Artificial Intelligence - Q&A #9

Key Takeaways

what's going on everybody Welcome to Q&A number nine this Q&A is all about AI this is the last Q&A of 2016 last video on my channel for 2016 and it's only appropriate to talk about AI I think AI is going to be big deal in 2017 not necessarily because of uh any advancements in AI specifically but more so the impact of the just all the way up to the current advancements in AI right now so a lot of questions on AI generally stem initially at least from the question of is AI a threat to to people in general okay so first this kind of like teared down response to what we should think about Ai and what the threats of AI are really we have to kind of break it down into almost like a tree argument so first of all the question is AI a threat you obviously you can just say yes or no to this question if you say no um the problem is we really just don't know with AI so I think it's really hard to hold the stance that no AI could not be a threat we we just don't understand AI we don't know what's coming with AI so you pretty much have to be in the camp of yes now most a lot of people are in yes because um as humans We Fear what we don't know but we also it's rightfully so we we really just don't know what AI will bring so AI in the current sense I mean there's nothing to fear um but the the fear is when we get to something like artificial general intelligence which is like intelligence that's on par with a human and then you've got artificial super intelligence or ASI uh and this is something that's beyond our intelligence so these two things are kind of I think the you've got AI right now which is like weak Ai and you've got strong AI which is like really truly intelligent AI because right now kind of what we have is just um brute forcing and fitting to problems that are you know infinitely dimensional but all we're doing is just fitting to infinitely dimensional problems really with AI today so so let's say your answer to the AI thread is no well you're done here I guess but you're also U probably just stupid uh is AI a threat if your answer is yes then we have to start thinking about okay well what can we do about this threat of AI so um a lot of people don't know we we don't really know what we can do about it but the naive approach is say um is there some way we could restrict AI or uh restrict the development of AI well especially like maybe AGI we could come up with some sort of system that would restrict the AGI artificial general intelligence so intelligence that's on par with a human we could we could probably devise a system that would restrain an AGI but could we devise a system that would restrain an a SI and artificial super intelligence uh I think it takes a little bit of huis to think that we could somehow restrict ASI so if and when ASI comes around really that's just going to be kind of a crapshoot like who knows uh and to think that somehow we could restrict ASI I I think is probably a mistake or or at least there there's just no way it could happen um so then you've got Place restrictions on AI use as in people using AI companies using AI governments using AI that kind of stuff well the problem with this is AI is so powerful um that everybody wants to do it everybody wants to use AI so if say the US government says you guys you can't use AI anymore well the other countries are getting to use AI so AI is still going to advance and because it's so powerful it it would just be silly to try to restrict AI also AI is an idea it's it's just math it's it's a lot like encryption where countries will say XYZ encryption is bad um so asymmetrical encryption for example um that's that's banned you can't use it well that's just law right law doesn't stop you from using it so and if anything all it does is stop you know people who are following who are willing to follow the law stops them from using it but it doesn't stop a criminal from using it so the same thing is true with AI would it is it really the greatest idea to be to tell to basically have all the law-abiding citizens the actual people who are part of society tell them no you cannot use Ai and then what's left is just the the outliers the criminals who would be working with AI the people that don't care about the law so I think that's a bad idea too so restricting the use of AI um for people and companies and governments is also probably mistake especially when it comes to something like artificial super intelligence where it's like a win or take all scenario the first person to reach artificial superintelligence is the Victor uh it's very hard to conceive of a way that one person one entity could reach artificial super intelligence and if that entity didn't want anyone else to reach it uh it's hard to see how anybody could avoid that situation so so so again coming back to to the original question is AI a threat no well you're stupid yes okay well what could we do about it we could we restrict it some somehow either restrict its use or restrict um restrict the actual software itself somehow so there are ways you can fool AI um but for the most part you know probably not especially ASI artificial super intelligence we are not super intelligent so how we would think about it's almost like um you know consider the difference between like an ant and you okay can you think of ways that ants might devise a way to to thwart something you want to do you know no it's just you're you're just on a different level of intelligence so to think that we could somehow uh restrict an artificial super intelligence I think is just silly and then also I think it's just a bad idea to restrict it via law or something like that regulation now the other option you have is pretty much the opposite of restriction and that's to just decentralize it because really while AI alone is a threat I don't know how we can protect ourselves from that threat other than to um to to make sure that not just one human or one small group or entity or uh country organization whatever has that technology so the best thing we can do is decentralize that technology through I think open software like tensor float