Wearable Computing

Editor: Bernt Schiele n MPI Informatics n [email protected]

Project Glass:
An Extension of the Self
Thad Starner

u r goal is to reduce t he
time between intention and
action. The first time I heard Google
CEO Larry Page say that phrase, we
were talking about the then-secret
Project Glass. However, this idea runs
throughout Googles products.
As an example, go to google.com
and type weara as if to start searching for wearable computer. After just
the first five letters, Google suggests
possible completions (including wearable shoe trees). By the time the letters
wearable com are entered, Google
has matched wearable computer
as the first completion result and has
already loaded results in the page below
the search box. Often, the information
the user needs will be on the screen
before the query is even finished. These
techniques save seconds in the interaction, completing the intention and making the Google search experience more
pleasant. On a mobile device, however,
fast interaction can be the difference
between a successful interface and a
device thats left in the pocket.

Hooked on HUDs
For the past 20 years, Ive been wearing
my computer as an intelligent assistant
in my daily life. Wearable computers
and head-up displays (HUDs) have
existed since the 1960s in some form,
but as far as I know, nobody has been
using them daily for as long as I have.
Ever since my first month of use in
1993, Ive been hooked.
The device enabled a compelling new
lifestyle in which I felt more powerful,

14

PER VA SI V E computing 

PC-12-02-Wear.indd 14

independent, and (most important to

me) in control of my time. I founded
the Wearable Computing Project at
the MIT Media Lab with students
who shared the wearable computing
vision, and we soon realized that the
functionality of most mobile consumer
electronics (music playing, video playing, voice calls, taking photos, recording video, and the nascent practices
of Web surfing and text messaging)
would eventually converge on one portable device.
Fifteen years later, smartphones
would take on this role, but they lack
immediacy. The process of retrieving
a device from the pocket, unlocking it,
and navigating the interface requires
around 20 seconds. This delay between
intention and action correlates with
how much a system is used in practice.1
I speculate that users perceive how
interruptive the retrieval and navigation process is and tend not to use
devices when it will require too much
interruption.
For example, at the end of my first
faculty meeting at Georgia Tech in
1999, the group was trying to schedule the next meeting. I was shocked to
realize that I was the only person who
was accessing his calendar in real time,
even though I knew other professors
had PDAs and paper-based day planners in their pockets. For me, the HUD
had removed the barrier to accessing
my computer, and there was little effort
involved in referring to it throughout
the day (as opposed to the pocketed
devices others were using). Even with

the specialized hardware, however,

I found it important to optimize the
user interface for fast interaction for
each function I intended to use. Otherwise, using the function was too much
trouble, just as with the PDAs my colleagues were neglecting.

Reducing Access Time

One frustration I had during that time
was with Web search interfaces. During
face-to-face meetings with a colleague
or student, I could refer to my calendar and personal notes quickly enough
to avoid interrupting the flow of conversation. However, if I needed some
information during a brainstorming
session, searching for it on AltaVista
required me to pause and interrupt the
conversation completely. I had to focus
on the screen, page through 14 different hits until I found a likely link, click
on the link, and then scan through the
page searching for what I wanted, succeeding only about half of the time.
The time between intention and action
was so long that I gave up doing Web
searches during a conversation unless
the information was absolutely critical.
Fortunately, I met two guys who were
working on the problem.
When Larry Page and Sergey Brin
first introduced themselves to me at a
conference in 1998, they asked for a
demonstration of my wearable and then
described their own research. While
graduate students at Stanford, they
had created a new algorithm, Page
Rank, that promised to provide better ranking of hits from Web searches.

Published by the IEEE CS n 1536-1268/13/$31.00 2013 IEEE

3/26/13 10:00 AM

I remember saying that if their algorithm could provide the information

I needed in the first four hits, it could
change the way I use my wearable computer. If fast enough, having Google
search on my eye would remove the
barrier to accessing the worlds information, making it available to me during
conversation and collaboration or simply
when I wanted to satisfy my curiosity.
Today, I use Google search on Glass
in a wide variety of contexts, both personal and professional (Figure 1 shows
a photo of the Glass HUD). Googles
search displays excerpts of the most
related content for each of its hits, which
means that the result of the search itself
often provides me with the information I need without even clicking on a
linksomething that I find critical in
my usage on a HUD.
Before I took a leave of absence from
Georgia Tech in 2010 to join Babak
Parviz, Steve Lee, and the Google guys
in creating Glass, my research group ran
many user studies focusing on how and
why a user might want a wearable computer with a HUD. We became more
convinced that the major advantage was
the reduced access time. When the time
between intention and action becomes
small enough, the interface becomes an
extension of the self.
An example is a man riding a bicycle.
If a car cuts him off, the rider doesnt
think, If I squeeze this lever on the
handlebar, it will pivot about its axle,
pulling a cable. Using a sheaf, that cable
will direct the force under the bicycle
and pull on a pair of calipers that will
then squeeze against the sides of my
back tire, and the friction will stop
me. Instead, the rider simply reacts
and stopsthe bicycle acts as an extension of the rider. To me, this example
demonstrates the difference between
doing research on human-computer
interfaces and doing research on interfaces that are extensions of the self. If
we can make mobile interfaces as quick
and natural as riding a bicycle, then we
make the user more powerful in a fundamental way.

