Ethics in the Age of Generative AI

Distinguishing responsible tech from human behavior

- Generative AI is already transforming every aspect of the human experience, and I'm
the kind of optimist who believes that these tools will make us more human, more
creative, inspired and connected. In the banking sector, AI and machine learning are
helping us identify people who might have opportunities to be more financially secure, to
save more, to increase retirement contributions and to have better pathways to
economic opportunity. In agriculture, AI models are helping to predict large weather
events so farmers and producers can understand when additional insurance is wise or
help them know exactly when to plant or harvest to best maximize their economic
returns. And maybe closer to home inside of organizations, AI is transforming human
resources, helping managers understand how to inspire better performance from their
teams, correct potential biases or discrimination and make sure that promotions are
truly merit based. But even as I'm excited about these potential opportunities ahead, I'm
also convinced that we need to make sure we build these tools with positive intention
with a grounding in ethical and responsible reasoning. And I'm not alone. From the very
beginning, developers of artificial intelligence have known the incredible power of these
tools and consider the ethical quandaries that might arise when we deploy them. In
recent years, these apprehensions have reemerged with a sharper edge. Researchers
and now policymakers are exceptionally concerned by the potential ways that AI could
perpetuate bias, could make the world more unequal and could do so in ways that are
invisible to us. Many of the ethical concerns that AI researchers have worried about are
coming to light in a very real way. Consider the idea of deep fakes, tools that might
create a persona or an avatar that's pretending to be a trusted person in your
organization, delivering fraudulent information to your customers, or even advising them
to take dangerous or potentially risky courses of action. We've seen the advent of
chatbots everywhere and we know that without ethical design, chatbots might give false
advice maybe to medical practitioners or to students. Information that sounds logical,
but is in fact based on inaccurate or untruthful information, and has never been audited
by a human being. And issues about fundamental ownership, questions of legal and
copyright. Who owns the ideas and products that are created by generative AI? We're
just beginning to resolve them. Join me for a moment in an example. Let's imagine that
your company has deployed a new AI system to support the HR function, scanning
resumes of applicants to identify potential interviewees. At first glance, the tool works
incredibly well, operating just as quickly and providing the same number of candidates
as your human support team. But as you dig deeper, some disturbing patterns emerge.
The tools prioritizing a particular gender, folks from a particular address or
neighborhood or with a particular pattern to their work history. And you realize that
these are the same biases that the humans in the dataset that trained the tool were
expressing. We have a challenge here. This algorithm now is providing bad
recommendations but humans are the ones to make the decisions about who they will
interview. In situations like these, we need to ask ourselves, what is the highest
standard of responsible human behavior? What actions can we take to best promote
fairness and dignity? And is it possible that we've trained an AI to provide an answer
that's lower than that highest standard? When this happens, what can we do to help
humans make better decisions based on the algorithm's recommendations? If you're
eager to learn more about how to answer these questions, stay with me. Later in this
course, we're going to cover step by step how to resolve dilemmas just like these.