Theo that kind stuff um open AI is a great one just just the decentralization of just knowledge Concepts and lowering the barriers to entry to to just working with this stuff so um I think that's that's it cuz you know AI is is really neither good nor evil inherently um and it's it's not even self-centered it's none of those things but humans by just Evolution we are self-centered we want to protect ourselves first and all this so one you know AI alone is is likely not a threat um or it's I guess I can't it's just neither good nor evil but it's probably a threat but but AI in the hands of a few is almost certainly a threat in my opinion so um just some more examples here uh like I thought I think it was Sam Harris that made the example of like a comparing um an ant to a human and where you know if you know as an ant walks along the pavement or something you know you you probably like you'll take you'll take steps to avoid stepping on the ant but if the ant is in your house you're probably going to just exterminate it like that's your first choice is like let's just exterminate it so in terms of like an artificial superintelligence it's probably not just going to one day just be like let's just kill humans for fun it's probably not going to do that but if we're getting in the way of some sort of major objective of the artificial super intelligence then it's it's conceivable that it would decide let's just exterminate the humans but like at least for me if ants were in my house and if I could communicate with those ants and tell them hey you need to you need to go or I'm going to kill you I would do that like I would I would try to to to give them like an eviction notice but I can't because they can't communicate with me but but on the other hand like with uh larger animals like dogs for example we can't communicate perfectly with dogs but we can teach dogs we can train dogs we can tell dogs and show dogs what we like and dislike and I think that with like an artificial super intelligence that would possibly be possible so it doesn't mean that an artificial super intelligence would deem it worthy to save the humans but um there you go so at least for me I think that's kind of how um how the argument kind of breaks down with um with AI and whether or not AI is a threat and then there's this kind of Al this this question that comes next which is how far away is AI I think this one's hard I think you know there's some some Believers in the whole Singularity like Ray kurile um I just I can't get behind the idea of a singularity fully Because unless it was like a mathematical breakthrough it's possible like with AI some sort of mathematical breakthrough happens or maybe a physical breakthrough with like components and Hardware which would probably still be a mathematical breakthrough maybe with that we could see something like a singularity but I just don't think that's what's going to happen I think that you know let's because the argument is like basically if we could make an artificial general intelligence something as smart as a human then because it's software it can just be infinitely copied right well that's not really true like let's say you come up with a great neural network that detects cats in video what what did you need for that well you needed it a big GP you so can you just infinitely scale your uh your cat detector no because you need the hardware and you need the power you need all these things to also be in Balance so the only way I think a singularity could occur cuz like right now today an artificial general intelligence was created just by stacking a bunch of gpus together well it wouldn't be it in theory it could be infinitely scalable but in practice it would not be because we would run out of power both in hardware and actual power to power these things so could you make you know seven billion of these agis no you you wouldn't be able to scale it that fast um also the argument because the because the argument goes so even though I'm going to have to just raise my hand and disagree with the whole AGI infinite scaling as soon as AGI is created if AGI the well the argument is that if that with AGI you could scale it in initely it would be human intelligence so it's the equivalent of all the humans possible and if if if if it's possible that we would ever create an ASI then then an infinitely scalable AGI could create the ASI okay sure I guess maybe but that that argument is also flawed just simply because stacking of intelligence does not necessarily mean an increase of intelligence right so the stacking of intelligence say say you take one dog you put a dog in a room and then you take five dogs and put five dogs in a room is there really that much of a difference in intelligence I mean some social constructs might actually form and there might be a slight change in Dynamics but overall collectively did will the dogs reach a new Peak intelligence probably not what if we take a 100,000 dogs put a 100,000 dogs in a room are we going to get like some new evolved doggo I don't think so right that just seems crazy but for some reason we think that who if we took AGI and we took a billion agis or 10 billion or 100 billion somewhere along those lines it'll produce an ASI maybe maybe that will but maybe not like we really just that we don't know so I think that's kind of a bad argument and so but that's that's what the whole Singularity pretty much hinges on is when we reach a gii it'll be infinitely duplicatable dupli anyway it'll be infinitely um scaled and Bam like just like that overnight will have ASI no we just won't unless it was like a mathematical breakthrough that on current on a current GPU you could like have like a billion agis or something like that then sure you might see Singularity maybe but but still even even today the intelligence that we have these Neal networks they don't do anything they just do kind of what we tell them they don't really actually they don't think for themselves at all uh and then they can they can do tasks like with reinforcement learning or something but but it's they're not going to come up with new and novel things