEUM

By Joy Lisi Rankin

 49

Computer scientist
Seymour Papert
created the Logo Turtle to
help cure what he termed
“mathphobia.”
50

Dartmouth:
Building a BASIC computing community
When mathematics professor (and future Dartmouth
president) John Kemeny made a presentation to col-
lege trustees in the early 1960s hoping to persuade
them to fund a campus-wide computing network, he
emphasized the idea that Dartmouth students (who
were at that time exclusively male, and mostly afflu-
ent and white) were the future leaders of the United
States. Kemeny argued, “Since many students at an
institution like Dartmouth become executives or key
policy makers in industry and government, it is a
certainty that they will have at their command high-
speed computing equipment.”
Kemeny claimed that it was “essential” for those
Above: At Dartmouth,
nascent power brokers to “be acquainted with the Right: Kemeny, the
mathematics profes- potential and limitations of high-speed computers.” co-creator of the
sors Thomas Kurtz In 1963 and 1964, he and fellow mathematics profes- programming lan-
(left) and John Kemeny sor Thomas Kurtz worked closely with Dartmouth guage BASIC, believed
pioneered the use of it was essential for
students to design and implement a campus-wide
computers in college his students to “be
education.
network, while Kemeny largely took responsibility for acquainted with the
designing an easy-to-learn programming language, potential and limita-
called BASIC, for students (and faculty) to use on that tions of high-speed
Now, in 2023, North Carolina is considering mak- network. Both developments were eagerly welcomed computers.”

ing coding a high school graduation requirement. If by the incoming students in the fall of 1964.
lawmakers enact that curriculum change, they will As Dartmouth’s network grew during the 1960s,
be following in the footsteps of five other states with network terminals were installed in the new campus
similar policies that consider coding and computer computer center, in shared campus recreational spaces
education foundational to a well-rounded education: and dormitories, and at other locations around campus.
Nevada, South Carolina, Tennessee, Arkansas, and And because the system was set up as a time-sharing
Nebraska. Advocates for such policies contend that network, an innovation at the time, multiple terminals
they expand educational and economic opportunities could be connected to the same computer, and the
for students. More and more jobs, they suggest, will people using those terminals could write and debug
require “some kind of computer science knowledge.” programs simultaneously.
This enthusiasm for coding is nothing new. In 1978 This was transformative: by 1968, 80% of Dartmouth
Andrew Molnar, an expert at the National Science undergraduates and 40% of the faculty used the net-
Foundation, argued that what he termed computer work regularly. Although incoming students learned
literacy was “a prerequisite to effective participa- how to write a program in BASIC as a first-year math
tion in an information society and as much a social course requirement, what really fostered the comput-
obligation as reading literacy.” Molnar pointed as ing culture was the way students made the language
models to two programs that had originated in the and the network their own. For example, the impor-
1960s. One was the Logo project centered at the MIT tance of football in campus life (Dartmouth claimed
Artificial Intelligence Lab, which focused on exposing the Ivy League championship seven times between
DARTMOUTH COLLEGE/RAUNER SPECIAL COLLECTIONS LIBRARY

elementary-age kids to computing. (MIT Technology 1962 and 1971) inspired at least three computer foot-
Review is funded in part by MIT but maintains edi- ball games (FTBALL, FOOTBALL, and GRIDIRON)
torial independence.) The other was at Dartmouth played avidly on the Dartmouth network, one of them
College, where undergraduates learned how to write written by Kemeny himself.
programs on a campus-wide computing network. Because the network was so easy to access and
The Logo and Dartmouth efforts were among several BASIC was so easy to use, Dartmouth students could
computing-related educational endeavors organized make computing relevant to their own lives and inter-
from the 1960s through 1980s. But these programs, and ests. One wrote a program to test a hypothesis for a
many that followed, often benefited the populations with psychology class. Another ran a program called XMAS
the most power in society. Then as now, just learning to to print his Christmas cards. Some printed out letters
code is neither a pathway to a stable financial future for to parents or girlfriends. Others enjoyed an array of
people from economically precarious backgrounds nor a games, including computer bridge, checkers, and
panacea for the inadequacies of the educational system. chess. Although learning to write a program in BASIC
 51

was the starting point in computing for Dartmouth

students, the ways they used it to meet their own
needs and forge community with their peers made
the system a precursor of social networking—nearly

What was intended as half a century ago. Coding in BASIC didn’t replace
their liberal arts curriculum requirements or extra-

computing for all ultimately curricular activities; rather, it complemented them.

amplified existing Different results:

The Dartmouth network expands

inequities. As it grew in popularity, other schools around New

England sought to tap into Dartmouth’s computing
network for their students. By April 1971, the network
encompassed 30 high schools and 20 colleges in New
England, New York, and New Jersey. All an individual
school needed to connect were a terminal and a tele-
phone line linking the terminal with the mainframe
on Dartmouth’s campus (often the greatest expense
of participating in the network, at a time when long-
distance phone calls were quite costly). Yet as BASIC
moved beyond Dartmouth into heterogeneous high
schools around New England, the computing culture
remained homogeneous.
Private high schools including Phillips Exeter,
Phillips Andover, and St. Paul’s were among the first
to connect, all before 1967. Within a few more years,
a mix of private and public high schools joined them.
The Secondary School Project (SSP), which ran from
1967 to 1970 and was supported by a three-year NSF
grant secured by Kemeny and Kurtz, connected stu-
dents and educators at 18 public and private high
schools from Connecticut to Maine, with the goal of
putting computing access (and BASIC) into as many
hands as possible and observing the results.
That these schools asked Dartmouth for time shares
reflected interest and motivation on the part of some
individual or group at each one. They wanted network
access—and, by extension, access to code—because it
was novel and elite. Some students were enthusiastic
users, even waking at four in the morning to sign on.
But access to the Dartmouth network was emphati-
cally unequal. The private schools participating in the
SSP were (at the time) all male and almost exclusively
white, and those students enjoyed nearly twice as much
network time as the students at coeducational public
schools: 72 hours per week for private school students,
and only 40 for public school students.
In these years before the expansion of educational
opportunities for girls and women in the United States,
high school boys were enrolling in many more math
and science classes than high school girls. The math
and science students gained access to computing in
those courses, meaning that BASIC moved into a sys-
tem already segregated by gender—and also by race.
What was intended as computing for all ultimately
amplified existing inequities.
 53

Logo:
Trying to change the world, one turtle at a time
One state away from Dartmouth, the Logo project,
founded by Seymour Papert, Cynthia Solomon, and
Wally Feurzeig, sought to revolutionize how elemen-
tary and middle school students learn. Initially, the
researchers created a Logo programming language
and tested it between 1967 and 1969 with groups of
children including fifth and seventh graders at schools
near MIT in Cambridge, Massachusetts. “These kids
made up hilarious sentence generators and became
proficient users of their own math quizzes,” Solomon
has recalled.
But Logo was emphatically not just a “learn to
code” effort. It grew to encompass an entire lab and
Left: In the ’60s, Dart-
a comprehensive learning system that would intro- Above: Logo’s founda- to program the Logo
mouth students had duce new instructional methods, specially trained tional philosophy was robot (above) as a
unprecedented com- teachers, and physical objects to think and play with. for the child to pro- pathway to larger
puter access thanks Perhaps the best-remembered of those objects is the gram the computer concepts and ideas.
to a time-sharing net- rather than be taught
Logo Turtle, a small robot that moved along the floor,
work that connected by it. Students learned
multiple terminals via
directed by computer commands, with a retractable
telephone line to a pen underneath its body that could be lowered to
central computer. draw shapes, pictures, and patterns.
By the early 1970s, the Logo group was part of the curriculum—something that would only have been
MIT AI Lab, which Papert had cofounded with the possible at a small private school like this one.
computer scientist Marvin Minsky. The kid-focused The Lamplighter project—and the publication
learning environment provided a way to write sto- around the same time of Papert’s book Mindstorms,
ries, a way to draw, a way to make music, and a way to in which the mathematician enthused about the
explore a space with a programmable object. Papert promise of computing to revolutionize education—
imagined that the Logo philosophy would empower marked a high point for Logo. But those creative
children as “intellectual agents” who could derive educational computing initiatives were short-lived. A
their own understanding of math concepts and cre- major obstacle was simply the incredibly slow-moving
ate connections with other disciplines ranging from and difficult-to-change bureaucracy of American
psychology and the physical sciences to linguistics public education. Moreover, promising pilots either
and logic. did not scale or were unable to achieve the same
But the reality outside the MIT AI Lab challenged results when introduced into a system fraught with
that vision. In short, teaching Logo to elemen- resource inequities.
tary school students was both time- and resource- But another issue was that the increasingly wide-
intensive. In 1977-’78, an NSF grant funded a yearlong spread availability of personal computers by the
study of Logo at a public school; it was meant to 1980s challenged Logo’s revolutionary vision. As
include all the school’s sixth graders, but the grant computers became consumer objects, software did,
DARTMOUTH COLLEGE/RAUNER SPECIAL COLLECTIONS LIBRARY; MIT MUSEUM

covered only four computers, which meant that only too. People no longer needed to learn to code to
four students could participate at the same time. be able to use a computer. In the case of American
The research team found that most of the students education, computers in the classroom became less
who were chosen to participate did learn to create about programming and more about educational
programs and express math concepts using Logo. games, word processing, and presentations. While
However, when the study ended and the students BASIC and Logo continued to be taught in some
moved on, their computing experiences were largely schools around the United States, for many students
left in the past. the effort of writing some code to, say, alphabetize
As that project was wrapping up, the Logo team a list seemed impractical—disconnected from their
implemented a larger-scale partnership at the pri- everyday lives and their imagined futures.
vate Lamplighter School in Dallas, cosponsored by
Texas Instruments. At this school, with a population Corporate coding
of 450 students in kindergarten through fourth grade, Schools weren’t the only setting for learn-to-code
50 computers were available. Logo was not taught movements, however. In the 1960s the Association
as a standalone subject but was integrated into the for Computing Machinery (ACM), which had been
54