Understanding Vilas' ethical AI framework

- I'm so excited about how quickly we're building new generations of AI tools, but I know
that we need to make sure we're designing tools that support the future we want to
create, equitable, sustainable, and thriving. And to do this, we're going to have to come
up with new frameworks for ethical creation just as quickly as we advance the frontier of
innovation. So how do we translate intuitions and hopes into clear principles for decision
making? I'd like to share with you a three part framework that I use for evaluating and
advising organizations on the creation of new ethically grounded AI tools and it works
equally well for technologists and non technologists. The three pillars of the framework
are responsible data practices, well-defined boundaries on safe and appropriate use
and robust transparency. Let's start by talking about responsible data practice. This is
the starting point for all ethical AI tools. Any new technology is only as ethical as the
underlying data that it's trained on. For example, if the majority of our consumers to date
have been of a particular race or gender when we train the AI on that data, we'll
continue to only design products and services that serve the needs of that population.
As you consider building or deploying any new tool you should ask what's the source of
the training data? What's been done to reduce explicit and implicit bias in that dataset?
How might the data we're using perpetuate or increase historic bias? And what
opportunities are there to prevent bias decision making in the future? The second part
of the framework is the importance of creating well-defined boundaries for safe and
ethical uses. Any new tool or application of AI should begin with a focused statement of
intention about the organization's goals and an identification of the population that we're
trying to serve. So for example, a new generative AI tool that can write news articles.
Well, it could be used to help tell the stories of a wider range of underrepresented
voices. We could use it in new languages or it could perpetuate misinformation. When
considering ethical use, you should ask who's the target population for this tool? What
are their main goals and incentives and what's the most responsible way to make sure
we're helping them achieve those goals? The third part of the framework is robust
transparency. We need to consider how transparent the recommendations of the tool
are, and that includes how traceable those outcomes are. This allows for human
auditing of ethical accountability. When it comes to transparency, you should ask how
did the tool arrive at its recommendation? And sometimes it's not possible to know, but
if so what are other ways we have of testing its fairness? Is it possible for decision
makers to easily understand the inputs, analysis, outputs, process of the tool? And
finally, have you engaged with a broad range of stakeholders to make sure that this tool
promotes equity in the world. As you embark on building and using increasingly more
complicated ethical AI tools, this framework of responsible data, well-defined
boundaries and robust transparency should provide you with a foundation for making
smarter, more informed decisions.
Organizing data with ethics in mind
- I remember standing at the top of the Burj Khalifa, the tallest building in the world. As I
looked out on that view, a part of me was wondering just how strong the foundations
must be to protect the people inside. These days, I look at generative AI models that
can do amazing things and I find that same part of me evaluating the safety, trust, and
design that ensures that these models will inspire and protect all of us. AI models are
built on top of data. So let's talk about the importance of ethically organizing and using
your institution's data. By taking an ethical approach, you'll reduce the risk to your
organization and you'll increase the value of that data as an organizational asset. There
are three goals in effective and ethical data organization: the first is prioritizing privacy,
the second reducing bias, and the third, promoting transparency. The first consideration
is prioritizing privacy. Almost every organization collects sensitive data about customers
and employees; things like personal healthcare information or financial and banking
details. And customers and employees trust the organization with this data so it's
important to handle it sensitively and ethically. Failing to uphold this trust can expose a
company to liability and reputational harm, and maybe, most importantly, erode trust
with your customer. So to test your company's practices, you can lead a privacy audit.
During a privacy audit, you build a comprehensive understanding of what data your
organization has, how it was collected, how it's stored, and how it's administered. The
results of a policy audit inform recommendations to create or adapt your existing privacy
policy to protect sensitive data. With a privacy policy in place, the next step is to create
a training curriculum for all employees that focuses on understanding why sensitive data
must be handled securely and advises them of their responsibilities. The second goal is
reducing bias in data collection and in data use. Bias in data can arise from a number of
sources and understanding how it makes its way into your dataset requires genuine
curiosity in your analysis. To start a bias audit, be curious about whether the data really
represents the population you're trying to serve. For example, I recently worked with an
organization building AI for cancer screening. And as they tried to deploy this tool, they
found that early models exclusively use training data from the global north, requiring a
retraining of the model to make it useful for a global population. So does your dataset
represent inputs from a diversity of individuals across race, gender, age, and more? Are
we asking the right questions when we collect data? You might also consider whether
your data collection process was accessible to differently-abled people. And finally,
once the data is collected, you might consider whether a team with relevant and diverse
lived experience has an opportunity to analyze and interpret this data to reduce the risk
of potential bias. Bias is especially important when we attempt to explain how our
algorithms make recommendations that have real impacts on people's lives. For
example, recent studies have shown that early attempts to automate hiring have
propagated existing biases and employment practices. Understanding the bias in the
data helps us minimize the negative impacts of bias in the algorithm. After you've
completed your privacy and your bias audits, transparency is the final step in the
process. You want to be able to explain to all stakeholders, your customers, your
employees, your suppliers, your regulators, how data is collected and used. You might
consider publishing a data governance framework or a data transparency statement to
help your stakeholders understand what you do with their data. And you should also
make it clear that individuals can access any data that you might have stored about
them and have rights on how you might use it on an ongoing basis. Organizing and
understanding your data helps you understand your customers better, ensure they're
well represented, weed out biases, and builds a stronger foundation for your AI products
and tools.

