ai.bythebay.io: Jur van den Berg Interview
Recording: ai.bythebay.io: Jur van den Berg Interview
you [Music] you yeah it is a very difficult question AI is a very broad term it has been used for for everything from a star search to just play in computers and now with this rebranded as as having to do with deep learning mostly so yeah I have personally always had the opinion that AI is not a good term because it is it sounds cool it sounds fancy but that is also the risk of the term because it creates inherently a promise that it may not always fulfill and we have had the AI winter as people know like where the promises were not being able to be fulfilled and that's too bad so if people would have called it different then maybe it wouldn't desert generate such a hype and I think ultimately these types are harmful so I am NOT a big fan of the term AI yeah it is actually it has been extremely successful with the with the event of deep learning where we really harnessed all the computing power we have to really train deep and large networks and we have seen results that are just remarkable where we can train networks to detect cars and images for instance at a level that is that it's equal to humans and that's just unbelievable no classical fusion system has ever been able to perform that well and now we see this new technology that suddenly has this enormous promise in this enormous application domains old not only in imaging but also offering or inefficient but also in sound processing and you name it it's everywhere now it's it's again a hype but this one has a lot of potential and it's from and it's producing results as we speak and that is really a really really a technology of the future like the research is very much in baby stages at this point the researchers that work on it don't really know what they're working with they're still trying to figure out why does it work so well and that makes it fascinating to see it so working so well in practice and of course that's always the first stage of research hey we have some new technique and it works fantastically and now of course over time people start wondering like why doesn't work so well what is the key what actually happens here that makes it work so well and at this point we as researchers or as industry leading scientists do not understand that law and that that is super exciting and that's of course where where we need to make progress to have a fundamental understanding of what's going on why does this work so well and if we ask that we can also only make it better from there yeah so I work really on motion planning and motion planning for in particular self-driving vehicles has to do with planning its steering and its application of the brakes in the gas and I do not directly use machine learning ultimately I use different techniques but we can use machine learning in the stages leading up to it so it's not a fool deep learning or machine learning system it's a system based on classical algorithm that is aided by machine learning various various cases and that is where the perfect speeds sweet spot is because of course the learning has been so successful but we shouldn't make the mistake to throw away because we can techniques that are proven to be very successful in the past but we have to make sure that we can use them together and to bring them to bring the most out of both of them and that's what I do and what keeps me up at night is now I sleep pretty well actually generally yeah [Music] you [Music] you