established as a professional organization in the panacea for racial inequality and the social instability
1940s, spearheaded similar efforts to teach coding it fueled. A group from a Delaware ACM chapter, a
to young people. From 1968 to 1972, ACM members conference report suggested, believed that “in these
operating through their local chapters established days of urban crisis, the data processing industry
programs across the United States to provide training offers a unique opportunity to the disadvantaged to
in computing skills to Black and Hispanic Americans. become involved in the mainstream of the American
During the same years, government and social wel- way of life.”
fare organizations offered similar training, as did If success is defined as getting a steadily increasing
companies including General Electric. There were at number of Black and Hispanic men and women good
least 18 such programs in East Coast and California jobs in the computing profession—and, by extension,
cities and one in St. Louis, Missouri. Most, but not giving them opportunities to shape and inform the
all, targeted young people. In some cases, the pro- technologies that would remake the world—then
grams taught mainframe or keypunch operation, but these programs failed. As the scholar Arvid Nelsen
others aimed to teach programming in the common observed, while some volunteers “may have been
business computing languages of the time, COBOL focused on the needs and desires of the communities
and FORTRAN. themselves,” others were merely seeking a Band-Aid Below: Activist and edu-
Did the students in these programs learn? The for “civil unrest.” Meanwhile, Nelsen notes, businesses cator Robert P. Moses
answer was emphatically yes. Could they get jobs as a benefited from “a source of inexpensive workers with established the Algebra
result, or otherwise use their new skills? The answer much more limited power.” In short, training people Project in the early ’80s
to address racial and
to that was often no. A program in San Diego arranged to code didn’t mean they would secure better, higher-
economic inequities in
for Spanish-speaking instructors and even converted paying, more stable jobs—it just meant that there math education.
a 40-foot tractor-trailer into a mobile training facil- was a larger pool of possible entry-level employees
ity so that students—who were spread across the who would drive down labor costs for the growing
sprawling city—would not have to spend upwards computer industry.
of an hour commuting by bus to a central location. In fact, observers identified the shortcomings of
And in the Albany-Schenectady area of New York, these efforts even at the time. Walter DeLegall, a
General Electrical supported a rigorous program to Black computing professional at Columbia University,
prepare Black Americans for programming jobs. It was declared in 1969 that the “magic of data processing
open to people without high school diplomas, and to training” was no magic bullet, and that quick-fix train-
people with police records; there was no admissions ing programs mirrored the deficiencies of American
testing. Well over half the people who started this public education for Black and Spanish-speaking stu-
training completed it. dents. He questioned the motivation behind them,
Yet afterwards many could not secure jobs, even suggesting that they were sometimes organized for
entry-level ones. In other cases, outstanding gradu- “commercial reasons or simply to de-fuse and dissi-
ates were offered jobs that paid $105 per week—not pate the burgeoning discontent of these communities”
enough to support themselves and their families. One rather than to promote equity and justice.
consultant to the project suggested that for future
training programs, GE should “give preference to The Algebra Project
younger people without families” to minimize labor There was a grassroots effort that did respond to these
costs for the company. inadequacies, by coming at the computing revolution
The very existence of these training endeavors from an entirely different angle.
reflected a mixed set of motivations on the part of During the late 1970s and early 1980s, the civil
the organizers, who were mostly white, well-off vol- rights activist Robert P. Moses was living with his
unteers. These volunteers tended to conflate living in family in Cambridge, Massachusetts, where his daugh-
an urban area with living in poverty, and to assume ter Maisha attended the public Martin Luther King
that people living in these conditions were not white, School and he volunteered teaching algebra. He
and that all such people could be lumped together noticed that math groups were unofficially segregated
DAVID RAE MORRIS

under the heading of “disadvantaged.” They imagined by race and class, and that much less was expected of
that learning to code would provide a straightfor- Black and brown students. Early on, he also identi-
ward path out of poverty for these participants. But fied computers—and knowledge work dependent on
their thinking demonstrated little understanding of computers—as a rising source of economic, political,
the obstacles imposed by centuries of enslavement, and social power. Attending college was increasingly
unpaid labor, Jim Crow violence, pay discrimination, important for attaining that kind of power, and Moses
and segregated and unequal education, health care, saw that one key to getting there was a foundation
and housing. Largely with their own interests in in high school mathematics, particularly algebra.
mind, they looked to these upskilling programs as a He established the Algebra Project during the early
 55

1980s, beginning in Cambridge public schools and

supported by a MacArthur “genius grant” that he
received in 1982.
In a book that he later coauthored, Radical

Arming Black students Equations: Civil Rights from Mississippi to the Algebra
Project, Moses clearly articulated the connections

with the tools of math between math, computing, economic justice, and
political power, especially for Black Americans. “The

literacy was radical most urgent social issue affecting poor people and
people of color is economic access. In today’s world,

in the 1980s precisely economic access and full citizenship depend cru-
cially on math and science literacy,” he wrote. “The

because it challenged computer has become a cultural force as well as an

instrument of work [and] while the visible manifes-

power dynamics. tation of the technological shift is the computer, the

hidden culture of computers is math.”
Moses had earned his bachelor’s degree at Hamilton
College in New York and a master’s degree at Harvard
University before teaching math at the Horace Mann
School in the Bronx from 1958 to 1961. For him, arm-
ing Black students with the tools of math literacy
was radical in the 1980s precisely because access to
technology meant access to power. “Who’s going to
gain access to the new technology?” he asked. “Who’s
going to control it? What do we have to demand of
the educational system to prepare for the new tech-
nological era?”
Moses mobilized students and parents alike to
ensure that algebra was offered to all students at the
Martin Luther King School. He devised new approaches
to teaching the subject, and drawing on his experi-
ence with grassroots civil rights organizing, enrolled
students to teach their peers. College admission rates
and test scores rose at the school, and the Algebra
Project spread to at least 22 other sites across 13 states.
It focused on math because Moses identified math as
the foundation of coding, and the stakes were always
connected to economic justice and educational equity
in an economy built on algorithms and data.
Moses made explicit “a number of issues that are
often hidden in coding discourse,” the historian Janet
Abbate has observed. “He questioned the implied
meritocracy of ‘ability grouping’ … he attacked the
stereotype that Black people aren’t interested in
STEM … [and] he emphasized that social skills and
community were an essential part of overcoming
students’ alienation from technology.”
Moses died in 2021, but the Algebra Project lives
on, now in collaboration with a group called the “We
the People” Math Literacy for All Alliance. The cur-
riculum he pioneered continues to be taught, and the
Algebra Project’s 2022 conference again called atten-
tion to the need for better public education across the
United States, especially for Black, brown, and poor
children, “to make full participation in American
democracy possible.”
56

Rewind, reboot:
Coding makes a comeback
In the past decade, a new crop of more targeted
coding programs has emerged. In 2014, for example,
the activist and entrepreneur Van Jones collaborated
with the musician Prince to launch #YesWeCode, Technological solutionism
targeting what they called “low-opportunity commu-
nities.” In doing so, they called attention to ongo- may persist, but there’s
ing educational and economic inequities across the
United States. an increasing recognition
One of #YesWeCode’s early efforts was a
youth-oriented hackathon at the Essence Music that coding training alone
Festival in New Orleans in 2014 that encouraged
kids to connect coding with issues that mattered to is not enough.
them. As #YesWeCode’s chief innovation officer,
Amy Henderson, explained, “A lot of the people who
develop apps today are affluent white men, and so
they build apps that solve their communities’ prob-
lems,” such as Uber. “Meanwhile,” she continued,
“one of our young people built an app that sends
reminders of upcoming court dates. That’s an issue
that impacts his community, so he did something
about it.”
#YesWeCode has since morphed into Dream.Tech,
an arm of Dream.org, a nonprofit that advocates for
new legislation and new economic policies to rem-
edy global climate change, the racialized mass incar-
ceration system in the United States, and America’s
long history of poverty. (Its other arms are called
Dream.Green and Dream.Justice.) Recently, for exam-
ple, Dream.org pushed for legislation that would
erase long-standing racial disparities in sentencing
for drug crimes. As a whole, Dream.org demonstrates
an expansive vision of tech justice that can “make the
future work for everyone.”
Another initiative, called Code2040 (the name
refers to the decade during which people of color
are expected to become a demographic majority in
the United States), was launched in 2012. It initially
focused on diversifying tech by helping Black and
Latino computer science majors get jobs at tech
companies. But its mission has expanded over the
past decade. Code2040 now aims for members of
these communities to contribute to the “innovation
economy” in all roles at all levels, proportional to
their demographic representation in the United
States. The ultimate vision: “equitable distribu-
tion of power in an economy shaped by the digital
revolution.”
Both Code2040’s current CEO, Mimi Fox Melton,
and her predecessor, Karla Monterroso, have argued
that coding training alone is not enough to guarantee
employment or equalize educational opportunities.
In an openly critical letter to the tech industry pub-
lished after the murder of George Floyd in 2020, they
noted that 20% of computer science graduates and
 57

24% of coding boot camp grads are Black or Latino,

compared with only 6% of tech industry workers. Fox
Melton and Monterroso observed: “High-wage work
in America is not colorblind; it’s not a meritocracy;
it’s white. And that goes doubly for tech.”
These recent coding education efforts ask import-
ant questions: Code for what? Code for whom?
Meanwhile, several other recent initiatives are
focused on the injustices both caused and reflected
by more recent aspects of the digital economy, par-
ticularly artificial intelligence. They aim to challenge
the power of technological systems, rather than
funneling more people into the broken systems
that already exist. Two of these organizations are
the Algorithmic Justice League (AJL) and the Ida
Below: Black Girls
B. Wells Just Data Lab. Above: Ruha Benjamin
Code works to Joy Buolamwini, a computer scientist, founded directs the Ida B. Wells
increase the number the Algorithmic Justice League after discovering as a Just Data Lab, which
of women of color grad student at MIT that a facial-analysis system she aims to rethink and
working in technology retool the relationship
was using in her work didn’t “see” her dark-skinned
by introducing girls to between power and
computer science.
face. (She had to don a white mask for the software technology.
to recognize her features.)
Now, the AJL’s mission is “leading a cultural
movement towards equitable and accountable AI,” such efforts back in 1980, writing in Mindstorms that
and its tagline reads: “Technology should serve all “a particular subculture, one dominated by computer
of us. Not just the privileged few.” The AJL pub- engineers, is influencing the world of education to
lishes research about the harms caused by AI, as favor those school students who are most like that
well as tracking relevant legislation, journalistic subculture.”)
coverage, and personal stories, all with the goal of Learning to code won’t solve inequality or pov-
moving toward more equitable and accountable AI. erty or remedy the unjust structures and systems
Buolamwini has testified to Congress and in state that shape contemporary American life. A broader
hearings on these issues. vision for computer science can be found in the
The Ida B. Wells Just Data Lab, founded and model proposed by Learning for Justice, a project
directed by Ruha Benjamin, a Princeton professor of of the Southern Poverty Law Center that works to
African American studies, is devoted to rethinking provide educational resources and engage local
and retooling “the relationship between stories and communities, with the ultimate goals of address-
statistics, power and technology, data and justice.” ing injustice and teaching students and the com-
Its website prominently features a quote from the munities they come from to wield power together.
journalist and activist Ida B. Wells, who system- The project’s digital literacy framework highlights
atically collected data and reported on white mob important focus areas far beyond a narrow empha-
violence against Black men during the 1890s. Her sis on learning to code, including privacy concerns,
BEBETO MATTHEWS/AP IMAGES; CYNDI SHATTUCK

message: “The way to right wrongs is to turn the uncivil online behavior, fake news, internet scams,
light of truth upon them.” One of the lab’s efforts, ideological echo chambers, the rise of the alt-right,
the Pandemic Portal, used data to highlight racial and online radicalization.
inequality in the context of covid-19, focusing on These new frameworks of digital literacy, tech
10 different areas: arts, mutual aid, mental health, diversity, and algorithmic justice go beyond coding
testing and treatments, education, prisons, policing, to prepare individuals to meaningfully question, eval-
work, housing, and health care. It provided data- uate, and engage with today’s array of digital spaces
based resources and tools and offered evidence that and places. And they prepare all of us to imagine and
these seemingly disparate categories are, in fact, articulate how those spaces and places can better
deeply interwoven. serve us and our communities.
Technological solutionism may persist in Silicon
Valley campuses and state house corridors, but individ- Joy Lisi Rankin is a research associate
uals, organizations, and communities are increasingly professor in the Department of Technology,
Culture, and Society at New York University
recognizing that coding instruction alone won’t save and author of A People’s History of Computing
them. (Even Seymour Papert expressed skepticism of in the United States.
58

