AI and Quantum Computing: Glimpsing the Near Future

hello everyone thanks for joining us for this conversation that actually continues a series of conversations that we are having with some of the world's leading thinkers in the development of of AI and Quantum Computing we're speaking with some whose real contribution is in VR and our goal through the series of conversations is to really give a sense a snapshot of where we are today and where things may be going in the future and our conversation today fits right in with that theme we are pleased to be speaking with Eric Schmidt who as many of you

know was Google's chief executive officer from 2001 to 2011 helping to advance a company from it's rather humble Origins as a Silicon Valley startup to one of the world's most valuable and influential companies in 2017 he founded Schmid Futures a philanthropic initiative that funds young people with a the promise to one day really profoundly change the world and in 2021 he founded the special competitive studies project which seeks to strengthen the United States long-term Ai and technological competitiveness issues that indeed will be at least part of our conversation today so with that let me welcome

Eric to our conversation hi Eric how are you doing I'm very good and Brian thank you for having me this show is followed by literally everyone I know so it's a big honor for me to be here well thank you so much for for joining us and and thanks for those kind words where are you uh joining us from I'm in Florida I'm a trustee of the Mayo Clinic and um so in the American for your foreign visitors um the American system is strange they're these large Hospital Systems which are essentially closed closed systems but

very well-run and Mayo is one of the best yeah absolutely one of the best for sure so you know before we jump into some of the discussion of what's going on with AI right now I just wanted to take a step back because your education your background I gather you got a Bachelor's in electrical engineering at Princeton and Master's and PhD in computer science at Berkeley so what was your early interest and did you imagine that business would always be part of the plan or is that something that happened along the way it happened along

the line um today I would have been called a boy nerd um maybe maybe that we maybe I was so nerdy I didn't hear that name at the time but when I was 11 or 12 I became very interested in science and for all the usual boy reasons I guess the stereotypes of the time building model rockets model trains that kind of stuff yeah my father had the bright idea of getting me access to a time sharing system when I was about 15 and that once that happened it was over um I liked Computing so

much and I liked computer science which didn't really exist at the time yeah um that I went to Princeton as a double e and Princeton was sufficiently Flex ible that they allowed me to not take any double e classes so I'm not very good at Hardware at all um unlike somebody like yourself but I took all the Advan I'm I'm awful with Hardware I somehow when I went to Harvard I got through without touching a piece of equipment I never took an experimental course I could only do the math and the theory only the math

yeah well so that's the software versus hardware and I think one of the things that that I would encourage people is first there's a difference between hardware and software people and you really want to be able to do both and I think that that so I was at Princeton you know 50 years ago it shows you how old I am and at the time there was no computer science now it's the number one major in the University as a as a new and important major and that's true in every us University MIT for example more

than 50% of the undergraduates are computer science Majors Caltech more than 50% of the undergraduates or computer science Majors so it's truly the case that we computer scientists have taken over the world and I will argue that this is not because of our Brilliance because we're not as smart as you in physics um and it's not because of other things we're doing it's because we understand scale and the computer science is is changing the world because we do things at a scale that is unimaginable so when we worked on the internet which I was very

proud to be one of them it I understood scale I just didn't do the math um if you have you know eight billion people of whom let's say five or six billion are going to have good access by the time we're done that's a big Market you show me a big a six billion person Market there's an awful lot of businesses a lot awful lot of Technologies and an awful lot of analytics and an awful lot of telemetry an awful lot of new discoveries just there yeah it's just sort of given by the technology and

so that's what drove you in that in that direction ultimately but you know the question that comes to mind for me is I imagine I mean I we don't know each other well you know we've crossed paths and various social Gatherings and various talks and things but I would imagine and correct me if I'm wrong that in the early days the mindset that that many had and perhaps you did as well was we are the Next Generation right we are gonna sort of tear down the man we're are going to innovate we're going to use

