There is a new law GDPR in Europe that requires algorithms, to explain themselves when they make decisions that impact humans. On the one hand that's a good thing, clearly when these impacts are negative, if I need a loan and an algorithm decides for me that I cannot have that loan, then it's desirable to understand why I cannot have that loan. But we should also not fool ourselves, that this is actually always possible.
When these algorithms become more and more complex, it becomes harder and harder for these algorithms to explain precisely why they come to a decision, so I think it's very hard for Alpha Go to explain to us why they made this famous move which won a game. We can come up with some approximation of an explanation, but it will not actually reflect the true complexity of the algorithm as it runs. At some point, there will also have to be a certain trust in the system. For instance, if a system that diagnosis you when you get to the doctor and sort of recommends a treatment, you can choose a simple algorithm that can perfectly explain itself or you can choose very complicated algorithm that performs a lot better and makes it more likely that you will recover but that cannot explain itself very well.
Personally, I would choose the more complex algorithm and simply trust it. But it depends also a lot on the application. In terms of the loan, I might actually want and demand an explanation for why it was rejected. But I think it will depend a lot on the particular application that we're looking at, and maybe we will just have to, sort of, be satisfied with simple explanations that do not actually reflected the full complexity of the decision.
It's very hard to look in a glass bowl, but I would think that there's 2 developments which are particularly exciting in terms of Artificial Intelligence, in the coming 5 years. The first one is probably Augmented Reality, so I think we'll see glasses, wearables which will project information on top of the real world, we can project all sorts of interesting things on top of the real world which can be very powerful device to have. The other one is probably interactive jet bots, so in the future will probably be possible, again in maybe is somewhat limited domain, to have a conversation with an AI system that sounds almost indistinguishable from a human conversation. That has also of course many applications.
In terms of skills that we need to learn, in schools to prepare for a modern future AI: I find it somewhat strange that even at this young age we don't learn about coding and software and more about technical skills. Currently we learn a lot of natural languages in school which is great, but we don't learn coding languages, which I think is strange, especially given the fact that digital, technical, innovation affects our daily lives so much. We spend a huge amount of time on our cell phone, and we don't understand how a cell phone works, and we don't understand what happens to the data trail that we leave on that cell phone. That's on the technical side. I think it's also important to train people less and less in very specific jobs, because the labor market is probably going to shift and change, and more about learning to learn, being creative and changing skills in an environment where that is necessary.