Teaching
the
biliterate
brain
to read
 L
inus Merryman spends
about an hour a day on
his laptop at his elemen-
tary school in Nashville,
Tennessee, mostly work-
ing on foundational read-
ing skills like phonics and
What’s best spelling. He opens the reading app Lexia
for kids with ease, clicking straight through to
lessons chosen specifically to address his
accustomed reading needs. This week Linus, who’s in
to toggling second grade, is working on “chunking,”
from finding the places where words are bro-
ken into syllables. The word chimpanzee
book to screen appears on the screen in large letters, and
Linus uses his mouse pad to grab cartoon
Roman columns and slip them into the
and back again?  spaces between letters, like little divid-
ers, where he thinks the syllable breaks
should be. The app reads his guesses back
to him—“chim-pan-zee.” He gets it right. 
After practicing these foundational skills
on the computer, he and his classmates
close their laptops and head to the rug,
each with a print copy of their class reader,
I Have a Dream, a picture book featuring
the text of Martin Luther King Jr.’s speech.
Students follow along in their books as the
teacher reads aloud, occasionally stopping
By Holly Korbey so they can ask questions and point out
things they notice, like how the speech is
written in the first person. 
Linus’s mom, Erin Merryman, an early
reading interventionist at another Nashville
school, initially worried about how well her
son would learn to read in a classroom that
made so much use of computers. He has
been diagnosed with the learning disabil-
ity dyslexia, and Merryman knows from
her training that dyslexic students often
need sensory input to learn how sounds
are connected to letters. Close oversight
from a teacher helps them as well. But since
his reading has vastly improved this year,
she’s adjusted her view. 
60

“I think a lot of what the app is doing is very and then stepping outside them to evaluate all
good, very thorough,” Merryman says. “I’m sur- the evidence and shake out the findings. 
prised by how effective it is.” “Knowledge has not progressed to the point
Like Merryman, a growing group of experts where we have the kind of evidence I feel we
and educators are trying to figure out what the need,” Wolf says. “What do the affordances of
relationship should be between digital tech- each medium—screens vs. print—do to the
nology and reading instruction. Both reading reading brain’s use of its full circuitry? The
and digital tech are world-expanding human answers are not all in.” 
inventions, and laptops and smartphones have But, she continues, “our understanding is that
arguably given humans unending opportunities
to read more; you can access pretty much any-
print advantages slower, deeper processes in
the reading brain. You can use a screen to com- Research-
thing in print within a few seconds. In terms of
“raw words,” the cognitive scientist Daniel T.
plement, to teach certain skills, but you don’t
want a child to learn to read through a screen.”  ers who
Willingham has said, kids read more now than
they did a decade ago. But many reading experts Which is best for comprehension, study young
suspect that the technology may also be changing
how they read—that reading on a screen is fun-
screens or books?
Once children have learned to decode words, readers’
damentally different from reading on the page. 
Researchers who study young readers’ brains
research on how they comprehend texts encoun-
tered on screens and paper gets a little more brains and
and behaviors are eager to understand exactly
where tech serves kids’ progress in reading and
decisive. Experts say that young readers need
to be reading alongside adults—getting feed- behaviors
where it may stand in the way. The questions are
still so new that the answers are often unclear.
back, asking questions, and looking at pictures
together. All this helps them build the vocabulary are eager to
Since the covid-19 pandemic closed schools in and knowledge to understand what they’re read-
2020, nearly all students have been organizing ing. Screens often do a poor job of replicating
their learning around a school-issued laptop or this human-to-human interaction, and scien-
tablet. But educators who are more dependent tists like Wolf say that the “reading circuits” in
than ever on digital tech to aid learning in gen- children’s brains develop differently when the
eral often have little or no guidance on how to young learners are glued to a screen. 
balance screens and paper books for beginning Studies on the inner workings of the brain
readers accustomed to toggling between the confirm the idea that human interaction helps
two. In a lot of ways, each teacher is winging it.  develop beginning readers’ capacity for under-
Figuring out how best to serve these young standing. But they suggest that reading paper
“biliterate brains” is crucial, cognitive scientists books is associated with that progress, too. In
say—not just to the future of reading instruction, one study, researchers found that three- and
but to the future of thought itself. Digital tech- four-year-old children had more activation in
nology has transformed how we get knowledge language regions of the brain when they read
in ways that will advance and forever alter our a book with an adult like a parent than when
species. But at the individual level, the same they listened to an audiobook or read from a
technology threatens to disrupt, even diminish, digital app. When they read on an iPad, acti-
the kind of slow, careful learning acquired from vation was lowest of all. In another study, MRI
reading books and other forms of print.  scans of eight- to 12-year-olds showed stronger
Those seemingly contradictory truths under- reading circuits in those who spent more time
line the question of how we should go about reading paper books than those who spent their
PREVIOUS SPREAD: GETTY IMAGES

teaching children to read in the 21st century, says time on screens.

neuroscientist Maryanne Wolf, author of Reader, For older students, significant research shows
Come Home: The Reading Brain in a Digital that comprehension suffers when they read
World. Wolf, the first to use the term “biliterate from a screen. A large 2019 meta-analysis of 33
brain,” is busy researching the relative merits of different studies showed that students under-
screen- and page-based approaches, adopting in stood more informational text when they read on
the meantime a stance of what she calls “learned paper. A study by the Reboot Foundation, evalu-
ignorance”: deeply investigating both positions ating thousands of students across 90 countries
 61

including the US, found that fourth graders who 30 countries showed that students who pre-
used tablets in nearly all their classes scored 14 ferred reading on paper scored 49 points higher,
points lower on a reading test than students who on average, on the Program for International
never used them. Researchers called the score Student Assessment (PISA)—and the study
gap “equivalent to a full grade level” of learning. hinted at an association between reading paper
Students who used technology “every day for books and liking to read.
several hours during the school day” underper- Baron also thinks there should be more prac-
formed the most, while the gap shrank or even tical attention paid to developing pedagogical
disappeared when students spent less than half approaches that explicitly teach the slower,
an hour a day on a laptop or tablet.    more focused habits of print reading, and then
Why do students understand more of what help students transfer those skills to the screen.
they read when it’s in a book? Researchers aren’t Reinforcing those habits would be helpful even
entirely sure. Part of the issue is distraction, for people who usually read books, because
says Julie Coiro, a researcher at the University someone reading a book can get distracted
of Rhode Island. Kid-friendly reading apps like too—especially if a phone is nearby. 
Epic! offer thousands of books that often con- The use of digital books and textbooks
tain images, links, and videos within the body exploded during the pandemic, and it may be only
of the text. These are meant to enhance the a matter of time before all educational publishing
reading experience, but they often drag chil- moves online. So it’s all the more important to
dren away from concentrating on the meaning keep making digital reading better for students,
of the text. Even in reading experiments where says literacy educator Tim Shanahan. Instead of

understand students weren’t allowed to browse the web

or click on embedded links, though, they still
trying to make the digital technology more like a
book, Shanahan has written, “[engineers] need to

exactly performed worse. 

Virginia Clinton-Lisell, the author of the 2019
think about how to produce better digital tools.
Tech environments can alter reading behavior,

where tech meta-analysis, hypothesized that overconfidence

could be another aspect of the problem. In many
so technological scaffolding could be used to
slow us down or to move around a text more

serves kids’ of the studies, students who read from a laptop

seemed to overestimate their comprehension
productively.” In the future, students might read
about history or science from something like a

progress in skills compared with those reading the paper

books, perhaps causing them to put in less effort
“tap essay,” where words, sentences, and images
are revealed only when a reader is ready and taps

reading while reading.  

Students self-report learning more and hav-
the screen to move on to the next piece of text.
Or maybe their reading material will look more

and where it ing a better reading experience when they read

paper books. Linguist Naomi Baron, author of
like a New York Times digital article, in which
text, images, video, and sound clips are spaced

may stand in How We Read Now: Strategic Choices for Print,

Screen, and Audio, says that when she interviews
out and blended together in different ways.

the way. students about their perceptions, they often say

reading from a book is “real reading.” They like
Hooked on computer phonics 
About two-thirds of American schoolchildren
the feel of the book in their hands, and they find can’t read at grade level. At least partly to blame
it easier to go back to things they’ve already read is a widespread method of reading instruction
than when they are reading from a screen. While that dominated classrooms for 40 years but was
they might prefer digital formats for reasons of not based on scientific evidence about how the
convenience or cost, they sense they have greater brain learns to read: “balanced literacy,” and
concentration while reading print.  its close cousin “whole language,” deempha-
But Baron says school districts and educators sized explicit instruction in reading’s founda-
often aren’t aware of the strong research connect- tional skills, leaving many children struggling.
ing books to better comprehension or confirming But over the last several years, a new method
student preferences for print. Baron’s research strongly focused on these foundational skills,
dealt with college students, but last year a study often referred to as the “science of reading,” has
by the Organization for Economic Cooperation brought sweeping changes to the US education
and Development (OECD) of 15-year-olds in system. Based on decades of scientific evidence,
62

