Joe Asks John Carmack "How Close Are We to Artificial Intelligence?"

the Joe Rogan experience how far away do you think we are from artificial general intelligence so by nature I'm an optimist I am you know I tend to I to underestimate how long things take but on the other hand as a programmer I've usually been able to say well maybe I missed my estimate by 50% while everybody else blew it by a hundred percent or something I think that we will have we will potentially have unclear signs of AGI maybe as soon as a decade from now Wow now lots of people disagree the majority of scientists working on it think it's like oh it's gonna be at least a few decades and you still have a few holdouts that say oh it can't happen at all but I'm a strict materialist I think that you know our minds are just our body in action and there's no reason why we can't wind up simulating that in some way I am so I don't think the question of how far off there's a lot of numbers that you can play like the the brain has something like 85 billion neurons in it in the room they have something like 10,000 connections between it and you can multiply those out and compare them to what we have in computer memory and processing time and you can say that yeah within 10 years those curves should have crossed but I would even go so far as to say most of those are probably not completely necessary we know lots of biological systems like we understand the processing that goes on a lot in the visual side and we don't need nearly as many computer transistors as neurons that are used for processing some of those early layers so I suspect that even today some of the government supercomputers that the biggest the top 500 list that they have those are remarkably I probably useful for doing Intel it artificial intelligence work where for a long time for decades I thought that was sort of just I am national chest-thumping the top 500 computers because so many of them they relied on replacing what used to be the old big iron Cray vector supercomputers and they really weren't very easy to program most programs people want to use you can't run it on a supercomputer and just be a lot faster one of the shocking things most people don't really appreciate is the fastest way to do most single threaded applications is an overclocked gaming computer today you can go spend a million dollars and buy a computer that will do many tasks faster than what you can just run on a gaming computer and this is not at all the way things were for decades where for a long time you would go spend your millions of dollars on a Cray supercomputer and all of your code would run faster than anything you can get but it turned out that the the processors that you wind up using in high-end gaming systems are in many cases the fastest or in all cases at least close to as fast for certain serial applications so the only thing else you can do is pile lots more of them together and these big computers are football field-sized AI systems they're just racks and racks of GPUs and CPUs and nowadays for a long time I'd be like well what would you I would think how could I make a faster quake map build or something on one of those because we would sometimes have hours and hours spent processing this and I at one point we had a computer that was almost in the top 500 I'm at it software just for making our Maps I but I looked at a lot of these supercomputers and like how these are terrible not very useful for what we want i but now as I look at AI work and I think that well if you're just doing a whole bunch of these I kind of general matrix multiplies that computer right there it's probably pretty good so I would suspect that you could do something if we had the right algorithms the right training schedule and the right time to run through it that it's probably possible on some systems today and it'll just still take many years for the right algorithms to wind up being developed the right training regimens to be run and faster cheaper hardware and why did it make it more economical to run all the experiments because in so many cases the trick is not that the minimum requirements exists but that a thousand people have thrown themselves at the wall of a problem most of the bounced off and failed but eventually somebody gets through now what about quantum computing is that something that could potentially break the bottleneck that we have with Moore's Law so I don't I am NOT an expert on quantum computing and I I think that is many times I beat myself up about it where there's some simulators online where you can go and work on it and I should work through the exercises of doing the basic factoring algorithms on quantum computing but I know my read on it right now is that it's probably directly useful for most of the artificial intelligence tasks the big things that people worry about that are for things like breaking cryptography breaking the different hashes and encryption methods that it's possible that in many ways that's almost a terrible technology because it's a technology that doesn't solve so many of the problems that you'd like it to solve and it does solve one of the problems you kind of wish nobody was able to solve I am so I fight breaking all the encryption you know like if somebody winds up I with a quantum computer that they achieve quantum supremacy and it runs past all of our traditional computers and all of a sudden they can break all of the all of the secure socket layer stuff break everybody's signatures impersonate any public key sign stuff there's no upside to this that's all downside and all bad things they're gonna come from that while it's not gonna make your video encoding go any faster and it's probably not going to help artificial intelligence in many ways so I haven't found a whole lot to get me really excited about quantum I am computing it may just be that yeah and with all these cases why I beat myself up about not learning more about it because in most cases when presented with some capability there's some way to figure out how to apply it usefully to the things that you really want in fact I consider that almost the essence of engineering engineering is figuring out how to do what you want with what you've actually got and if somebody gives anytime somebody gives me new hardware usually I can figure out some useful way to do things that I want with it even if it's not immediately obvious and maybe quantum computing plays out that way but I am it is still definitely the domain of big labs with cryogenic cooling and all that stuff so hmm it just hasn't been at the top of my radar

Channel: JRE Clips

Views: 523,189

Keywords: Joe Rogan, JRE, Joe Rogan Experience, JRE Clips, PowerfulJRE, Joe Rogan Fan Page, Joe Rogan Podcast, podcast, MMA, Joe Rogan MMA Show, UFC, comedy, comedian, stand up, funny, clip, favorite, best of

Id: CYPQeN72LuM

Length: 6min 6sec (366 seconds)

Published: Thu Aug 29 2019