without being guided to to try to come up with new and novel things so getting back to the question of how far away are we really it's unclear right it could be tomorrow or it could be like in a hundred years or could be never right ASI might not actually be achievable at all we really just don't know but if it is achievable we'll almost certainly reach it because again going back to the whole should we restrict AI it just won't work because AI progress is just simply going to happen we are just going to Simply continue improving on gpus it may not always be at the rate of Mo's law the whole double every two years that's probably not going to continue forever but maybe it will or maybe it'll increase so no matter what happens it's going to keep continuing and no matter how scared people are or whatever it doesn't matter steps will continue to be taken so instead what we have to do is think about what can we do to ease the blow We Can't Stop ASI from deciding that humans are bad for the planet Earth and to just exterminate us that really is never like you're just not going to outsmart ASI so what can you do I think again as I said before I think the whole decentralization make sure it's not just one group that has artificial super intelligence right what stops a bad guy with ASI a good guy with ASI anyway let the fight begin no um so there's that but also even now there's M even larger threats that no one's talking about or at least I don't see people talking about everyone's like arguing about how do we stop an ASI like that's a that's a deep philosophical question and anybody who claims they have the answer that is just BS you you can't know right because we have general intelligence so using general intelligence to stop super intelligence not going to happen so on the chances that this goes smoothly we instead have to think about what's what kind of transitions are we going to go through and already today we have serious problems that are coming so the number one uh I think it's either one of the largest or it's actually the largest especially probably in the last year it's the largest job in the United States is commercial driver say truck driver taxis Uber driver lift whatever all these things commercial driver but what's happening and what happened in 2016 and what's been leading up to 2016 is just uh self-driving cars and Uber Auto all these companies are trying Google of course um are making self-driving cars I suppose Tesla Apple's trying to get in the mix self-driving cars are coming and Commercial delivery has already taken place I think it was Otto that did it it's over it's that battle's lost so all these people that were doing commercial truck drivers have to go somewhere else now I used to kind of always think like big deal this this this sort of transition has always occurred throughout history where like one group of of workers uh just simply transitioned to another career but now it's getting hard because maybe you can but I can't think of a single career that cannot be replaced by Robotics and AI I can't think of one right all of them could be replaced now there might be a demand for uh for human-made things right man-made objects or man-made art or stuff like that um kind of like made in the USA products versus if you're in the United States um that's a big deal to some people but not everyone not everyone cares about that and if people don't have money they don't care uh so I think like what the bigger question is um you know not worrying about how do we restrict something we can't fundamentally restrict instead it's a question of if this all goes well what are we going to do with ourselves because all the jobs can be replaced every job could be replaced and again you could say well the government could restrict companies from automating um cashier jobs or something like that so at least at stores people could work but it would almost have to be the case that all countries did this or something okay it's it's really that's not an option I don't think it's not a long-term solution to this problem and so I think the the real challenge is figuring out with company like say uto for example as Auto Cuts all of these jobs for truck drivers is auto doing anything I'm not saying they have to I'm not saying it's their their job that's not the argument I'm making I'm just simply asking a question are they going are they also simultaneously working to try to solve the problem of these displaced truck drivers because sure they could go uh try to be a cashier at Walmart well Walmart already has automated checkout lines uh could they stock shelves at Walmart well Walmart well Walmart doesn't do it yet but Amazon's already stocking shelves automatically and automatic shelf stocking is is that like that's old technology right that's coming uh it's going to be cheaper and as a company is it fair to restrict a company and like say you must hire humans right I'm not sure that's fair either uh and like I said I'm not sure that auto should be in charge of solving this problem or other companies making self-driving cars or any other automation technology but if they aren't going to do it who's going to do it and if no one does it then we will end up with a situation where like five companies rule the entire Globe basically of services and jobs and make all of the money have all of the AI and that's something you want to avoid and how you avoid that problem that to me is the main problem the main problem is not how do we stop AI from killing us it's not going to be the AI I just don't think it's going to be AI that kills us or restricts us or enslaves US is it's going to be humans right the humans are the ones that are evil and self-centered and uh lust for power and all this kind of stuff it's humans that do that so an ASI may not even be possible but what is possible is the automation of every job on this planet that's what's possible and that's what's happening already it's already happening and the question is how are we going to handle that how are we going to handle that transition because I don't know I don't have the answer to that one that's the good question