the “science of reading” approach is organized In the Long Island schools Borkowsky works
into five areas: phonemic awareness (learning with, it’s more likely that students do more
all the sounds of the English language), phonics reading work on laptops because schools pur-
(learning how those sounds are attached to let- chased expensive technology and feel pressured
ters), vocabulary, comprehension, and fluency. to use it—even if it’s not always the best way to
Learn-to-read apps and digital platforms have teach reading skills. “What I’ve seen in schools
the potential to teach some of these foundational is they have a program, and they say, ‘Well, we
skills efficiently. They’re especially well suited bought it—now we have to use it.’ Districts find
to phonemic awareness and phonics, making it hard to turn back after purchasing expensive
learning letters and sound combinations a game
and reinforcing the skills with practice. Lexia,
programs and materials,” she says.
Some platforms are working to bridge the Digital
arguably the most widespread digital platform
devoted to the science of reading, teaches basic
gap between online and in-person instruction.
Ignite! Reading, an intensive tutoring program platforms
and complex foundational reading skills, like
letter-sound blends and spelling rules, using
launched after the pandemic closed schools,
teaches foundational reading skills like phone- can
responsive technology. When learning a specific
skill, such as figuring out how to read words
mic awareness and phonics through a videocon-
ferencing platform, where reading tutors and reinforce
like meal and seam with the “ea” vowel combi-
nation in the middle, students can’t move on
students can see and hear one another. 
Ignite’s instruction attempts to blend the certain
until they’ve mastered it. 
A new wave of predictive reading platforms
benefits of digital tech and human interaction.
In one tutoring session, a first grader named specific
goes one step further. Companies like Microsoft
and SoapBoxLabs are envisioning a world where
Brittany in Indianapolis, Indiana, sounded out
simple words, prompted by her reading tutor, reading
students can learn to read entirely via computer.
Using AI speech recognition technology, the
whom she could see through her laptop’s cam-
era. Brittany read “map” and “cup,” tapping the skills, but it’s
companies claim, these digital platforms can whiteboard in her hand each time she made a
listen closely to a student reading. Then they can sound: three sounds in a word, three taps. At the
identify trouble spots and offer help accordingly.  same time, a digital whiteboard on her laptop
As digital tech for learning to read spreads screen also tapped out the sounds: one, two,
into schools—Lexia alone serves more than three. As Brittany sounded out each word, the
3,000 school districts—some reading experts tutor watched the child’s mouth through the
are wary. Research on its efficacy is limited. computer’s camera, giving adjustments along
While some see technology playing a useful the way. 
role in reading-related functions like assessing Ignite cofounder and CEO Jessica Sliwerski
students and even training teachers, many say says she’s building an army of remote reading
that when it comes to actually doing the teach- tutors to assist teachers in helping kids catch up
ing, humans are superior.  after the pandemic years. Students get 15-minute
Digital platforms can reinforce certain spe- sessions during the school day, and when ses-
cific reading skills, explains Heidi Beverine- sions are over, tutors get coaching on how to
Curry, chief academic officer of the teacher make the short bursts more effective. 
training and research organization The Reading Sliwerski believes technology can be incred-
League, but it’s the teacher who is constantly ibly useful for giving more students one-on-one
monitoring the student’s progress and adjusting attention. “We are taking a different approach
the instruction as needed.  to the technology,” she says. “We are centering
Faith Borkowsky, founder of High Five the child on a human who is highly trained and
Literacy, a tutoring and consultancy service in accountable. That’s the core of it, and there’s not
Plainview, New York, is not bothered by read- really anything tech about that.”
ing instruction apps per se. “If it happens to be
a computer program where a few kids could go Preserving deep reading 
on and practice a certain skill, I’d be all for it, if Once students can decode words and compre-
it aligns with what we are doing,” she says. But hend their meaning, the real work of reading
often that’s not how it plays out in classrooms. begins. This is what Wolf calls “deep reading,”
 63

a specific set of cognitive and affective pro- It’s possible, of course, that kids’ attention
cesses in which readers are able to take in spans haven’t actually changed that much with
whole chunks of text at a time, make predic- the advent of digital technology. Instead, argues
tions about what comes next, and develop Willingham, the cognitive scientist, in his book
lightning-fast perception. These interactive The Reading Mind: A Cognitive Approach to How
processes feed each other in the brain, accel- the Mind Reads, it’s their expectations for enter-
erating understanding.  tainment that have changed. “The consequence
But since the vast majority of the reading of long-term experience with digital technolo-
that today’s young people do—let’s face it, gies is not an inability to sustain attention. It’s
the majority that we all do—is skimming an impatience with boredom,” he writes. “It’s an
online article, a Facebook post, or a text from a expectation that I should always have something
friend while hopping from one tab to another, interesting to listen to, watch, or read, and that
deep reading as a cognitive process is at risk. If creating an interesting experience should require
today’s kids read only from screens, Wolf says, little effort.” Deep reading, on the other hand,
they may never learn deep reading in the first requires “cognitive patience”—an entirely dif-
place—that elaboration of the brain’s reading ferent set of skills in which kids often have to
circuit may never be built. Screen reading may put in great effort for a payoff that is sometimes
“disrupt and diminish the very powers that it many pages down the road.
is supposed to advance.”  Yet in Wolf’s view, getting rid of all read-
“We are amassing data that indicates there ing tech would be as ill-advised as relying on
are changes in the reading brain that diminish it exclusively. Instead, she’s hoping to spur
its ability to use its most important, sophisticated a conversation about balance, gathering evi-
processes over time when the screen dominates,” dence about which ways of using digital tech-

the teacher Wolf says. Deep reading is something that came

naturally to many readers before digital tech and
nology work best for diverse learners and for
different age groups—information that could

who is personal computers, when they had lots of time

to spend doing nothing but reading a book; but
help districts and teachers guide the decisions
they make about teaching reading. A five- to

constantly it can’t be assumed that today’s young readers,

with their biliterate brains, will automatically
10-year-old child who is learning to read has
different needs from a 12-year-old, or from

monitoring learn the process. 

Some educators are paying more attention to
a high schooler whose smartphone is loaded
with five social media apps. Young children just

the student’s how to help students begin to learn deep read-

ing. Doug Lemov, a charter school founder who
beginning to build their reading circuit benefit
most from books and human interaction. Older

progress now trains teachers full time with his “Teach

Like a Champion” books and courses, is acutely
kids can cultivate the “digital wisdom” to make
smarter choices while working on developing

and concerned that many middle and high school

students no longer appear to have the attention
the ability to toggle effortlessly between print
and digital worlds. 

adjusting span to concentrate on a text for long periods of

time. So he encourages the teachers he trains to
Some kids, though, may be tired of all that tog-
gling. Matt Ryan, a high school English teacher

the adopt “low-tech, high-text environments” inside

their classrooms, with paper books, pencils,
in Attleboro, Massachusetts, doesn’t allow any
e-books in his class—when he assigns a novel,

instruction and paper. In such a setting, students slowly

build up their attention spans by doing nothing
it’s paper only. Not only does he not get any
pushback, he says, but he senses students are

as needed. but reading a book or scratching out a piece of

writing, even if that means beginning with just
somewhat relieved. 
“Distractions are a very real issue, so read-
a few minutes at a time. ing on a device will not be effective for most of
“Build on that until they can go for 20 min- them,” Ryan says. “My sense is that so much
utes, either in a group or individually—just of what they do is on a device—they welcome
reading the text, sustaining their attention and something off of it.”
maintaining focus,” Lemov says. “Writing does
Holly Korbey is an education and parenting
the same thing: it improves the focus and atten- journalist and author of Building Better
tion that students will need to do deep reading.”  Citizens: A New Civics Education for All.
64

Living with
We can build places that are easy to escape from, or places that are easy to defend. By Susie Cagle
GUTTER CREDIT HERE
 65

In 2007, the Witch Fire burned 1,125

homes in San Diego County. But when the
firestorm hit the edge of a development
named The Crosby at Rancho Santa Fe,
not a single house ignited.

T
he first sparks that ignited
in the Montecito hills
above Santa Barbara,
California, on November
13, 2008, were stoked
by ferocious sundowner
winds gusting at up to
85 miles per hour, pushing the flames
down into the densely populated canyon.
Troy Harris, then the director of institu-
tional resilience at Westmont College in
Montecito, rushed from the other side
of town to the campus, nestled in foot-
hills dense with chaparral and eucalyptus.
Within minutes of entering the canyon,
the Tea Fire had already reached the
school. But the students did not evac-
uate. Westmont, with a legacy of large
canyon wildfires over decades and only
two winding roads as routes of escape,
had planned for just this kind of disaster.
They stayed put.
“We had parents calling the sheriff’s
office and the sheriff’s office was telling
people—incorrectly—tell your kid to get
out of there,” says Harris. In fact, there
would have been no way to move 1,000
people down the hill faster than the fire
was moving in on them. Instead, students
and staff gathered in the fire-resistant gym
on the southwestern corner of campus. 
Nine structures on the campus burned,
but the sheltered students were unharmed.
It was, says Harris, “a spectacular win,” but
a highly unusual one.
With each devastating wildfire in the
US West, officials consider new emergency
management methods or regulations that
might save homes or lives the next time.
In the parts of California where the fire-
ready hillsides meet human development
and where the state has suffered recurring
seasonal fire tragedies, that search for
new means of survival has especially high

wildfire
stakes. Many of these methods are low cost
and low tech, but no less truly innovative.
With climate change bringing more com-
munities under wildfire threat across the
GUTTER CREDIT HERE

world, adaptation may require more social

change than materials engineering.
“When people think of wildfire, they
think of getting away as quickly as possible,
66

Nine structures on the Westmont College

campus burned as the 2008 Tea Fire
swept through Montecito, California. But
students who sheltered in fire-resistant
structures were unharmed.

right? Like that’s the messaging that every- means planning for a worst-case scenario tech or expensive, but it is counterintu-
one hears—evacuate, evacuate, evacuate,” but not a truly rare one: a fire that moves itive to how we have long thought about
says Jason Tavarez, Harris’s successor at faster than one can flee. That is the kind wildfire. In the 1970s, when Jack Cohen
Westmont. “And that’s 99 times out of 100.” of fire California has seen time and again.  pioneered the concept of “defensible
But the other scenario is this: a con- In response to the increasing threat, space,” a zone cleared of flammable veg-
flagration too fast and violent to escape, some institutions and communities are etation or other fuel around a structure,
with no better option than to hunker. It is taking a cue from Australia, where offi- the US Forest Service largely ignored
a “shelter in place” or “stay and defend” cials have employed a policy of “leave early him. It was a paradigm-shifting innova-
approach to wildfire. Evacuations from or stay and defend” since the 1990s. But tion—an easily implemented retrofit, at
western US wildfires have routinely caused even Australia has had second thoughts least wherever the space was available—
significant casualties themselves, with since the 2009 Black Saturday fires, when but it meant considering wildfire from a
fleeing people trapped on narrow roads more than half of the 173 people killed defensive position instead of the offen-
behind debris or in traffic jams. For that
reason, coupled with the more destructive
pace of recent fires, there is a new spotlight
on the shelter-in-place strategy. Despite
some notable successes, however, it is The basic science of preventing a
not very popular. building from burning is not especially
“In the US it’s something people are
struggling to wrap their heads around,” says
high tech or expensive, but it is counter-
fire researcher Crystal Kolden, a professor intuitive to how we have long thought
at the University of California, Merced.
“When is it okay to shelter in place? And
about wildfire.
more importantly, what is the minimum
need in terms of the facility, and how
do you do that risk-benefit trade-off in a
moment of crisis?” had been sheltering inside a home. And sive one the Forest Service had adopted
In order to effectively live with fire, we for the most part, the US has been slow for nearly 100 years.
can build places that are easy to escape to adopt shelter-in-place policies for wild- Today regulators have come around,
from or places that are easy to defend. fire. The optics are not good—even the and California building standards for wild-
These are by no means mutually exclu- best-laid plans can look like abandonment land areas at high and very high fire risk
sive, but the US West hasn’t done either. or imprisonment, like leaving people to now require 100 feet of open space around
Meanwhile, the population has grown into nature’s violent whims. Fire researchers structures, at least where there is 100 feet
the spaces on the rural edges of cities and and officials can’t agree on the science available to clear. Other home-hardening
suburbs, in the foothills and canyons and that should guide the planning. And with measures are comparably small scale, even
drainages where fire lives—what’s called little adoption, there is little data on how cheap: replacing flammable roofs, closing
the wildland-urban interface. While fires well the approach works. Experts point window seams and junctions, using fine
have grown in size and destructiveness repeatedly to the same handful of success wire mesh to cover vents where sparks
PREVIOUS SPREAD: DON BARTLETTI/LOS ANGELES TIMES/GETTY IMAGES

