Elon Musk and other AI experts are calling for a pause in the development of powerful ai ai tools but is this a case of trying to put the genie back into the bottle we'll explore this next up on today in Tech [Music] hi everyone welcome back to the show I am Keith Shaw an open letter coordinated by the future of Life Institute which lists Elon Musk as an external advisor calls for a six-month moratorium on the development of generative AI tools like chat GPT in order to give the industry time to set safety standards for AI design and also prevent potential harms of the riskiest AI Technologies joining me on the show to discuss this idea is Jason Mars PhD he's the associate professor of computer science and engineering at the University of Michigan and co-director of um's clarity lab yeah he was on our show uh about a month ago one of my favorite guests welcome back to the show Jason absolute pleasure to be here uh awesome topic very very interesting yeah so what were your thoughts when you so you were telling me before the show that that you received this invite to sign the letter uh so just kind of give me your thoughts on you know do you think this is a good idea what did you think when you received the the invitation to sort of sign it yeah no and so yeah so I re received the invitation um a couple days before it hit news in a big way when you won and uh Steve Wozniak also uh signed on uh yeah I I didn't sign it um so I think uh it's it's there's a lot of these interesting and and some somewhat valid uh concerns around how this kind of technology is going to shape reality yeah however there's there's a massive I I believe it's part of an overreaction I believe it it it it adds to the fear-mongering it adds to um and it's also uh somewhat misplaced there's a lot of Downstream implications on making any kind of attempt that can only be implemented through a congressional uh legal means uh as policy for the nation is the only practical way you can implement this and the downstream effects are way more like incredibly disruptive uh in and of it themselves and also ineffective right you you can't you in the realm of software you it's where basically all of the Creations come from Human Minds uh you can't really put a moratorium on the things people people invest their time into and work on and develop with access to technology and mass it'd be impossible to Define what a large language model is in respect to the kinds of things that are dangerous um and it also put a pause on the development of a technology that many folks globally are developing right which which would would stunt essentially we're already at risk of um losing uh some of uh America losing its Monopoly or it's you know it's uh Advantage when it comes to technological development in the world economy yeah this would this would stunt our ability to be relevant we're already losing the social media game uh with Tick Tock being one of the largest Platforms in America right uh surpassing all of the copycats that are trying to replicate It in America yeah um so you can imagine what an opportunity this would create for other nations to to advance and overtake the technology especially given that it's it's public how it was built yeah if this was like a a us-based moratorium then all of a sudden like you know I know I'm pretty sure China is not going to sign that and you know other nations are developing AIS are not going to sort of you know jump into this what what are their concerns about the risks of of sort of this this race because like my first impression was like it it felt like with with Elon Musk and Steve Wozniak it was just like whoa whoa guys like you know we weren't involved in this so we want we want to have a it just felt like hey look guys the race already started like you can't just pause a race in the middle and you just have to keep going or this trainers catch up exactly this train has left the station right this is not one you you know there's two kinds of ways you can think about these problems um so there are potential risks so let me first state there are potential risks and then I'm going to talk about the the approaches that I subscribe to on how to reason about what to do given those risks so the risks are real so uh right now we live in a world where people are consuming the information they would like to consume and the more interesting and Boombastic that information is there's almost a propensity to believe it right so okay now we have a step function in our ability to fabricate things that look real these models are trained to mimic real based on what it has been trained uh as real and so we're seeing this in many modalities I recently saw a video where someone was using Obama's voice uh to say what they wanted to say and it was in real time rendering a voice that was convincingly Obama could you imagine someone uh mimicking Biden's voice and calling upon all Americans to do something uh hateful and atrocious right right and you know so so you have an opportunity for misinformation uh fabrication and and and to to mislead the the public but this isn't novel we've had Photoshop for a very long time right right there wasn't a moratorium on Photoshop and we could fabricate photos for a long time we can do it a little bit more convincingly and across more modalities now uh and people are likely to misuse it uh and we have to educate the public now to be even more thoughtful and suspicious of what they see uh and there's an education Journey there so that's that's a real risk right yeah there's also more economic risks with um yeah go ahead well you know there's always been bad actors from whatever Technologies come out with like like you said we didn't ban Photoshop because some people were putting you know somebody's head on someone else's body um exactly obviously you know you don't you don't halt AI just be you know just because someone's using it to fake uh Obama or Biden or Donald Trump or whatever like that they're doing out there it's it's more about educating the the users of such stuff uh to do that so so is that their main is that