Preparing technology teams to make ethical decisions

- Technology teams face some special challenges when it comes to ethical decision
making. You know, teams are asked to build new technologies for business challenges,
but they're not always asked to consider upfront the social expectations and challenges
that come with that design. Together we'll explore some ethical dilemmas that might
face technology teams and learn together how to create a culture that promotes ethical
decision making and accountability. Technology teams are unique in large
organizations. First, they're often composed of individuals with really specific skills and
expertise, which is frequently not so well understood by others in the organization, and
sometimes even by the team's own managers. Second, technology teams often work at
an extremely fast pace under tight deadlines. This means they have little extra time and
resources to audit or to reflect on the consequences of their decisions on users or on
the wider society. And finally, technology teams are often subject to specific regulatory
requirements, for example, GDPR and associated laws in Europe. All of these
considerations mean that it's important that technology teams have a strong, internal
ethical culture, along with external oversight and accountability, to make sure that we
are making decisions ethically. Ethical decisions that technology teams might face
include: ensuring the security and privacy of data collection; storage, use and reuse;
creating and auditing algorithms to ensure that they're fair and they're free from bias;
and understanding the data storage and other environmental impact of their decisions
and considering opportunities to reduce technology's carbon footprint. Can you think of
any others? Every team will face their own unique ethical challenges, which is why it's
vital to foster a culture of ethical decision making, one where teams can respond to an
array of challenges as they arise, or even, hopefully, prevent them from arising in the
first place. Here's a few steps you can take to create a culture of ethical decision
making. First, foster a culture of ethical communication within your team. Encourage
every team member to openly raise questions and concerns about the ethical use of
technology. Here's an idea. Start meetings by focusing on a recent ethical challenge
that your team faced and discussing how it was resolved. Reward and celebrate team
members who appropriately raise and resolve ethical issues and make it a part of your
day to day. Next, you might consider establishing a technology-specific training
curriculum for your team, focusing on emerging technologies and the ethical challenges
that team members might face when they begin to deploy them. You should explicitly
consider ethical challenges at the start of every project. Before launching a new
initiative, your team should come together to discuss possible ethical dilemmas and
consider potential remediations and decide on a path forward. And when you feel like
you need extra support, you should reach out to academics or philosophers, individuals
who can help your team understand the full spectrum of ethical challenges that might fit
within your work and bring them into a framework focused on advancing both your
product and social wellbeing. Giving your technology team the tools to ensure that they
can make decisions that align with the company's values and wider social ethics is
essential. It sets your team up for success and it empowers them to solve new
challenges as they arise in the future.

