Designing Artificial

Intelligence Products
Workbook
Your AI Mascot will guide you through
this week, so watch out for these icons:

Read

Write

Consider

Visit the website

Design/Create

1
 Workbook 7: The Sustainability,
Feasibility, and Responsibility of GANs

Using GANs for Good

Week 7

As you’ve seen throughout this module, AI has raised the stakes

in the realm of images, video processing, and video alterations.
You’ve explored how GANs can enhance as well as mislead.
Undoubtedly, this technology has been phenomenal in many
areas, including 3D medical imaging for better diagnosis. But
when left to its own devices and without ethical guidelines, these
very processes can be detrimental in areas such as social and
mainstream media. Where do we go from here? How can we you
use GANs for good?
For this task, you’ll use a GAN to solve a business problem for
the good of society. Follow the steps below to complete this
workbook entry.

2
 1 Step One: Select a Problem

Select a problem that needs to be solved or a process that needs to

be improved. This problem can be something in your own business,
organization or personal life.

Keep in mind just how powerful GANs are. GANs have been described
as robot artists that can learn to mimic any data distribution.

Type your response here

3
 Step Two: Assess the Target Market
2 (Sustainability)
Based on the choice you've made in Step One, conduct some research
and determine your target market.

The following questions.

Will your How will How will How will Can you
product or you address your product you measure maintain
service the needs or process progress? this product
serve of your compare or process
a relevant target to the over time?
need? market? competition?

Type your response here

4
 Step Three: Assess the Finances
3 (Feasibility)

Now, consider the following questions.

What are Do you have Which What are the How much
the current a unique distribution technology will it cost
market value channels investment to train
trends for proposition? will you costs? your AI
your product use? process?
or process?

What kind Is the data How much Does your How much
of data do publicly or will it cost organization will it cost to
you need? readily to maintain have the outsource
available? the AI technology the project?
technology? talent to
develop
the resource?

5
 Step Three: Assess the Finances
3 (Feasibility) (Cont.)
In this step, sketch out a cost analysis for the project.

Type your response here

6
 Step Four: Assess Societal Matters
4 (Responsibility)

Microsoft’s Office of Responsible AI along with the AI, Ethics, and Effects in
Engineering and Research Committee have developed the six principles
of responsible AI. They include:

Fairness: Our society is unfair and biased in many different ways. The whole
point of focusing on fairness in AI systems is to make sure that the system
that we develop and deploy reduces unfairness in our society rather than
keeping it at the same level or making it worse. AI systems can be used to
allocate and withhold opportunities and resources in different areas, such
as criminal justice, employment, and finance. Another example would be
cultural denigration or over- and under-representation. The problem is that
there isn’t a single definition that we can easily quantify and integrate into
our systems. Fairness relates not only to the technical context of the system
but also to the societal context in which the system is deployed. Therefore,
fairness is a socio-technical challenge that can be overcome by having a
team with a greater diversity of people developing AI systems.

Reliability and safety: Reliability and safety is a concern for any system.
You need to make sure that the systems you are developing are consistent
with the design ideas you have and that they work in a way that is
consistent with your values and principles. To achieve this, the models
must not create any harm in the world. In situations where products may
make mistakes, their deployment should be preceded by sharing the risks
and harms clearly with the users. This is the principle of ensuring the
reliability and safety of any AI product. For example, risks of self-driving
cars are not limited to physical systems but are also about human lives,
as wrong systems can create harmful situations for humans.

Privacy and security: Privacy is a fundamental right. We add complexity to

AI and ML systems and increase reliance on using data to develop and train
them. This increasing reliability on data adds new requirements for keeping
the system secure. For example, you need to ensure the protection of that

7
 Step Four: Assess Societal Matters
4 (Responsibility) (Cont.)

data so that it doesn’t get leaked or disclosed. One approach could be not
to remove data from a user’s device by running the models locally on the
device. You need to think about where the data is coming from, if it’s a
user-submitted data or publicly outsourced. Finally, you need to check if
the data is corrupted.

Inclusiveness: When developing an AI product, you must make sure that

the product can engage with every community in the world. You must be
intentionally inclusive and diverse with the approaches taken towards AI.
Moreover, you need to make sure that designing a product is done with an
inclusive perspective towards the future of AI – this can be achieved by
ensuring that people from minority communities are involved with the
process of both designing and testing an AI product.

Transparency: Transparency and intelligibility can help achieve a really

diverse range of goals: from mitigating machine unfairness in machine
learning systems to helping developers debug their AI systems and
gaining more trust from the users.
There are two sides to transparency: The first one is that the people
developing AI systems should be open about how and why they decided
to use AI and the limitations of their systems. Transparency also means that
people should be able to understand the behavior of AI systems in terms
of interpretability or intelligibility.

Accountability: Despite all the complexities with these new models and
with new technology that can be unpredictable and difficult to interpret,
you are still accountable for how a technology impacts the world.
Accountability is also the structure that you put in place to make sure
that you are consistently embedding the principles in everything you
design and deploy. Accountability also means helping users be accountable.
For example, you must develop, sell, and advocate for any technology, such
as facial recognition or any other product you develop.

8
 Step Four: Assess Societal Matters
4 (Responsibility) (Cont.)

Consider the following questions (based on Microsoft AI principles).

Does your Is your AI Is your Does your

AI process process AI process AI process
treat people reliable? safe? guard the
fairly? privacy and
security of
your target
population?

Is your Is your Are you

AI process AI process willing to be
inclusive? transparent? accountable
for the
actions of
your AI
process?

9
 Step Four: Assess Societal Matters
4 (Responsibility) (Cont.)

Now, create general guidelines for addressing ethical and social issues
related to your AI process.

Type your response here

10
 Designing Artificial
Intelligence Products
Workbook