their main reason and it's not because again there's a quote here that I wanted to read you um and this is from one of the guys uh that I think signed the uh no he's one of the organizers of the letter and this is the quote quote it is unfortunate to frame this as an arms race it is more of a suicide race it doesn't matter who is going to get there first it just means that Humanity as a whole could lose control of its own destiny I mean when you talk like that then you're talking like Taylor robots Terminators and you know we're doomed because until you know AI is going to get smarter and it's like whoa like that's not what this is doing right exactly exactly maybe I'm just naive and maybe this is something to do it it does feel a little bit like the sky is falling from from from this group yeah yeah I mean and and this is what gets clicks and this is what this is what gets uh attention right um especially when it's coming from folks uh like Elon Musk and but but the the the the real issue is it's absolutely not that's absolutely not the case we are going to have to adapt so there's two approaches there's prevention where we stop progress right well let's stop it let's figure out what to do uh and then we'll restart we'll unpause right progress but but really the attention and the energy should go into adaptation uh and how do we now live in this new world how do we anticipate how things will change and then stay ahead of the implications in so far that some might be dangerous right I I think there's other implications that are more important than uh we have a misuse of information uh yeah there's other implications that are much more foundational there's a lot of like for instance there's a lot of specialized skills in the world right we've got journalists that win Pulitzer prizes for example for how they exert their stories uh that kind of thing needs to be re-understood because the barrier to entry to writing a phenomenal story is going to be reduced because now more people are going to be able to access the help of some of these systems right to write incredibly brilliant stories so that actually applies on an economic scale now the the things where the the economy of it the the wages that certain employees get being high for their specialized skill now more people will be able to do that thing right and so that might create differences in the demand Supply those are the kinds of interesting questions I would want leadership to be pondering and thinking about how do we stay relevant and and understand five years down the line how this is going to change the economy and so forth but this knee-jerk reaction kind of you know Sensational yeah uh uh you know standpoint is yeah not useful what uh what concerns you about AI at the moment I mean obviously you your your your Enthusiast of the of this idea and a lot of these tools that come out you you mentioned on the last show that you use it all the time in your programming like when you're coding and and you know it helps you become more productive in in in the work that you're doing and I think and I think I think you said you're using it in your classes as well you're using it to to teach um absolutely so and and your students are allowed to use it as long as they sort of reference it right yeah yeah so so what are your concerns about some of these tools at this point is it is it the you know because another big concern I've heard is about explainability that there's a lot of people that still don't understand why the AI sort of spit out what it did like it didn't it the explainability has been been explained to me for uh three to five years now and I still don't understand it so if I still don't understand it it's probably an issue right yeah no absolutely well and the thing is my concerns really are manifested in what we're seeing happening currently uh my concerns is most people won't understand it uh Congress people don't know what a website is is YouTube a website uh you know like folks don't understand what's going on and so bad choices are might happen Sensational things like this may cause a congressional hearing Elon Musk might be invited to say something votes get passed and then yeah the the dynamic Stops Another really important thing to understand about AI that a lot of folks don't understand um is and this might sound like a vote towards a moratorium which still I I think is crazy but when we develop as a community of scientists when we in the art and the discipline of innovating when it comes to AI models deep learning models that kind of science is becoming more of a Discovery Science understanding nature than it is an engineering science where we're trying to build something to solve a goal a lot of times when we build these models and we train them in various ways we actually don't know what that model is going to be able to do we'll have some thoughts as to why this mode of training and why this style of neural network should have the capacity to do something but once we're done training it we've been blown away and it manifests itself in the papers in the papers themselves they'll say often things like we never expected the model to be able to do this but look at what it did right and so and so every every GPT three the you know what the paper does published doesn't talk about it being able to write code but then we we discover after the fact this thing can actually write code yeah chat TPT was primarily motivated to create a PC gpt3 unbiased doesn't say offensive things yeah but it actually produced a GPT a a conversational AI that did things Way Beyond what I anticipate the the creators uh develop so it's a very Discovery Science and you know who knows what will discover these models being able to do once we take Tinker with them yeah do you understand that that's how that that is what could freak out some people um or you know you know if if you told me for example that you engineered and designed a a train and then you wrote something about it says yeah we were really surprised when that train started to fly you're like exactly exactly so maybe we