Preparing C-Suite in directing responsible AI

- CEOs and the C-Suite play a critical role in building cultures of responsible AI. They
set the tone by establishing practices and principles and they ensure that every
individual in the organization feels like they're a part of making ethical decisions. Earlier
in this course, we discussed the example of Alice Wong and her company that faced a
critical dilemma around the deployment of an AI chat bot. In that instance, if we were
advising the C-suite of the company, we'd start with the following recommendations.
First, to make sure that a responsible AI policy and governance framework is in place.
This is a statement from the C-Suite about how the organization should design and
manage AI technologies. It should describe how to make ethical decisions. It should
protect privacy, and it should focus on the elimination or reduction of bias. For example,
the C-suite might mandate that AI tools are trained with diverse data sets or they might
require that chatbots are always identified as an AI and not impersonating a human
customer support agent. These guiding principles create a shared set of values that
everyone from data scientists to supervisor, to field staff, can use to evaluate and guide
the deployment of artificial intelligence. Next, we might advise the C-Suite, provide an
maybe even mandate responsible AI training and education for every person in the
organization. This method of democratizing decision-making around AI tools can be
very powerful, can bring business knowledge, present and frontline service fields to help
train and develop internal models. For example, in Alice Wong's example, Alice relied
on customer service agents with years of direct customer experience, helping solve
consumer challenges to validate the recommendations of the AI models. Empowering
these agents to understand the limitations of the model can increase the quality of their
feedback. Then, C-Suites should insist on building ethical AI elements into all of their
technologies and conduct regular audits. The C-Suite can identify specific metrics such
as customer satisfaction and create regular reporting mechanisms to ensure that the
company's AI practices are aligned with responsible AI principles. Here's an example.
That might look like setting up monthly standups, where technology executives join the
C-suite and present ethical challenges that have emerged in the past month. This could
be the start of a dialogue with C-Suite executives understanding and documenting
ongoing interventions and improving the ethical nature of the product. Much like safety
practices in other industries, this practice socializes and makes ethical AI development
a shared and accountable responsibility. Finally, the C-Suite might consider hiring a
chief AI ethics officer. The company might establish a specific senior role, focused on AI
ethics that can develop and oversee the use of responsible AI practices and serve as a
central audit for other departments. They should understand the intersection of the
business the technology and the customer experience and they could provide a check-
in balance for technology development to ensure that community voices are also
present and ensure that potential risks are identified early in the creation process. The
C-Suite sets the tone for responsible AI across the organization, creating strong
policies, ensuring that there's appropriate training, establishing monitoring and reporting
mechanisms, and potentially creating roles focused on AI Ethics. With the C-Suites
primary responsibility for guiding responsible AI, we can build cultures that focus on
ethical decisions, even as we deploy great new products.

Preparing the Board of Directors to manage risk and opportunity in AI

- I work with a lot of board directors, and they're having trouble sleeping right now
because I can tell you, even the smartest AI can't possibly predict all the ethical
dilemmas that might arise from these new technologies, and yet it's the board's job to
make sure that organizations are prepared, to make sure that new technologies are
deployed in the best interests of all stakeholders. Board directors have a legal and an
ethical obligation to act in the best interest of an organization and its stakeholders.
Board members have different responsibilities from the organization's C-Suite. As you
might recall from the previous video, the C-Suite's responsible for the day-to-day
operations of the company, making real-time decisions about when and how to use new
technologies. For a board of directors to ensure that an organization's use of AI aligns
with ethical values and regulatory requirements, they should first make sure that the
organization has policies and procedures in place to identify and address ethical
concerns that might arise in the use of AI. These policies should be designed to mitigate
risks, including bias, privacy, and security, but they should also create opportunities for
individuals with ethical concerns to come forward, including directly to the board when
and if necessary. The board should make sure that the organization has their resource
and the expertise necessary to manage the ethical risk posed by AI effectively. Do you
know if there's a board policy on the ethical use of AI at your organization? If not, it may
be worth it to find out. Board members also have specific responsibilities and obligations
to regulators. They're accountable for ensuring that the organization complies with
statutory requirements related to the use of AI as well as monitoring all of the new
regulations and understanding the possible impact for the organization. To be effective
at these responsibilities, the board of directors should establish a dedicated AI
committee to oversee ethical AI practices within the organization. This committee
should receive outside advice from experts in the fields of AI, ethics, and law, and
should be responsible for providing guidance on issues such as bias, transparency, and
accountability. They should, in particular, be available to advise the C-Suite on
significant decisions about these matters. The board of directors has a critical role in
managing ethical risks within organizations. They must ensure that the organization has
appropriate policies and procedures in place to address ethical concerns, but they also
have to provide guidance and oversight to the executive team, ensure compliance with
regulatory requirements, and they should establish a dedicated committee to oversee
ethical AI practices within the organization. By taking these actions, boards of directors
can build trust with stakeholders and they can ensure that organizations are focused on
the long-term success of their AI efforts.

