Responsible Artificial Intelligence (RAI): A Primer
Updated: Mar 8, 2021
AI, broadly defined, is ingrained almost everywhere we look. Understand what it is and how and why we all can and should address it responsibly.
As we began our journey, we had to decide our focus and our language. It quickly became "Responsible AI". But what does that mean?
We identified Responsible AI as both the right opportunity and term that encompasses all we hope to achieve. However, as we started to talk to others, we quickly realized that many people had different understandings of what Responsible AI is, the opportunity it presents, and when that opportunity will exists (hint - it is already here!).
So, we want to begin our journey with you by laying out how we think about Responsible AI and its immense opportunity. In our first post, we hope to:
Demystify what Responsible AI
Help readers better understand that AI is much more common and accessible than one might think
Illustrate why we need to manage AI responsibly to mitigate unintended societal outcomes.
Finally, we want to define the business opportunity we believe Responsible AI can deliver
We believe that, by following Responsible AI principles, you can both grow your business and mitigate unintended outcomes. You don't need to choose. It's a win-win for all of us!
While this topic could seem intimidating, we don’t believe it needs to be. As we make our journey down this path, we continually find ourselves more excited about this opportunity ahead. We hope you do as well!
Before talking about “Responsibility”, let’s define AI
It’s a lot more than virtual assistants, facial recognition, and futuristic robots
To set the stage, we believe in a very broad and accessible definition of Artificial Intelligence.
While many may think of Artificial Intelligence as something that is very futuristic (i.e., those robots), it is actually much simpler.
Artificial Intelligence is any system that:
Takes and interprets inputs
Generalizes its knowledge
Applies its knowledge to new applications
A simple version might be that any time a computer makes a decision, it is AI
A couple key items to note in this definition. First, AI is a “system” – as such, it isn’t only some programming code and technology, it also includes things like objectives and governance. Second, it doesn’t require automation and, thus, allows for (and we could argue requires) human interaction and oversight. As we move forward, these items are key in understanding the broad applications and implications of AI.
Great, we have a definition. What does that mean then?
Let’s bring this to life with a few examples.
Leaving robots and things like virtual assistants behind for a moment, let’s look at the below sample list of business decisions that we make all the time.
Which of these business decisions include AI component(s)?
Whom to market to and who not to?
Whom to offer credit to?
Whom to recruit for jobs?
Where to place your next retail location?
Defining functionality of and how to launch a new product?
Of course, it is a trick question. In modern business, to answer each of these questions, systems, AI is used in some form. These systems used to make these decisions include components that take and interpret inputs, generalize knowledge, and apply the knowledge to new applications.
While, over time, we will cover many of these examples and more in detail, let’s start with the question of marketing. Modern marketing has many, many systems that:
Takes Input: Brings in data on who has responded (or not) to past marketing efforts
interprets Input: Analyzes this data for patterns
Generalizes Knowledge: Build models to predict future responses , and
Applies Knowledge to New Applications: Selects who to market to
You could apply a very similar view of the other examples. For credit, we look at who did not default in the past, for recruiting, we look at who we hired (and who performed well). For retail, we look at the characteristics of where we’ve been successful in the past. For new products, we gather data on where past products were successful and build new products for similar ‘like’ markets.
What, then, is Responsible AI – and why does it matter?
Understanding how “unintended outcomes” can happen at scale
Think about how we all make decisions every day that are informed with the best information we can find – and we look to make with the best intentions.
If you are like me, it is rare when the results of those decisions are exactly what I expected. For example I may have say something someone else may not interpret how I had envisioned because they have a different perspective than I do. These mismatches of expectations relate to cognitive biases we all have.
There has been much written and analyzed about cognitive bias – when someone has a preconceived notion of someone or something, based on information we have, perceive to have, or lack. We know these biases lead to unintended outcomes. Because of these cognitive biases, I (and most of us) look to be responsible in our decisions, but we are imperfect. I review what I’ve done before, try to learn, and look to improve.
Now, imagine an artificial intelligence system machine making millions to billions of decisions all the time. As this system was designed by humans who are imperfect and have cognitive biases, we know it will have unintended outcomes. However, because of the exponential number of decisions that the AI system is making, the impact of these unintended outcomes is astronomical.
So, Responsible AI, is then implementation of governance across the designing, building and managing of artificial intelligence to best mitigate unintended outcomes.
A famous historical example of this are some of the systems that determined credit worthiness. A responsible practice that was put in place were regulations that ensure that data points around ethnicity, for example, are not used to determine credit worthiness. While this is a great responsible practice, we know there is more work to do. It typically can be easily seen that even if these data points on ethnicities are not included, other data actually proxies them quite highly - which, essentially, leads us to the same unintended outcome.
So, what can we do to mitigate unintended outcomes?
This is where we started to realize this is actually achievable
The wonderful thing is that there is much knowledge about how to govern AI in a responsible manner. It does take some intention and effort, but, nicely, it layers very well onto current processes as a better lens and way to make decisions.
At Eagna, we've delved through this thought leadership and built out systems to help our clients implement Responsible AI. Through this work, we've been able to, we believe, simplify all that is out there into a small group of focus efforts to achieve Responsible AI.
As we tell more stories here, we will illustrate these focus areas . For our primer, the key tidbit is that Responsible AI is relatively easily achievable with:
Broader organizational participation
More focused planning
Commitment to some testing and improved data
A couple unique “data science” approaches, and
Some smarter reporting
But, don’t I have to sacrifice business results?
Good News…. You don’t!
The big question I often get is if we can have both – great business outcomes and great social outcomes. I believe the answer is a definitive, unequivocal YES!
An “a-ha” moment I had what that Responsible AI best practices actually align wonderfully with the approaches needed to expand into new audiences of your company. A key to this is that Responsible AI centers on not just keep doing what we’ve done in the past - and that we test into new areas. We need to actively seek out new data where we have blind spots and test into new audiences. That’s where the beauty lies… the more we employ Responsible AI best practices, the more we test into new audiences, the more we learn, and the more we can grow.
How do I get started?
I hope you are intrigued by this primer and it inspires you to do your own learning. A great way to start (as seems to be the case these days) is to educate yourself through some online research and reading. A lot of really great content is out there.
Of course, we’re happy to be a part of your journey as well.
Follow us on LinkedIn to keep up to date on our blog and what we are learning.
If you are even more intrigued, reach out to us to discuss how we can help.
We are looking for partners in our journey. Have stories of your own? We would love to work with you on including them here with our stories. Contact Us!
We’d love to understand where you are and discuss how our Discovery Series and Responsible AI Design solutions can help you grow your business!