should maybe we should pump the brakes a little bit Yeah but or is it something that you know maybe it's something that can't be understood or because when I I try to read some of these papers and it's like well it goes right over my head it's just like yeah no yeah I'm hoping that there are smarter people than me that that are reviewing these things and going like when you read that where you were like wait what you were surprised that it did this or yeah yeah I'm also it a lot of it is actually breathtaking even to the researchers as to as to what these Technologies are are able to do uh when trained the the fundamental difference is when it comes to the practicality from a societal perspective of taking up an approach of prevention or slowing down of this kind of progress It's fundamentally impractical and it's impractical in ways where it would be more practical for let's say uh nuclear uh engineering right like when it comes to oh well we definitely have to put a moratorium on the the production of plutonium uh and splitting it because you know there's a difference in the digital realm where everyone has the tools to if everyone could dig in their backyard and the you know access plutonium the practicality of putting a moratorium on the whole world doing that yeah is it doesn't work right with in software we're in that realm where people will do it underground uh and then it'll be even more dangerous you know it's it's almost like people can do it in secret and then produce false information much more effectively if it's done in secret China and Russia uh can produce fake media that's epically more real yeah and if you have a stunted America that doesn't even understand this technology because we have a moratorium on it we're more at risk so so if there's a practical element to that and and this is where prevention versus adaptation is an interesting debate I just had a very deep debate with a climate scientist one of the best in the world yeah uh he was over for dinner and we were talking about climate change and he's like oh absolutely this is a real thing it's it's it's absolutely produced you know by humans uh it's crazy but he thinks it's it's it's irrational to Think We're Not Gonna raise the degree 1.5 degrees two degrees it's like that's going to happen everyone is focused on prevention of that happening where we should understand how we adapt societies to live in a world that's two degrees hotter right right and so there's and so we don't hear in public policy an adaptation Focus right because public policy loves and now we all have to pause the world and then and and that's that that's almost more um appealing to the I guess the psyche than what's cognitively harder which is almost engineering re-engineering the world assuming that these things are going to happen but that's the right approach right you know because if All Is Lost if we don't prevent something and we're not prepared we're way more at risk than if we use our ingenu annuity yeah to re-understand what the world should be like yeah and this this letter did sort of raise awareness of of sort of this thing it was a high profile you're like I said yeah this is going to generate clicks do you think that this then turns into something where uh you know a conference is held or some sort of like high level discussion and people people start talking at this thing but that's still not going to prevent like others from not attending and just doing what they want to do and you know now you've got big companies like Google and Microsoft involved and they're like well you know we're doing this because we want to stay in business and we want to you know help our customers get better results and get better answers you know it's almost like trying to to get people to understand a common standard for you know networking for example or right you know interoperability issues and that took years for people to sort of come together um are do you think that maybe there this will lead down to a road of a standard AI development or some sort of yeah this is what we yeah absolutely right so the the interesting thing about what's happened is the AI is here now and now we have to understand what to do yep now you mentioned you know conferences there's a lot of different viewpoints and and forecast as to how the world changes right do we have this centralized AI that I mean there's a lot of funding of companies and a lot of thought that the future is we're gonna have these centralized AI uh uh resources that are kind of regulated by the government and they're like a utility that's going to be consumed by Society right that's one thought process because it's really expensive to train they're huge uh and it's it's a lot like a power delivery to a grid right um so I don't think that's gonna happen that at all right that's a very thinking in the now all of the scientists the computer scientists in the world I won't say all let me say much of the energy of the global computer scientists of the world is understanding how we make these Technologies smaller more wieldable easier to train faster to train easier to get as smart with less power with less energy that's all of our Focus the first computer was two you know buildings big uh before we had a PC right and now we have phones in our pocket that very Society will understand how to take these Technologies and wield it so I don't think it's a practical uh solution to think that we'll have uh not this decentralized everyone democratize access to AI right but the the high level point though is what you would suggest is exactly the kind of energy we see to exert we should have special conferences we should have new kinds of think tanks about just answering the questions of how will society change right because deriving the solutions will then be a straight line in my opinion but it's making that prediction as to given what we know about how technology evolves given what we know about what we observe the new digital uh ecosystem being where everyone's getting their news from social media Etc when you put together these interdisciplinary