over the past two decades, so has the pop- stories like the one at Westmont College. might enter. And the latest fire-resistant
ulation in these hazard areas—roughly “We have to get over this idea that it’s materials won’t save a house where the
doubling between 1990 and 2010, with the always the best thing to actually evacuate,” gutters have been allowed to fill with dry
more dangerous areas growing the most. says Kolden. “We used to have community kindling. Form tends to follow function: flat
In fact, the wildland-urban interface is the bomb shelters, right? These are function- roofs, steel windows, clean lines that leave
fastest-growing land-use area in the US.  ally community fire shelters. Those are no harbor for a stray ember. Each devastat-
Sheltering is not passive but active, the sort of conversations that we haven’t ing fire is bound to encourage a new inno-
whether it involves advance preparation had. And if we really want to build fire- vation as fresh weaknesses are revealed.
in open-air safety sites and enclosed build- resilient communities, we have to have California’s strictest fire code applies
ings or, in some cases, fire defense as the those going forward.” only to homes in a clearly designated
flames move in. In rural areas with few high-risk area (where, according to the
routes in or out, a shelter-in-place plan Our shelters California Department of Forestry and
can mean the difference between life and The basic science of preventing a build- Fire Prevention, roughly one in four resi-
death in the face of a fast-moving fire. It ing from burning is not especially high dential structures lies)—and only to those
 that are newly built. In Paradise, where a nearly a century ago is essentially off the “The conversation turns to not whether
fire in 2018 killed at least 85 people and table—it is work that isn’t required under we’ll develop these areas, but how shel-
destroyed more than 18,000 structures, state codes, and no clear funding source ters are becoming part of it,” says Cova. In
nearly 40% of homes built after 1996 sur- is available. Even where communities are California, “they’re trying to chart a course
vived, versus just 11% of those built before. wiped out by fire, existing roads don’t fall where development in these areas can
The incremental addition of more and under the purview of minimum fire reg- continue. You end up with public-safety
denser housing in flammable dead-end ulations when it comes time to rebuild. and affordable-housing goals conflicting.” 
canyons is a concern, says Thomas Cova, But entirely new housing tracts are held
an evacuation researcher and professor of to much higher standards. Stay and defend
geography at the University of Utah. The “I’ve always thought of shelter-in-place Even among shelter-in-place advocates,
space between houses, or lack thereof, is as a backup plan in emergencies, and it there is broad agreement that it is always
a significant predictor of whether or not would be really wise to consider what better to evacuate if there is the time and
they’ll burn. Building suburban infill is in options you might have,” says Cova. “But ability to do so safely. The problem is with
many ways good housing policy for a state now, I think it’s also entering into the discus- wildfires that move so fast there’s no time
suffering from a severe lack of affordable sion associated with [new] development.” to get out. A secondary, no-evacuation plan
homes, but it is bad land-use policy for a That’s especially true in light of could mean the difference between guar-
state with recurring intense wildfires. Still, California’s acute housing affordability anteed death and a chance of survival. It
there’s little clear incentive for local offi- crisis, which has put the state under severe may be as counterintuitive a cultural inno-
cials to prevent the construction of new pressure not only to continue building new vation as defensible space, forcing us to
homes, even ones that will increase the risk homes but to build them on cheaper, more look at wildfire as an even greater threat.
DAVID MCNEW/GETTY IMAGES

for the entire community. One more flam- rural, more fire-risky land. A new guidance “We don’t have formal methods for des-
mable structure on the hillside, one or two issued in October 2022 by the California ignating safety zones for the public. But
more cars on the road—but also revenue state attorney general explicitly calls for the concept has been used,” says Cova. In
collected from one more property tax bill. local agencies to “avoid overreliance on past blazes, firefighters have, for example,
Extensive retrofitting of the built envi- community evacuation plans” and con- moved people to golf courses and turned
ronment in towns and cities established sider shelter-in-place options.  on the sprinklers.
68

One of the first shelter-in-place suc- school has had a shelter-in-place plan for ignition; fire hydrants were spaced every
cesses in the US was a result of quick decades, but some officials were none- 250 feet along roads in and around the
thinking rather than advance planning. theless critical. “This shelter-in-place community; a defensible zone and other
In 2003, with the Cedar Fire whipping policy is going to have to be reassessed,” open spaces such as golf courses and parks
across San Diego, fire officials chose to state senator Henry Stern told a crowd were maintained to buffer the neighbor-
lock down the Barona Resort and Casino at a community meeting shortly after hoods from the chaparral and eucalyptus
instead of attempting to evacuate the the fire. Even when it works as intended, hillsides expected to burn; and homeown-
hundreds of people inside. The fire chief choosing to stay while a fire rages is not ers’ associations were set up to enforce
parked his truck across the sole exit, “so popular public policy. and maintain fire protection measures. 
that if anybody got the idea of leaving, they “It is just a bad plan for people to leave Each home was considered to be
weren’t going anywhere,” says Cova. “The Pepperdine when they already are in the built to shelter-in-place standards, with
fire burned around the casino’s parking safest location you can be for survival,” ignition-resistant construction and mate-
lots on all sides, all the hills around it. And
the people just stayed there and gambled.”
Westmont College began its shelter-
in-place planning that same year, at the
urging of the local fire department. In
2009, just six months after surviving the
Tea Fire, Westmont was threatened by the “Shelter-in-place was really a theory
Jesusita Fire. This one was a little further and it’s still a work in progress.”
away, and slower moving—so there was
time to leave. That’s when Harris real-
ized “we had a stay plan, but we had yet
to develop a go plan.” In evacuating from
Jesusita, “it was clear it was a multi-hour
thing. There’s just no real fast way to get
1,000 people off the hill.”  says Drew Smith, LA County’s assistant rials—a cutting-edge approach for the
Tavarez is quick to point out that the fire chief. time, though the standards have since
Westmont students are not held against LA County fire officials reevaluate the been adopted into state and local codes.
their will. But most everyone at the school plan annually and haven’t found it want- They are little fortresses of tile roofs,
at this point has bought into sheltering ing. But Smith is skeptical of expanding stucco walls, hardscape patios, and cov-
in place. And if anyone hasn’t, he says, the concept to smaller institutions or com- ered eaves. Early evacuation is still always
“we explain very kindly but firmly that munity buildings—there is not enough the primary emergency plan, and the roads
with the number of students that we space in those structures for enough peo- are designed to facilitate it. But the heavy
have here, and the plans that we have in ple to weather the extreme heat and smoke fortification gives the communities—both
place, and the contingency that we built of a wildfire, he says. His measure is 50 the structures and the people who shel-
into how we do things on campus, this people for 15 acres, or about four peo- ter in them—an extra chance to survive.
is actually a lot safer than trying to fight ple in the space of a football field. Some “One of the core principles is that it’s
the fire down the hill.” state fire planners, though, use standard community wide,” says Brandon Closs,
Nonetheless, college populations are occupancy measurements to determine fire prevention specialist for the Rancho
easier to keep contained than other com- shelter-in-place capacity, resulting in a Santa Fe Fire Protection District. San
munities, and Westmont isn’t the only standard closer to a few square feet per Diego’s building code has long been at
example. In 2018, ahead of the massive person. The scarcity of data means there’s the vanguard of fire safety—it was used
and fast-moving Woolsey Fire that burned no consensus. as a model for the state regulations, and
through the Santa Monica mountains, it is still more stringent than the state
officials evacuated a quarter-million peo- Fire-planned communities requires. 
ple from their homes while Pepperdine Individual homes can also serve as shel- “Shelter-in-place really was a theory,
University in Malibu sheltered hundreds ters given the right conditions. In 2004, and it’s still a work in progress,” says
of its students on campus. They were five communities in Rancho Santa Fe, Closs. He and others are confident in
protected by wide defensible spaces, an affluent, semi-rural part of San Diego Rancho Santa Fe’s design, but the com-
expansive irrigated lawns, and hardened County, were designed with this in mind. munities haven’t yet been thoroughly
buildings equipped with sprinklers. The Thousands of homes were built to resist tested by a blaze. 
 69

And nearly two decades after Rancho services from their own or contracted fire with homeowners responsible for their
Santa Fe was built, it is still an outlier in crews—or at least they used to be. Even own go plans and for the full cost of any
the state. in perhaps the best-designed fire-ready hardening measures or landscape man-
Cost alone is one likely hurdle. The wildland community in California, insur- agement. This also makes wildfire a deeply
nonprofit Insurance Institute for Business ance companies are canceling policies to unequal problem: some high-risk areas
and Home Safety estimates it costs 4% to reduce their risk load. “The dollar is going are filled with multimillion-dollar homes
13% more to build a home to the highest to move a lot of things quicker than regu- surrounded by plenty of open space,
level of fire safety, far exceeding current lations can,” says Closs. whose owners have the means to keep
state standards. But achieving the level them updated with the latest construction
of community hardening now in place A cultural shift innovations, while others are packed in
in the wealthy, gated neighborhoods of It is infinitely easier to upgrade one’s own on small lots overgrown with trees that
Rancho Santa Fe requires a much larger roof or vent mesh than it is to implement residents can’t afford to cut down. Every
investment. community-scale hardening measures. step toward putting the burden of safety
Homes in these developments are The factors making California’s wildfires at the community level relieves some of
priced in the low millions at the least: a more acutely destructive to people and that inequality.
2,400-square-foot three-bedroom house their homes are more socioeconomic than “Civilization has always progressed
in Rancho Santa Fe sold for $3.2 million in they are climate driven. based on community cooperation,” says
2022. Lower-priced homes and communi- “We’re not accustomed to thinking Kolden. “And we need to do this for fire
ties in equally fire-risky parts of San Diego about what shelter-in-place looks like, to have any chance of averting a lot of the
County, and across California, have none of because the term is most commonly asso- disasters that we’ve seen the last few years
this protection. Many homeowners in the ciated with people’s individual houses,” as we move forward.”
area are also covered by insurance policies says Kolden. Preparing for fire is in many Susie Cagle covers climate change
that offer private mitigation or firefighting ways treated as an individual problem, and inequality in California.

