Computer Vision News - September 2023

11 Computer Vision News depths. For instance, for deep neural networks, I would like to understand how they work and why they work, and in that way, make them also trustworthy and reliable. I don’t know how my washing machine works, but it works for me. It does the laundry. Why do you dedicate so much effort to something that, for whatever reason, works? Yeah, that’s a very good question, but your washing machine does not pose any threat to you, I imagine. However, if you have a self-driving car that doesn’t work properly and makes wrong turns, it could cause serious harm to humans, animals, and so on. In that sense, I think it’s a very different situation. I also don’t understand my washing machine. I mean, to a certain extent, I think I know how it works, but I don’t have any, let’s say, worry about it. But AI is also used, for instance, for reaching decisions like, is this person sent back to prison or not? I mean, really key decisions. In that sense, it needs to be absolutely reliable. We need to understand how it reaches decisions. Also, we need to understand how it relates to the training data, which is usually something not that much put into focus. The training data needs to be clean, and it needs to be fair, and so on. From the data science side, there needs to be a lot of research in this direction. In that sense, from my perspective, it’s absolutely key. Also, this EU AI Act correctly requires that we understand it in depth. There’s this right to explain, a right to explanation. If AI technology is approved, we need to understand how it works. The customer needs to know or has the right to get an explanation for this decision. Gitta, you have succeeded in scaring me a little bit that AI could find things that may not be exactly in the direction we want. Has this already happened, or are we not thereyet? If you think of self-driving cars, they were already involved in accidents, so it’s known that they sometimes brake strangely. I think if there is a car on the right-hand side at the “ … they said there are these great methods now, which often easily outperform the state of the art, but we can’t really understand them. They are like a miracle, and the mathematics is missing!” Gitta Kutyniok

RkJQdWJsaXNoZXIy NTc3NzU=