AprilJune 2013

PC-12-02-Wear.indd 15

Figure 1. The Glass head-up display (HUD). (Figure courtesy of Google.)

While researching this principle, one

of our inspirations was the wristwatch.
The wristwatch requires around two
seconds to access, 2 which seems to
remove the barrier to use (what my
colleague Brad Rhodes calls the twosecond rule). Consider an interesting experiment that helps illustrate
the point. Next time you see someone
glance at their watch, wait a beat, and
ask again what time it is. Often, I see
the wearer glance back at the watch, as
if the time has changed somehow. Why
doesnt the user just say the time? Surely
he has just read it!
In fact, the users intention in looking at his watch may have been to see
if he was late to his next appointment.
He checked the time, realized he has
another 10 minutes, and dismissed the
issue from his mind without retaining the actual numbers. When asked
to report the time, it is easier to check
again than try to reconstruct it. The
wristwatch has become an extension of
the self. The wristwatch provides utility with such a minimum investment
of user attention that the wearer has
delegated that function to the device,
freeing his mind for more important
tasks. In fact, time is such an important function in our modern lives that

weve made a clock the default screen

for Glass (see Figure 2), and I would
argue its the most used feature on
the device.
The important question is, can we
make more wristwatch-like interfaces
that provide a maximum amount of
utility for the minimum amount of user
attention? We call using such interfaces
microinteractions and discovered that
these can be enabled through a coupling
of HUDs and gesture or speech input
(see the Microinteractions sidebar
for more information). For example, a
flick of the wrist could be recognized
by a wristwatch to trigger the display of
a contact list in a HUD. The challenge
is creating interfaces that dont falsely
trigger during the users everyday life.
Theres a delicate balance between
gestures that are simple enough to be
both memorable and socially acceptable
yet distinct enough to be recognized
accurately.2
While gesture can be used silently
and perhaps subtly when working with
others, speech input requires that the
wearable be brought into the conversation. For example, when discussing
methods to create multitouch tabletop surfaces, the wearable user might
need a specific piece of information

PER VA SI V E computing

15

3/26/13 10:00 AM

Wearable Computing

Wearable Computing

Figure 2. A user on a hot-air balloon ride, wearing Glass. A clock is the default screen
for Glass, so the user sees the time in the upper right corner. (Figure courtesy of
Google.)

Microinteractions
Unfortunately, smartphones have very limited potential for microinteractions, which
leads to users being mired and head-down in their touch screens. Once the user has
invested the 20 seconds needed to retrieve the device and navigate to the appropriate function of the user interface, she tends to devote even more attention to the
device while she has it in hand.
For example, when reading an incoming SMS, which should be a microinteraction,
the user is soon lured into a scan of the email inbox. To make matters worse, smartphone user interfaces require significant hand-eye coordination, with the users eyes
typically cast down toward the floor. The users attention is tunneled away from
the world around her, into the virtual. However, this attention-tunneling problem is
alleviated once an interaction becomes micro, with an interface designed to get
the user in and out of the interaction as quickly as possible. Even better, a head-up
interface, when used for a microinteraction, keeps the users eyes pointed outward
toward the world, which also helps avoid attention tunneling. Compared to current
devices, wearable interfaces will help the user pay attention to the real world as
opposed to retreating from it.

to continue the conversation and say,

Let me google the refractive index of
acrylic.
A particular advantage of this technique is that it both communicates the
request to the computer and informs
the conversational partner as to the
wearers use of the machine.3 I prefer
such interactions, because they encourage socially appropriate use. Through
this technique, other functions of the
wearable computersuch as taking a
picture or recording a videobecome
transparent to bystanders. Having a

16

PER VA SI V E computing 

PC-12-02-Wear.indd 16

see-through display provides further

cues to conversational partners as to
the use of the HUD (looking closely at
the display in Figure 3, made for me by
Mark Spitzer in 1997, you can almost
read the text on the screen).

Figure 3. Having a see-through display

provides cues to conversational partners
regarding the use of a HUD. Looking
closely at the HUD Starner is wearing
here, made by Mark Spitzer in 1997, you
can almost read the text on the screen.
(Photo courtesy of Sam Ogden; used
with permission.)

the past 20 years. I encourage you to

participate in exploring this new lifestyle, enabled by Glass.
References
1. T.E. Starner et al., Use of Mobile
Appointment Scheduling Devices,
Extended Abstracts on Human Factors
in Computing Systems (CHI EA 04),
ACM, 2004, pp. 15011504.
2. D. Ashbrook, Supporting Mobile
Microinteractions, doctoral dissertation, School of Interactive Computing,
Georgia Institute of Technology, 2009.
3. K. Lyons, Improving Support of
Conversations by Enhancing Mobile
Computer Input, doctoral dissertation,
School of Interactive Computing, Georgia Institute of Technology, 2005.

Thad Starner is a technical

lead/manager on Googles
Project Glass and is a profes-

oon, Google Glass will be worn

by many more users as part of the
Glass Explorers Program. This deployment will be the largest ever of its type,
and we expect to learn more in a few
months than my teams have learned in

sor at Georgia Tech. Contact

him at [email protected].

Selected CS articles and columns

are also available for free at
http://ComputingNow.computer.org.

www.computer.org/pervasive

3/26/13 10:00 AM