ADVERTISEMENT

Discover what’s coming

next in technology.
SUBSCRIBE NOW for access to:
• In depth reporting on AI, climate change,
biotech & more
• Trusted insights you can’t find anywhere else
• Science & technology news shaping the future
• 6 print and digital issues a year
• Discounts on MIT Technology Review events

Scan this code to subscribe or

learn more at technologyreview.com/subscribe
70

There is no shortage of popular

books and lectures on math—
but they can take us only so far.
By Pradeep Niroula
Rocky
journeys to
Illustrations by Jenny Kroik

the land
of mathematics
How to Bake Pi:

M
athematics has long been the vocabulary of research and analogies
Easy Recipes for
Understanding Complex Maths
presented as a sanctuary from physics to explain myself. But despite
Eugenia Cheng from confusion and doubt, having been close to math for most of my
a place to go in search of life, I continue to be bewildered by math-
answers. Perhaps part of ematics research. What motivates it, and
the mystique comes from the fact that what is its ultimate endgame? What does
The Joy of Abstraction:
biographies of mathematicians often paint the world look like to someone steeped
An Exploration of Math,
Category Theory, and Life
them as otherworldly savants—people in the culture of mathematics? So when I
Eugenia Cheng who seem to pull nature’s deepest truths discovered that Terence Tao, a living leg-
from thin air and transcribe them in prose end of contemporary math, was offering
so succinct and self-assured it must be an online class on his approach to “math-
read meditatively, one word at a time. ematical thinking,” I had to check it out.
The Art of Logic
As a graduate student in physics, I have The movie-length course, distributed by
in an Illogical World
Eugenia Cheng seen the work that goes into conducting MasterClass, starts out invitingly enough.
delicate experiments, but the daily grind Tao exudes calm and confidence. A math-
of mathematical discovery is a ritual alto- ematical mindset, he says, makes “the
gether foreign to me. And this feeling is complex world a bit more manageable.”
only reinforced by popular books on math, He suggests that his class might be “even
A Divine Language:
which often take the tone of a pastor dis- more suitable for those without formal
Learning Algebra, Geometry,
and Calculus at the Edge pensing sermons to the faithful. math training.” But very soon, the futility
of Old Age In physics, the questions we ask and the of this attempt to pierce the mystique of
Alec Wilkinson theories we come up with aim to explain mathematics becomes inescapable.
the underlying reality better. Indeed, cer- For most of the session, Tao is seated
tain concepts—like the fact that oppo- in a white armchair; there are no black-
site charges attract or that disorder or boards, no pens, no paper. “Mathematics
entropy tends to increase—are so uni- is a language of precise communication,”
versally ingrained in our experience that Tao says, and yet here, he is without the
they creep into everyday language as met- most powerful tools for achieving that.
aphors. I often catch myself resorting to Although he tries to be approachable,
72

talking about how he once did poorly in When Wilkinson is not hunched over abstraction, why are you in mathematics?
an exam and struggles to assemble win- textbooks, he is dazzled by the mysticism Perhaps you should be in finance, where
dow curtains, I felt no closer to the world surrounding math. The mathematicians all the numbers have dollar signs in front
of math. After 90 minutes of watching, he talks to speak of their profession with of them.” Math’s only promise in return
the pithy takeaways I was left with were quasi-religious sentiments and think of for unwavering fealty is that of a higher
indistinguishable from what I might learn themselves as mere prospectors of a tran- plan, much as in a cult. Wilkinson is left
at a mindfulness retreat: ‘‘Everything is scendental order. When Wilkinson com- as dazed and exhausted as a victim of a
united” and “Embrace failure.” plains to his niece that math is not yielding shipwreck stranded in the Arctic.
I am not the only person who has tried— to him, he is told, “For a moment, think My frustrations and Wilkinson’s high-
and failed—to break into the church of of it as a monastic discipline. You have to light the inadequacies of the mediums
math. Recently, Alec Wilkinson, a writer for take on faith what I tell you.” Where his usually employed in teaching mathematics.
the New Yorker and a longtime believer in niece and others see patterns and order, Textbooks aren’t always written with acces-
self-improvement, took on a yearlong proj- he perceives only “incoherence, obfusca- sibility in mind. They vacillate between
ect to conquer some of the basic mathemat- tion, and chaos”; he feels like a monk who pedantry and hand-wavy dismissals, and
ics that evaded him in his youth: algebra, sees lesser angels than everybody around the exercises they present can appear to
geometry, and calculus. In his 2022 book A him. He is now reproachful of his educa- be a series of pointless drills. At the same
Divine Language, he describes his journey tion and his younger self: Why hadn’t he time, attempts at an overview can feel
as a quest for redemption after those strug- learned all this better when he had the frustratingly empty. What Wilkinson and
gles with high school math. “It had abused impressionability of a child? I really needed was a sympathetic voice—
me, and I felt aggrieved,” he writes. “I was A year later, Wilkinson can solve some the testimony of one who has climbed the
returning, with a half century’s wisdom, to calculus problems, but the journey was dif- heights of abstract math but also has the
knock the smile off math’s face.” ficult, the terrain harsh and often unwel- patience to guide a newcomer.
Wilkinson has a better plan than mine: coming. Math often gets talked about as The mathematician and concert pianist
he starts with standard textbooks. And a language with logic as its grammar. But Eugenia Cheng is the closest I’ve come to
he has help. His niece, a math professor, when you learn a language like Spanish, finding such a voice.
agrees to hold his hand through this jour- you can casually pick up some words and I got into Cheng’s books because I share
ney. But even the first steps through alge- immediately unlock a new culture. The her love of baking. For the proud owner of a
bra are backbreaking. The skepticism of an introductory steps to formal math, on the stand mixer and several pastry brushes, the
adult gets in the way; he cannot seem to other hand, demand a commitment to title Cakes, Custard and Category Theory
accept the rules—the way variables can be rigor and abstraction while withholding sounded too delectable to pass up. Its
added and multiplied, how fractions and any usefulness. Among mathematicians, first chapter, about the epistemic nature
exponents work—as readily as children as Wilkinson discovers, there is even a of mathematics, starts with a recipe for
do. What’s more, he finds the textbook general derision toward those who seek brownies. Cheng tells you that stumbling
writing atrocious. useful application. There is G.H. Hardy’s into new ideas in math is like screwing up
“There is a boosterish quality to the famous jeer in 1940, “Is not the position of a soufflé recipe so badly that you end up
prose, as if learning math is not only fun! an ordinary applied mathematician in some with cookies. I was easily drawn in.
but also obscurely patriotic, the duty of an ways a little pathetic?” Or a more recent Many popular books on mathematics
adolescent citizen-in-waiting,” he writes. remark by John Baez: “If you do not like try to be approachable by talking about
“In addition to leaving things out, they
were careless about language, their sen-
tences were disorderly, their thinking was
frequently slipshod, and their tone was
often cheerfully and irrationally impatient.”
Revisiting algebra as an adult,
Though he wrestles algebra with decidedly Wilkinson declares, is “like meeting someone
determined rigor, six hours a day for six to you hadn’t seen in years and being reminded
seven days a week, and obsesses about it
the rest of the time, simple competence
why you never liked him or her.”
continues to elude him. Revisiting algebra
as an adult, he declares, is “like meeting
someone you hadn’t seen in years and being
reminded why you never liked him or her.”
 73

that extreme precision is a part of mathe-

matical research; elsewhere, we learn that
there is really no right way to make a cake
and that we should embrace flexibility in
ingredients as well as techniques. These
analogies can feel tenuous, sometimes
even forced. But thankfully, they quickly
fade away to make room for a casual con-
versation about mathematical topics.
Cheng’s latest book, The Joy of
Abstraction, builds on similar themes but
feels more like an undergraduate textbook.
Its chapters, with titles like “Isomorphism”
and “Functors,” provide a fairly rigorous
introduction to category theory and are
replete with theorems, proofs, and exercises.
Occasionally, Cheng goes on a tangent about
how certain concepts have etymological and
semantic parallels to real life—a “function”
can be thought of as a vending machine, a
“set” may represent a group of people (and
you can divide that set into “partitions” of
“friendships”). But having set up the stage
using familiar objects, she quickly gets to
the hard work of manipulating them using
logic. In short, her books are a humane
introduction to foundational math, and
they paint a good picture of what mathe-
maticians spend their time thinking about.
There is, however, one major way her
Cube of privilege math books markedly differ from undergrad-
Cheng illustrates the idea uate textbooks. A persistent theme running
of intersectionality with across Cheng’s writing is that the world is
representations like this one, showing
privilege along multiple dimensions.
best understood in a stripped-down form,
and that insights from abstract math can
even nourish empathy and a sense of justice.
Friends trying to be sympathetic to a heart-
break by prying into painful details should
stock markets or poker odds. Others wax is simply the ivory-tower equivalent of content themselves with simply knowing
poetic about prime numbers and the mys- building a Lego spaceship from tiny, inde- that a) there was something you loved and
tery of infinity. Cheng’s books lift readers structible pieces. More important, they are b) you recently lost it. It’s as simple as that;
to the rarefied heights of mathematical an invitation to change your worldview, all other details are superfluous.
abstraction by teaching them category to simplify thinking with abstractions, to Cheng’s more unorthodox contention—
theory, which she believes is the most interpret and analyze the world in math- one best presented in her book The Art of
foundational kind of math. ematical terms. Logic in an Illogical World—is that category
Category theory may seem esoteric, but Reading Cakes, Custard and Category theory can, in fact, be deployed in our daily
it is the underlying grammar of mathemat- Theory (also issued as How to Bake Pi), lives to make discussions around privilege,
ical logic. Cheng’s books pull back the cur- one soon discovers that the desserts are sexual harassment, racism, and even “fake
tain to show how pedestrian mathematics mere gambits. Each chapter begins with news” less divisive. For instance, she thinks
research can be; the act of chaining simple a recipe followed by an analogy between that the debate about social welfare can
inviolable axioms into complex arguments math and baking. Puff pastry is a reminder be described in terms of “false positives”
74