uh uh fields of study and understanding to predict what the world what are the 10 changes to World 10 years from now yeah having the best minds think about those changes is the first step and then as we observe the living system of the natural development and once we understand those 10 things then we understand how to adapt the world uh to it I think the best luminaries shouldn't spend time writing these open letters to to excite dramatic fear but they should be organizing that conference Elon Musk uh and um the organization that slips my mind yeah the future of Life Institute yeah future of Life uh they should be organizing that conference now and getting people to sign up to attend and present yeah yeah this is going to change the world as opposed to drama I like what it's what you said about um sort of the it won't be a centralized AI I think I tend to think that maybe that would be better but then I start thinking well wait a minute if you know down the line everyone might have their own individual AI in their pocket and then they can then you know sort of again saving power saving saving uh resources and processing time but then I start thinking well then all of these little AIS are going to start fighting with each other and then we're going to have like a big AI warrant and then my mind goes off into science fiction again um yeah some some yeah a lot of things to think about so and again I think the education part of it is so important as well um getting people that might not understand what an AI is at the moment start start telling everybody that you know it's like this is this is what's out there this is what might happen this is what could happen um be alert about every picture that you see every video that you watch it might not be real uh you know it was created to either fool you again we've got April fools coming up like that's going to be like you know I can't wait to see what comes out tomorrow like so speaking of that totally speaking of that you know since we didn't you know the the whole news about the AI image fakery with uh Donald Trump you know allegedly getting arrested like those photos and then last week the the Pope I fell for the Pope I don't know if you did it I was like that's a cool so yeah and so you know I'm all I'm always uh aware of potential political images and political things because I know that there are people that are working on that so I'm ready like any image you send me of a politician I'm like all right I think that's fake but a pope wearing a cool jacket like if I was like oh yeah that's great you know that's so hilarious the first thing just last night I was chatting with my wife and uh I was like you know what you know what I'm shocked bye you know there's this Nashville shooting epic tragedy yeah crazy uh and there's there's a lot of interest in seeing what's in the manifesto right like we want to understand what caused this right and I was like you know Ling ja they're not releasing the manifesto okay just count down the the minutes and hours before someone leaks a fake Manifesto right generated by AI right and that could also cause more issues than if they they release it right right like dude you're absolutely right like with it but you see that the the insight to even make that prediction is what we want every American and every person in the world to understand so that when it does happen they're like oh let me first think is this real or fake right right as opposed to taking it a real cloth but when I saw that those pictures that's exactly uh the the kind of thing that's causing um anxiety societal anxiety right uh it was a very compelling uh picture of the Pope uh you know Trump being arrested was incredibly timely yep everybody wanted it to happen and this is the danger when you really really want something to happen in the world that's when you're you're your critical thinking skills go away exactly and you want to believe it yeah and then you'll believe it and then you'll act on it right which is which is but you know the thing is I call it an overreaction because we've seen this story this movie has played many times over right where there's a novel new technology uh and you know there's this societal anxiety and it's never manifested as bad yeah as the fear is right um and you know Photoshop I think is a a great example because frankly a great Photoshop technician could have done that Pope picture before right there was you know uh uh uh GPT but but that that photoshop would have taken how long for for a skill technician to do it would have taken at least you know 20 20 minutes or more and and this thing can get no yeah it was it was still good it would have taken longer than 20 minutes I'm trying to figure out how long it would like I would never be able to do it just with Photoshop like that's why yeah yeah that's why I like this this idea because it does democratize the idea of like well you know what I really want to I Wanna for my poster or for whatever I want a picture of a pope wearing a cool jacket and just being able to sort of uh say it verbally and prompt it into a you know an application that gives me power to do things now again I'm not doing it for an evil purpose I just want a really cool image you know to sort of express the idea that I want to do as a journalist or content creator things like that um I don't think I would ever Veer into the the negative part of it because I'm I'm right I generally and most won't and most most won't but you know rather than spending you know a week learning how to do Photoshop and doing all that stuff it's like that's what excites me about Innovative Technologies but I'm a good person so I understand that there are people that aren't so good as well so um yeah all right so yeah have you experienced gpt4 since it came out I think you know I think it was right on the cusp the last time we talked yeah yeah so I so I so I've tinkered with it I think tinkered with it a bit yeah um and you know and I've read a good bit about it I think it's uh I think it's interesting right because they're including images which