Consulting your customers in building AI

- In the previous videos, we've explored the importance of technology teams, of C-

suites, of boards of directors in ensuring responsible and ethical AI practices. But what
about the most important stakeholder of all, our customers? Designing great products
means that we have to understand and incorporate their preferences, their needs, and
their wants into our product design. I'd like to share with you a powerful framework for
listening to our customers, an acronym that I call LISA. First, we listen to users before
we start to build. Developing and launching new technologies requires a clear
understanding of our users' goals, their needs, and their fears. It can be difficult to
create a product when we haven't heard what our customers expect. Research has
shown that users care deeply about the experience and usability of the technology
products they use. In a recent survey conducted by the Nielsen Group, 85% of
respondents said they would not return to a website or a product if they had a poor user
experience. The second part of the LISA framework, how do we involve our customers
in design decisions? We know that our customers want to feel that their opinions matter,
and we want to include them in design decisions that can be crucial to building our
products to meet their needs. This can be especially helpful when we're seeking to
ensure that our decisions reflect the full diversity of our user base. Here's an example.
In 2016, Airbnb launched its Community Commitment initiative. They gathered input
from users on ways to make their platform more inclusive and welcoming for people
from diverse backgrounds. This simple practice led to the creation of brand new
features, such as the filters for gender neutral pronouns or the ability to search for
wheelchair accessible listings. Another way to involve customers and design decisions
is creating a user advisory board, a group of users who are invited to provide feedback
and input on new features, designs, and other aspects of product development, even as
you're in the design environment. For example, Microsoft has done this with a customer
advisory board made up of customers from a range of industries and backgrounds who
are invited to provide feedback on Microsoft's products and services, even while they're
still in development. By including users from diverse backgrounds and experiences, we
get a much wider range of perspectives on how to design and build products that
actually meet the needs of all users. The third part of the LISA framework, sharing
simple and transparent privacy policies. By prioritizing user privacy, we focus on
building trust with our users and we create a more loyal user base. This is so important.
According to a survey by Pew Research Center, 79% of adults in the US are concerned
about how companies use their data. This can be a barrier that keeps users from
engaging with your products, even when these products might actually help them
improve their lives. There are ways that we can do this that include using plain language
to explain data collection practices, providing customers with clear opt-in and opt-out
options, and implementing privacy by design principles into your core technology
development process. The final part of the LISA process is auditing our work and
inviting outsiders in to help hold us accountable. Every existing and new technology
product should be audited on a regular cycle, a process where you review the purpose
of the product, potential risk to users, and maybe most importantly, the possible
unintended consequences that might happen because of that product. Here's an
example where this works well. Google developed an AI principles framework which
guides their development and use of AI technology. That framework includes principles
like fairness, privacy, and accountability, and it's used as a guide for conducting regular
audits of their AI systems, identifying potential risks, and coming up with remediation.
There are a number of risks we should be aware of. They could include bias or data
privacy concerns, or even security vulnerabilities. And we know that organizations that
bring in users and audit these risks do a great job of responding to them. At OpenAI, the
trust and safety team is responsible for identifying potential risks associated with AI
technologies. The team includes experts in the fields of computer science, law, and
philosophy. They work together to ensure that OpenAI's technology is ethical and
responsible. Once potential risks are identified, then we have to step back and conduct
a risk assessment to evaluate the likelihood and possible impact of those risks. This
assessment should consider the potential impact on users as well as the business
impact of the risk. For example, at IBM, there's a separate AI governance board that's
responsible for conducting risk assessments for AI systems. The board evaluates the
potential risks and makes recommendations to mitigate those risks and improve safety
for users. Building great products means listening to our customers, and using the
framework we've described here, affectionately termed LISA, lets us listen to our
customers, lets us involve them in decision making, shares privacy practices, and
ensures that we're living a practice of regular audit and accountability. These practices
mean that we can build better trust with our customers and ensure that technologies
meet the needs and preferences of communities, not just the ones we serve today, but
the ones we aspire to serve in the future.