and “false negatives”: “a false negative in

this case is someone who deserves help
but doesn’t get it; a false positive would
be someone who doesn’t deserve help
but does get it.” The debate, her argument Cheng thinks the steely vocabulary
goes, isn’t about whether we should help of logic can help people caught in
people (of course we should!) but rather
about the extent to which we accommo- a heated argument realize that the divide
date such false positives and false nega- between them isn’t so irreconcilable.
tives. Someone who wants to reduce the
amount of money spent on social welfare
is probably bothered by the idea that false
positives are abusing the system by col-
lecting benefits they don’t deserve. Cheng diagram shows how combinations of priv- reinforce any kind of thinking, however
thinks the steely vocabulary of logic can ilege in multiple dimensions can form contrarian, problematic, or silly.
help people caught in a heated argument complicated hierarchies, so that people Yet there is still a sense in which
realize that the divide between them isn’t with three types of privilege are neces- Eugenia Cheng’s mission of demystify-
so irreconcilable (after all, they both want sarily better off than people with only two ing math is extremely noble. Her books
to help people) and steer them toward a types. But when Cheng uses this diagram try to replicate the humdrum ritual of
more nuanced conversation of “to what for insights on thornier questions, like why constructing arguments from ironclad
extent” and “under what circumstances.” “white men who did not grow up rich” proofs, and—more important—they show
Cheng believes we can encourage empa- may feel particularly aggrieved by non- what a math-inspired view of the world
thy through logically related analogies. white men who are richer and better off, could look like, both in its oddity and in
Initially befuddled by men who protest her answer is unsatisfying: in the cube of its permissiveness. You may find such a
sweeping accusations of privilege or aggres- privilege “there is no arrow from rich non- worldview odious and disagreeable, but
sion, she finds it helpful to compare their white men to non-rich white men (the two the key lesson from Cheng’s books is that
protests to the exasperation she feels when groups inhabit disconnected corners of a communicating a complex thought from
people resent graduates of elite schools (like diagonal), so the theory of privilege does one mind to another, let alone across cul-
herself) for having success handed to them not say anything about the relative situation tures and languages, is no easy feat and
by parents even though Cheng herself had of these two groups.” The cop-out may be that the art of expressing ideas charitably
to work hard. This, we are told, has made logically consistent, but it is certainly not and with clarity is something we all would
her more empathetic toward men: emotions the rhetorical coup de grâce one hopes to benefit from getting better at.
rightfully flare up when individual experi- learn after engaging with abstract reason- What I find most inspiring about
ences contrast with group generalizations. ing for several weeks. the culture of mathematics is how it has
However, applying such arguments One may equip people with rigorous endured through the ages, needling a com-
to more complex cases feels increasingly tools to avoid the slipperiness and ambiguity mon thread across civilizations. Math has
suspect. A diagram that appears in many of of everyday language, but these tools don’t managed to unify disparate discoveries
Cheng’s books is the “cube of privilege.” In always come with ethical guidelines. The across the globe, and the puzzles raised
one corner of the cube is the empty set {}. Malthusian panic over population explosion, centuries ago are still being pondered. One
Starting from that corner (bottom front left for example, emerged from observations reason this culture may appear mystifying
in the illustration on page 73), you can move about the exponential function and has for a beginner is that contemporary math
in three directions to collect one of three been used to justify anti-immigration pol- has whittled down millennia-old ideas,
types of privilege: white, male, and rich. If icies as well as genocides. Mathematically once rich and vivid, into terse symbols
you move in all three directions, one after inspired computer models are routinely and esoteric terminologies that aren’t
another, you end up in the opposite corner shown to have bias. A highly controversial always easy to master. Popular math books
with all privilege points: {white, male, rich}. book from 1994 hid its dubious efforts to seek a fresher take on these old ideas, be
To a category theorist, this is the most connect race with intelligence behind the it through baking recipes or hot-button
succinct description of intersectionality: mathy title The Bell Curve. As in the Bible, political issues. My verdict: Why not? It’s
the idea that attributes like class, gender, Tocqueville’s Democracy in America, and worth a shot.
and race can interact to produce com- other revered tomes, there is enough in Pradeep Niroula is doctoral candidate
plex manifestations of inequality. Cheng’s the vast literature of math to justify and in physics based in Washington, DC.
 How will generative AI
change your business?
Help your team stay ahead of the curve with a new series
of reports from MIT Technology Review
76 Field notes
 Field notes 77

Opposite: Marcin Below: The Maltron

Wichary and his enables people with
keyboards. “It sounds special needs to enter
really boring, right?” computer data much more
After discovering that a history of he says. “But if you easily and quickly
keyboards—from typewriters to look at it throughout than with conventional
the ages ... it keyboards.
iPhones—had yet to be written, becomes a lot more
this designer/typographer got to work. interesting.”

When the designer and typographer Marcin Wichary

stumbled upon a tiny museum just outside Barcelona
five years ago, the experience tipped his interest in
the history of technology into an obsession with a
Shift
happens
very particular part of it: the keyboard.
“I have never seen so many typewriters under one
roof. Not even close,” he shared on Twitter at the time.
“At this point, I literally have tears in my eyes. I’m not
kidding. This feels like a miracle.”
He’d had a revelation while wandering through
the exhibit: Each key on a keyboard has its own sto-
ries. And these stories are not just about computing
technology, but also about the people who designed,
used, or otherwise interacted with the keyboards.
Take the backspace key, he explains: “I like
that [the concept of] backspace was originally just
that—a space going backward. We are used to it eras-
ing now, but for a hundred years, erasing was its own
incredibly complex endeavor. You needed to master
a Comet eraser, or Wite-Out, or strange correction
tapes, and possibly all of the above … or give up and
start from scratch whenever you made a typo.”
The deeper he researched, the more fixated he
became. Amazed that no comprehensive book existed
on the history of keyboards, he decided to create his
own. When not working at his day job as design lead
for the design software company Figma, he began
producing Shift Happens, a two-volume, 1,216-page
hardcover book—and raised over $750,000 for the
project on Kickstarter in March of 2023. Wichary
was only a bit surprised by the support and the key- By Allison Arieff
board’s wide appeal. As he points out, “It’s such a
crucial device that occupies a lot of our waking life.”
78 Field notes

Some of these keyboards

are adjustable and
others fixed. The
ergoLogic FlexPro and
the SafeType (below)
eliminate pronation
of the forearms as well
as wrist extension.
 Field notes 79

These days,even the But some keyboards

cheapest desk go further than most,
keyboard is in some rotating or tenting
way “ergonomic,” their respective
allowing for reduced halves to allow a
effort and improved less stressful hand
response compared and arm position.
with even the best of
the mechanical and
electric typewriters
that preceded them.
80 Field notes

Other keyboards Clockwise from top

go even further, left, the orbiTouch,
questioning the Intellikeys, Big
very nature of keys. Keys, and DataHand
are each designed for
people with physical,
cognitive, or visual
disabilities.
GUTTER CREDIT HERE
 Field notes 81

Some keyboards a typewriter making

enable communication it possible to print
between people who in Braille; and a
might find that machine that allowed
difficult. Shown here people who are hard of
are a simple keyboard hearing to type over
connected to a telephone wires.
singular Braille cell;
GUTTER CREDIT HERE
KEVIN WALKER
 83

In the late 1980s, Stacy Horn By Nika Simovich Fisher Opposite: Stacy Horn, photographed
in 1994, still lives in the West
launched ECHO as a place for Village apartment that was her home
quirky, tech-savvy New Yorkers when she launched ECHO.
to congregate online. A few users
never left.

The lasting impact

of an online salon

O
ne January afternoon And now Rose, whose handle was KZ, interests were central to establishing the
last year, a bouquet was presented the same question, six and a vibrant cultural personality of this online
of balloons arrived half years into a battle with lung cancer that community.
at Karen Rose’s resi- she’d documented on a section of ECHO ECHO stands for “East Coast Hang
dence in Delray Beach, devoted to health. When she notified the Out,” and when Horn founded it, she
Florida. She wasn’t community that she was turning to hos- wanted to create a digital space that
expecting a delivery, pice care, her fellow “Echoids” responded was social and unequivocally New York.
since it wasn’t her birthday or wedding with the balloons, along with flowers and Members had to meet two requirements:
anniversary, and she thought someone had chocolate. they had to be geeky enough to navigate
made a mistake until she noticed the words Then last spring, ECHO’s founder, a cumbersome, text-based digital plat-
“AND NOW?” printed on each balloon. Stacy Horn, announced KZ’s death on form in the early days of the internet,
“AND NOW?” is the prompt that fol- ECHO. KZ was one of the platform’s 20 but culturally in tune enough to foster
lows every action on ECHO, a 34-year-old inaugural members, having joined in its the types of conversations you might
text-based social network that still hosts first year at Horn’s invitation and remained hear at a West Village dinner party. Horn
a community of former and current New until her death at 72. She was the host enlisted her graduate school friends
Yorkers. When you log in: AND NOW? of the network’s sex conference, a real (she was a recent graduate of New York
After checking who’s online: AND NOW? estate agent, artist, self-proclaimed “dance University’s interactive telecommuni-
Upon joining one of ECHO’s chat rooms, snob,” taiko drummer, tennis player, and cations program), as well as members
called conferences: AND NOW? general “doer.” People of such eclectic of other bulletin-board-style platforms.
84

One primary source of inspiration was orn, who is now 66, still lives in the
the California-based online community
known as the WELL (for “Whole Earth
’Lectronic Link”), started by Stewart
H same West Village apartment that
was her home when she launched
ECHO. When I met with her to discuss
Brand in 1985. Brand is well known for its origins, she had a neatly trimmed bob
being a counterculture impresario in the with bangs and wore denim jeans with a
Bay Area during the 1960s, editing the fitted black T-shirt—conjuring both Steve
widely distributed Whole Earth Catalog. Jobs and downtown “it girl.” She said the
Just as the WELL brought together exper- idea for ECHO came out of her day job as
imental, self-sufficient individuals who a telecommunications analyst at Mobil,
foresaw the endless possibilities of com- where she was the only woman in her
puters, ECHO defined the New York department.
web scene and influenced the design of Week after week, she’d pitch the idea
contemporary social networks, creating of “computer conferencing,” an efficient
lifelong friendships in the process. strategy to manage machines in different
When ECHO was founded, the World time zones that would post updates to one
Wide Web was still being invented, and continuously synced document. “I would
browsers weren’t a thing. Users congre- stake my entire future that this is going to
gated in interest-based forums, but Horn be the thing,” Horn enthusiastically told
found most of them to be male-centric, the team of corporate men about her plan.

