AI Governance: What to Consider

Welcome. My name is Martin Kok. I'm principal Analyst at Koko. My topic for this podcast is AI governance. So AI for artificial intelligence. So all this stuff, which includes machine learning and some of the other art topics these days and governance, which is hot, you know, another sense. So to speak a little bit more boring, maybe, but also very important. And if you look for these terms, you will not find that much about it.

However, that combination is one I believe is super central AI governance means how do we apply appropriate governance for what we do with tools, technology that uses some sort of AI not to discuss what exactly is AI or not. And I also had some conversations in the past couple of months with auditors about this topics and all of them, I would say share the same opinion that this is in field, which will become more important, which is important factually already, and which is also about easy to solve. But there's one, I would say one essential aspect within AI governance.

And that is, we need to understand, we must understand why the AI we use draws which conclusion this is the really essential aspect in that. So don't blindly trust the AI, but understand why it does recommend something. Why it comes to serving conclusions may be decisions may be false, positive, false, negative, whatever. So how does it work, unfortunately, and this is one of the challenges we are facing most of the solutions, which make use of AI in some ways, still build on some sort of secret sauce. So there's the AI. It does some things some later on there's a result.

If we can't explain it, it's a challenge. So the better we can understand why AI draws, which conclusion the better it is. Clearly the challenges AI is not deterministic. It is not that I say, if this rule applies that rule, that rule, then this will be the result, but apparently it can indicate which factors flow into certain decisions. And the other thing we can do apparently is we can try to prove or to check what, what, so are these results really well.

And So if we take certain use cases, certain scenarios, certain test cases, and we know, and we know if that is the case, then that should be the solution. Then we can compare the validity of results provided by there's this sample about cats and docs. Very common thing. So how long does it machine learning take to understand what is a cat and what is a dog for humans? It's very easy.

So for us, it's relatively easy to say, okay, cat, dog, cat, dog, and compare with the results. AI delivers to understand whether the results are good, or apparently we also can do that with other decisions. So if it's about insurance contracts, loans, or other stuff, we can look at the hard facts and look at all the factors. So influencing a decision and look at okay, is the result really well or not? And if we do it with enough test cases, we can understand how good or bad the AI is working.

And the closer we come to a hundred percent, the better it is, but we need to figure out ways to under better understand how is this working? How are these conclusions strong? There are few vendors which provide some tools, which help us understanding how an I AI works. And I think this is a way which is super important.

We need to get better guidance also of the, all the providers of AI power technology that help us trust in AI, because then we have made a great step forward towards AI governance in having the understanding of how AI works, that it works correctly and that it delivers because at the end, if it works correctly, it delivers on what we expected to do. So it's latest time that we not only start looking at AI as a super cool stuff, which helps us solving all the problems of the world, which is which it doesn't by the way.

But that we also look at, if we use AI, which are the controls we need to implement, how does the governance look like? And that we start doing that, then AI will deliver even bigger benefit than can do without appropriate governance. Thank you for listening to.