ai.bythebay.io: Francois Chollet Interview

[Music] so area in general is the automation of so area in general is the automation of thought which of course is it's very thought which of course is it's very vague and uncompress is many things for vague and uncompress is many things for instance you could say that just instance you could say that just software engineering is a are in some software engineering is a are in some way because you are automating thought way because you are automating thought processes when developing programs you processes when developing programs you automatically the automation of thought automatically the automation of thought will lead to the automation of the human will lead to the automation of the human mind right with artificial recreational mind right with artificial recreational human mind but we are still very very human mind but we are still very very far from this and I would say that what far from this and I would say that what we're currently able to do are just we're currently able to do are just essentially baby steps on the way there where do you think AI is successful where do you think AI is successful right now and where do you see it right now and where do you see it improving in the future improving in the future so that's an easy question clearly where so that's an easy question clearly where we are already successful right now is we are already successful right now is mission perception so things like mission perception so things like computer vision being able to recognize computer vision being able to recognize objects in image things like speech objects in image things like speech communication in so in general mission communication in so in general mission perceptions of processing of images and perceptions of processing of images and sound and the reason we already sound and the reason we already successful at it is because this is all successful at it is because this is all just pattern matching and with steep just pattern matching and with steep learning with right so depending we learning with right so depending we found a really powerful and an easy way found a really powerful and an easy way to do pattern matching even you know to do pattern matching even you know super human data machine however very super human data machine however very machine has very stock limitations and I machine has very stock limitations and I don't think intelligence can be reduced don't think intelligence can be reduced to pattern matching so the next stage to pattern matching so the next stage next steps for AI will be to go beyond next steps for AI will be to go beyond me me to attack all things that abstraction to attack all things that abstraction and unreasoning but we reform always on and unreasoning but we reform always on you what is your you've kind of touched you what is your you've kind of touched on this but what is your key focus with on this but what is your key focus with AI and machine learning what's something AI and machine learning what's something that keeps you up at night so my key that keeps you up at night so my key focus would be actually I don't I focus would be actually I don't I couldn't say I have a key focus and I'm couldn't say I have a key focus and I'm very much unfocused I work on many very much unfocused I work on many things I do a bit of everything I've things I do a bit of everything I've done quite a bit of the traditional I done quite a bit of the traditional I would say deep learning data matching would say deep learning data matching approach you know trying to tune carbon approach you know trying to tune carbon and architectures figure new designs for and architectures figure new designs for current architectures I've also worked current architectures I've also worked quite a bit on trying to apply deep quite a bit on trying to apply deep learning to formal reasoning which is I learning to formal reasoning which is I think a very interesting and also very think a very interesting and also very new field I only have like one specific new field I only have like one specific focus one specific area of interest but focus one specific area of interest but I do believe that the right program to I do believe that the right program to be working on is the form of abstract be working on is the form of abstract modeling abstract reasoning so the modeling abstract reasoning so the manipulation and development of abstract manipulation and development of abstract models and which cause you know far models and which cause you know far beyond pattern matching and just far beyond pattern matching and just far beyond our current approaches I was he beyond our current approaches I was he planning a new training and I we still planning a new training and I we still don't know you know what's the right way don't know you know what's the right way to approach these problems so even if to approach these problems so even if you identify which problems you should you identify which problems you should be working on that's not telling you you be working on that's not telling you you know which approaches you should be know which approaches you should be using so now I don't really a specific using so now I don't really a specific focus I'm just keeping my eyes open for focus I'm just keeping my eyes open for anything anything you you [Music]