“There wasn’t a velvet rope to get in, but you had to have certain chops
to be able to hang with those people.”

heavy in technical jargon, and, just like the She got the impression they thought the chair, brusquely called out, “Stacy Horn!
WELL, centered on the West Coast. She idea was laughable, and the answer was a Stacy Horn!” Horn interpreted the tone
craved a destination like the vibrant and firm no. Her boss suggested that obtain- as ominous, but to her surprise, Burns
artistic 20th-century salons of Gertrude ing a graduate degree would help Horn embraced her and said her paper was
Stein’s era, where users could exchange climb the corporate ladder, and while her more fun than anything she’d read in
ideas and meet one another while getting interests were shifting toward writing, she years. In that moment, Horn’s worldview
lost in discussion. thought graduate school sounded exciting. changed. “Oh my god, I can just go crazy
What she ended up making was a She picked the NYU program because it and somebody might actually like it,” she
hotbed of culturally minded early inter- had “telecommunications” in the title, so recalls thinking. Technology didn’t have
FREDERICK DUPOWERS; MAGAZINE FROM THE COLLECTION OF MIT LIBRARIES

net enthusiasts—a social network before Mobil would cover it as a work-related to be cold and impersonal. She became
there was a term for that. Through the expense. dedicated to experimentation and play
evolution of this ecosystem, users would Horn expected the program to be as in her work.
meet one another and contribute to the dry and technical as her job designing Back at Mobil, Horn decided that if her
changing digital economy by starting telecommunications networks, but she team could see a social network in action,
businesses and cultural programming. was taken by the school’s experimental they’d never go back. She started a trial
They would forever transform their lives philosophy. She wrote a play called Corpse program called MoNet (a portmanteau of
in a way that wouldn’t otherwise have in Space that took the form of a conver- “mobile network”), but to her dismay, it
been possible, all while making a lasting sation between a talking sofa, a praying flopped. (A few years after she left Mobil,
mark on New York’s budding tech com- mantis, and a dead saint. As she went to she says, the head of the telecommunica-
munity. ECHO was a blueprint for the turn it in, a pang of doubt overcame her; tions department told her that everyone
larger-scale social networks that we see she sheepishly placed her draft at the on the team had agreed to tank the project.
today, and it serves as a reminder that bottom of the stack of assignments and They were concerned that the platform
behind all networks are people, with a quickly left the room. The next time she would expose everyone’s work habits and
lot of words to exchange. was at school, Red Burns, ITP’s renowned amplify their mistakes.)
 85

building. But with the internet beginning to a tenth of the online world. Membership
take off, the phone company realized that snowballed after the appearance of a 1990
soon she wouldn’t be the only one needing New York Times story headlined “Coming
additional capacity. NYNEX ripped up the to the East Coast: An Electronic Salon,”
street and installed new cables that would placing ECHO at the forefront of New
support not only ECHO but the neighbor- York’s “Silicon Alley.”
ing buildings’ communication needs for
the foreseeable future. o extend the artistic component of
Horn recalls her neighbors being irri-
tated with the logistics of an internet busi-
ness running out of the apartment complex,
T the platform, Horn did outreach at
art openings and museums. She and
David Ross, then director of the Whitney
particularly during the cable installation, Museum of American Art, created an ongo-
but in the end, their neighborhood was ing series in which they’d pick a topic related
one of the first with stronger internet to visual culture and invite a panel of experts
connections that everyone could enjoy. to discuss it at a nonprofit performance
Back in her apartment, ECHO’s modem, space in the East Village. Other events the
housed in a custom cabinet with crimson community organized included “Dinner
Theatre of the Mind,”
a monthly seminar of
philosophical discourse
held by two members
Social butterfly known as “Neandergal”
From a 1993 profile in Wired (above): and “Miss Outer Boro
“I was pissed off that everyone was
exploiting this incredible communi- 1991”; an indepen-
cations device except women.” Right: dent film group; and
demoing ECHO on Charlie Rose in 1994.
the World Wide Web
Artist Consortium,
Upon leaving Mobil, Horn used the where participants met
software behind MoNet, known as Caucus, in real life to talk about
to set up ECHO. She pitched it as a social the internet.
community where interesting, thought- Ky l e S h a n n o n ,
ful New Yorkers could connect about the an actor and graphic
books they were reading and the places designer who founded
they were going, and ultimately get to sequins along the edges and gold tassels the consortium, says he initially joined
know one another on a deeper level. She in the front, would get so hot it warmed up ECHO to surround himself with people
wanted to create a “small town” feeling the whole space. ECHO’s server bounced who knew more about the web than he
where residents had a sense of pride. around New York before ultimately moving did. “There wasn’t a velvet rope to get in,
None of the investors she approached to a more stable facility in Oregon. but you had to have certain chops to be
were interested. At the time, she says, the At its peak in the late ’90s, ECHO had able to hang with those people,” he says.
consensus was that the only people who 3,500 members. Among them: writers, When Shannon and his wife, Gabrielle,
would want to talk to others online were artists, musicians, actors, therapists, and tried to post an e-zine called Urban Desires
socially inept weirdos. So she started the even, briefly, Robert F. Kennedy Jr. Horn online, his images didn’t load, and he
platform with $20,000 of her savings and hand-picked early members to help seed logged on to ECHO to see if anyone could
ran it out of her apartment. the community. To make women feel wel- help. A fellow Echoid, Chan Suh, responded
In those days, most people had only one come, she gave them free one-year mem- and revised his code. A month after the
phone line at home, while businesses would berships (ECHO cost $10 a month and e-zine launched, in 1994, Shannon learned
have a few more. Horn asked NYNEX, the $4 an hour for online time when it first that the French newspaper Libération had
VIDEO STILL VIA PBS

local phone company, to connect additional launched) and made sure to assign women run a full-page article about it. “The dis-
lines to her apartment. Before long she to host various conferences. Those efforts tance between putting something in the
needed up to 24 lines, which was more paid off—40% of ECHO’s users were world and having an impact just went to
than the maximum available for the whole female at a time when women made up zero,” he said after seeing the publication
86

on a newsstand in Times Square. By 1995, politics and taught entrepreneurship at a human touch became clear. ECHO’s pop-
Urban Desires had 100,000 site visits a nonprofit helping former drug dealers ulation was always small enough to afford
a day. He later partnered with Suh to start businesses, he moved back to New a more casual style of rule enforcement.
found an online marketing business called York. Living in the Brooklyn neighborhood Most incidents could be resolved with a
Agency.com. The company was bringing in of Fort Greene, which was going through face-to-face meeting, says Horn, who is
$200 million in revenue at its largest before a Black renaissance, he became part of a still involved in everyday administration
Omnicon acquired it in 2002, Shannon large community of Black professionals. of the social network—“babysitting us
says, and made interactive websites for He realized he wanted to create a digi- senior citizens,” as one user recently put it.
Fortune 500 companies, including British tal space that was inspired by ECHO but Today, with just 43 active users, ECHO
Airways’ first ticketing system and Sirius reflected his own interests and experience is a much quieter destination than it was in
Satellite Radio’s online player. of the city, for a community he connected the ’90s, but members still chat and bicker
There were weekly “F2F” (or face-to- with beyond the screen. with one another. After one recent dispute
face) sessions at downtown watering holes. In 1994 he started New York Online, a between three members, two of them were
After the parties ended, members would social network focused on highlighting a demoted to “read only” and Horn con-
log back on and keep chatting, sometimes multicultural experience in New York City. sidered closing the platform. When she
in private conferences. Online romances In a New York Times article from the year announced what she was thinking, users
blossomed. In her 1998 book Cyberville: of its launch, he compared the platform to balked: “If a plea would help you change
Clicks, Culture, and the Creation of an the subway: “It’s a network that connects your mind, ECHO has seen me through
Online Town, Horn described cyberspace you to the whole city, and you are always some of the most dramatic times of my
as the most erotic medium because of the surrounded by a really eclectic mix of folks.” life, and that is entirely due to your vision

Horn described cyberspace as the most erotic medium because

of the anticipation and thrill that messaging provided.

anticipation and thrill that messaging A few years later, as the internet became and your patience. I hope you will find a
provided. “Stacy always likes to say that more widespread, he launched BlackPlanet, solution,’’ a user named Schuyler Sue wrote.
[there are] children [who] wouldn’t exist a platform focused on Black Americans When not online, Horn spends her time
if it wasn’t for ECHO,” says Jim Baumbach, that became a precursor to social media working at the ASPCA and writing; her sev-
who met his wife, Liz Margoshes (a.k.a. platforms that updated in real time, like enth book is in progress. Although she is
Neandergal), on the platform in the early Myspace and Facebook. When Wasow sold not yet ready to step away from ECHO, she
’90s. Now they’re married, 70-something the site, in 2008, it had around 20 million has considered passing it on to someone
therapists who still use ECHO daily— even members and was the fourth-most-visited else to administer. And when it ultimately
going as far to send “YO’s,” ECHO’s ver- US social network. Kanye West mentioned fades away, she plans to donate ECHO’s
sion of a direct message, to one another flirting with women on BlackPlanet in his archives to the New-York Historical Society,
while in the same East Village apartment. 2004 song “Get Em High.” Wasow says securing any private conferences from
Shannon attributes ECHO’s success to that both ECHO and his platforms chal- release until those who participated in
its rootedness in a specific local scene. “A lenged the dominant ideas about who these them are no longer living. She takes pride
strong culture, by definition, has exclusion technologies were for, why they should be in the online culture she helped foster, one
criteria, whether they’re explicitly stated used, and who should use them: these com- in which language documents a communal
or not,” he says. munities “were prototyping the future in experience of passing time.
Omar Wasow, an assistant professor at which the internet belonged to everyone.” “On ECHO you own your own words,”
UC Berkeley, was on ECHO in his college Both Wasow and Horn have experi- Horn says. It’s one of a handful of guide-
years, when he was a student at Stanford. enced the pains of legislating a social net- lines that help keep the peace.
He’d grown up in New York, so the regional work. On BlackPlanet, there was a “fuck
aspect of the platform intrigued him, as did filter” that searched for curse words in Nika Simovich Fisher is a writer,
the focus on discussion. But Wasow says screen names and blocked them. But when graphic designer, and assistant
professor of communication design
he was more of a lurker than a participant. a user with the last name Bowcock was pre- at Parsons School of Design in New
After college, where he studied race and vented from accessing the site, the need for York City.
88

Screen time
We’ve been trying to figure out
how best to use computers in the
classroom for a very long time.

“Around the world, children have entered

a passionate and abiding love affair with
the computer. Notoriously, as with most
other love affairs of their children, par-
ents fail to understand it. In this case, they
are frightened by its intensity and espe-
cially distressed by what many perceive as
the mindlessness of the games that have
become its most visible manifestation and
a peephole into the future. The intensity of
feelings reflects children’s avid response
to a new medium that matches their ways
of knowing better than either print or tele-
vision. If the word ‘mindless’ applies here,
it is not to the children but to the adults—
especially to School, a sleepwalking dino-
saur of an institution that lumbers along a
set path, more and more out of alignment
with the society it thinks it serves and less
and less able to channel the energy and
vision of teachers who try to work in it.”

—SEYMOUR PAPERT, 1993

MIT Technology Review (ISSN 1099-274X), May/June 2023 issue, Reg. US Patent Office, is published bimonthly by MIT Technology Review, 196 Broadway, 3rd floor, Cambridge, MA 02139. Entire contents ©2023. The editors seek
diverse views, and authors’ opinions do not represent the official policies of their institutions or those of MIT. Periodicals postage paid at Boston, MA, and additional mailing offices. Postmaster: Send address changes to MIT Tech-
nology Review, Subscriber Services, MIT Technology Review, PO Box 1518, Lincolnshire, IL. 60069, or via the internet at www.technologyreview.com/customerservice. Basic subscription rates: $120 per year within the United States;
in all other countries, US$140. Publication Mail Agreement Number 40621028. Send undeliverable Canadian copies to PO Box 1051, Fort Erie, ON L2A 6C7. Printed in USA. Audited by the Alliance for Audited Media.