makes total sense yeah uh into these large language models so we're seeing what they would how we would describe it technically is now we're seeing these multimodal uh large language models being built in various ways um so I think that that's the big uh advancement of course it's it's better trained uh it's it's able to grab more understanding and insights uh around the world so it outperforms DPT 3.5 in the same realm of things that 3.5 does but the fascinating thing is that now it's able to capture these um high level understandings about the images of our world right so beyond just reading the text of the Internet it's also understanding and interpreting the images that you'll find on the payment internet too so it's a model that both can listen to what people are saying and see see uh what people are saying but uh there's a distinction between the GPT fours of the world and the chat gpts of the world they're kind of two yeah types of AI GPT 4 and 3.5 and those gpts are models that are trained to understand everything it can about the world and then it can be applied for many different tasks many different things you might want to wrap that model in you might want to wrap it in a product that solves a particular problem and the intelligence of what's in the world is now applied to your solution to that problem something like chat GPT is essentially a wrapper around something like gpt4 okay or gpd5 where it's it's actually the model conditioned to do one thing well in the case of chat GPT it's question and answering interactions right so I do ppt4 whole cloth wouldn't wouldn't produce a chat bot as as compelling as chat GPT out of the box right but it has in it an incredible amount of knowledge that would then if the same GPT chat GPT wrapper was applied to it would be Chad gpt2 right right right right right so yeah um so you're still excited about the you know the technology and where it's going yeah absolutely absolutely I think it's uh I think it's a phenomenal Direction I think the only problem that exists now is there's a bit of a monopoly of the of of that those kinds of Technologies realizing these advancements in open in open AI yeah uh it you know like when you look at like even the Googles and and even Microsoft's own other version of a large language model that was a thing they're still using that yeah yeah no no I mean and it's not chat GPT like as you get it on Open AI like those are attempts to catch up and to to participate because they see a market that they're losing but it's not democratized where where many many different um many many different institutions whether it be universities or or companies are actually all innovating at the same time so I actually want to see a world where more folks are innovating on these Technologies and we're not just waiting to see the next thing from open AI right right but yeah I'm absolutely excited about it well yeah so talking about the competitors you know we saw that Google came out with Bard and and you know did you have you tried that at all yet or yeah yeah so so so yeah so I've seen its output right so I haven't played with it directly myself and tinkered it but I've seen a b comparisons between what's produced yeah it's clear that it's clear that the meth so the methodology the the way that chat GPD was trained it's clear that that was not applied to Barn okay Bard is is is is much more the old school GPT 3 style of training which didn't condition it uh for phenomenal understanding of conversational Ai and a question and answering system yeah that's practically why these other systems aren't as good uh in my opinion yeah uh you know the way that chat GPT was trained is they they use a lot of humans to write what the model should say okay to improve it upon gpt3 right so you have your model that learned from the internet yeah but then you have to have a human in the loop and Coach the model right and train it and indeed they took the coaching and they were able to take that coaching and turn that into a model and then you had two models training each other it's called it's using a process called PPO but uh where it's reinforcement learning so they had a they had a trainer model teach a student model right like uh a child I'm just uh simplifying um a gbt3 and then that's how it got good the systems we're seeing like barred and the the Bings thing uh uh it's very clear from their outputs that that kind of reinforcement learning with humans in the loop yep was not applied to those models and so that's why they don't seem smart yeah they didn't get the extraction it did feel like from so I I got an invite and it was amazed of how quickly I was allowed in and uh because you know usually it takes a few weeks for me to get invited into anything and um I tried it and and it just you know and of course Google there's so many levels of warnings about like the output like they are trying to protect every angle and so yeah the the results that you get are just kind of like dull and boring and you know so I I equate like you know chat GPT and some of the stuff that was in the Bing was like your crazy uncle at Thanksgiving where Bart is more like your dad and you know so you know safe and reliable and you know doesn't want to make any waves whereas you know crazy uncle over there he's gonna just do whatever he wants and yeah um and that's what Microsoft was sort of pulling back too they're like we're gonna limit this so that we don't get a lot of the hallucinations and really freak out people that of where they could go so um that's my thoughts on that so um Jason one you know again thank you for joining us on the show today uh of course just I love talking about this stuff with you so we'll we'll definitely have you back whenever you know whenever all right awesome thanks man all right that's awesome here thank you yep that's all the time we've got for today's episode uh don't forget to like the video subscribe to our Channel and add any comments that you have below and join us every week for new episodes of today in Tech I'm Keith Shaw thanks for watching foreign
2023-04-09