Communicating effectively organizationally and globally

- In this course, we've talked about the decisions you will make as a leader in defining
responsible AI practice and the roles of those around you in firms and organizations.
But as we speed into the transformation the generative AI reflects, we know these
products will touch every person on the planet, and it's important for us to consider the
interests of various stakeholders. I use a convenient acronym called ethics to remind
myself of the specific responsibilities of each stakeholder group. Using this mnemonic
can help ensure that you're fulfilling your responsibilities to core AI action and including
stakeholders across the globe. The ethics framework outlines six key stakeholder
responsibilities for responsible AI. First, E for executives and board members. Top
management has a responsibility to establish ethical AI cultures across organizations.
This includes setting ethical guidelines and standards, ensuring that ethical
considerations are integrated into decision-making processes, and allocating resources
for ethical AI deployment and development. The T in ethics, technologists, engineers
and developers who have a responsibility to design and develop products that are
transparent, explainable, and accountable, avoiding bias in data and algorithms,
ensuring that systems are secure and safe, and developing AI systems that are
compatible with existing ethical frameworks. H, human rights advocates. Human rights
advocates have a responsibility to ensure that the systems that technologists build
respect human rights and dignity. This includes monitoring how AI systems are being
used by vulnerable groups, identifying potential human rights violations, and advocating
for the ethical use of AI. Is for industry experts. Industry experts have a responsibility to
share their knowledge and expertise on the ethical implications of AI. This might include
providing guidance on developing tools, identifying potential risks, and collaborating with
other stakeholders to address ethical concerns. C, customers and users. Customers
and users have a responsibility to provide feedback and insights. This could include
communicating concerns, feedback to relevant stakeholders, participating in user-
testing and feedback sessions, and staying informed about the ethical implications of AI.
And finally, S for society at large. Let's acknowledge that this is a shared journey that all
stakeholders have a responsibility to consider how these tools are changing the ways
that we interact as humanity. This includes identifying and mitigating potential risks,
promoting and advocating for transparency and accountability, and making sure that AI
is used in a way that benefits society broadly. It's vitally important to coordinate these
different stakeholders to create new spaces and forums for groups to come together. It's
not enough to have each stakeholder playing their part. We have to coordinate across
different stakeholder groups. Everybody needs to know what the others are doing and
that means, we need to create new forums and new participatory mechanisms to make
sure that stakeholders are working together to maintain ethical AI. Here's a few
suggestions of what you might be able to do to promote the ethics framework. You
could establish new mechanisms of clear communication between the stakeholders in
your work, your customers, your executives, and your technology teams. You could
develop training programs to educate employees and stakeholders about ethical
considerations in AI. You could advocate to create a cross-functional team within your
organization, or even a cross-organizations within an industry, bringing folks together
from different departments to develop and implement guidelines and standards. You
might consider developing a system for collecting and addressing user-feedback
particularly around concerns and risks about AI systems. And you might consider
engaging formally and informally with external stakeholders, human rights advocates,
industry experts in civil society to ensure that we're considering the broadest possible
implications of AI. We're at a moment in time where building products feels like the most
important way to explore what generative AI can do for humanity. And yet, if we build
products without also asking how those products will be used, what needs they serve,
and how they'll impact vulnerable people, we miss an opportunity to use AI to make
humanity better. The ethics framework gives us a way to encourage and involve
stakeholders from across society, to make sure that as we build products, we're also
building an AI ecosystem for the future of humanity.

Setting an intention of continual questioning

- You've completed the course and in this time together you've learned to evaluate
ethical risks, to design organizations that can handle them, and build mechanisms to
seek broad societal inputs on thorny issues. You should now be able to build products
that are grounded in both ethical principles and ethical practice, but the journey's only
starting. Here are three ways you can take what you've learned to grow as a
professional and be a leader in the ethical AI transformation. First, I encourage you to
find innovative ways to incorporate communities in what you're building. Can you
partner with local nonprofits to find problems worth solving, and can you use your
expertise to expand the impact of the tools you're already building? Second, I invite you
to focus on building skills beyond technology. Even as you become a better coder or
designer, what if you could improve your capacity to understand how your product
choices impact economic opportunity, help communities solve local problems, or even
address some of the greatest challenges facing humanity? And third, and most
important, I ask you to make an intention in your daily work to become a steward of a
human-centered future, one that recognizes the critical moment we're in, where our
decisions will shape the future for generations. Working together, we can shift a
conversation about building ethical AI into one about building an ethical society that's
powered by AI, a future that inspires us to be our best human selves.