Can you maybe pick from your experience some examples of where AI systems, AI engines have failed to be fair and maybe talk about technically why those problems occurred?

Yeah, I mean, I think there's a classic one, which we said we maybe wouldn't measure, but I have to mention it.

And it's the Amazon's hiring campaign in the 2018s, where the algorithm was trained on the existing profile of the workforce sitting within Amazon, which was predominantly male, which meant when they got CVs from people who didn't look like the people who were working there, then the bias just continued.

And that is the key point.

It's about this continuation of bias.

And if you don't really focus on the training data that's being used to inform the engine, that's being used to enable those patterns to be identified, which are at the heart of the algorithm, then you just end up essentially with a force magnifier because you've now just allowed this engine to do so much more.

If you think about that in the context of Many of us may have experienced a class in a college situation, and maybe there's 20 or there's 60 of you in a class, and you might have some bias or some error that somebody teaches you from the front.

What's special or different about AI? Why is it hard to make reliable AI systems?