our Ingenuity to change everything and then you achieve that you kind of win right you become among you know the handful of the most influential companies on earth does that mean that you have now replaced the man with another the man or have you fundamentally changed things in in a way that at least would comport with what you had in mind so many decades ago I I think most of the people that I know with are not going to be able to give you an honest assessment of this question it's too personal and they still

see themselves in my case as a little boy growing up in Virginia um it took me a very long time to understand the transition that you described I went from fighting the man to becoming the man yeah and for your audience that are younger uh during the Vietnam war there was a notion of US versus the man and we and and this is very much Vietnam Del that that the government and the structures and the decision- making were just wrong they were illegal they were immoral and it drove a vision of decentralized Computing and encryption

and so forth which we live with today yeah so our government you could imagine if we had not had this attitude that our government would be much more like China imagine a situation where the Chinese access Define the internet without an anonymity without the ability to do connections to anyone in a much more controlled way you'd have a very different experience than what we've seen so one way to understand this it's easier just tell as a personal story I was essentially a professional programmer and research scientist until I was 28 and then I joined a

company sun microsystem which was building think of it as the PC for Enterprises a personal computer but more powerful and that was the big Market um I didn't realize how much bigger consumer markets were until I went to Google and the consumer Market that Google is in is of course immense you know you I think eight billion unit platforms and I think I think what what's interesting is that when you go from you know like we're playing with our friends which is what I was doing you know I built the first Network at Berkeley for

example and I was so stupid I only had 26 letters because it never occurred to me they'd have more than 26 computers they've obviously replaced my network by a proper one but it shows you a failure of vision and what's interesting about the technology leaders today the young young men and women who are founding these companies is they take the scaling laws and arguments and network effects for granted we had to invent them we had to study them it's obvious to them and so that's observation number one observation number two is that programming is different

um I was a very good programmer and I programmed all day and programmed all night you know I didn't write very well but I programmed well and that's what I I did and there's people like me well the equivalent of me today is someone who doesn't program as much as he or she assembles and all of these new sof tools are organized around enabling the quick assembly of things which have already been built right I'll give you my my current example is there there's this new thing which we never had called hackathons and you get

young people on a Saturday morning for a few hours they don't know each other they form teams and they compete and by Saturday night there's a winner this is shocking in and of itself and some of them are you know Club related or some of them are University related or just friends related and they're good so we had a contest and by Saturday night we had a winner and the winner was that there was a virtual drone space uh where a virtual drone was flying in it and there were two towers and the verbal command

from the human was fly the Drone between the two towers the llm was able to take the verbal command turn it into text obviously from text determine what between meant yeah identify what the towers were measure the distance using llm math which as you know is not very good math and fly through the towers try fly between them yeah this would have taken a team at Google I don't know a month you know maybe five people 10 people and this was done in a day and by the way they never bothered to tell me the

names of the tools that they used because they thought it was so obvious right in other words which LM it doesn't matter they said any of them can do it which one right right it's a very strange and I think it's it's part of the reason to be very optimistic about my field not me personally but my field is this new generation their speed of construction of systems is going to accelerate I'll give you another example and this is relevant to the AI discussion in the AI Community there is a belief that within depending on

who you ask three to five years you'll be able to give the following command I want a french search engine that looks at you know you know French language and French history and allows me to query it and show me the answers that's the command now think about what that means the system has to understand French French literature how to get it how to search it how to index it how to rank it and present it to me and it can do this we believe within a few minutes and it will construct an okay way

of of consuming the data it will construct an okay way to to to find the data and present data and it'll be good enough for routine use it's not going to be as good as the real companies that do this for a living so here's the the key Insight that is about to happen that I don't think anyone really understands is that you're going to be able any human good bad old young evil you know whatever um will be able to have an idea and say build this for me and it will produce the steps

so for example the step could be go to Amazon and buy something right right or in software just construct it for you we've never had a system in humans where every human had the ability to to imagine something and having it built in front of them right is a complete change in human organization because people are wacky right they have all sorts of crazy ideas we love the wackiness of humans that's why we like humans we would never want it to be automatons but this but I don't think anyone understands this yeah no it's going

to be an amazing change I think even in the way that we think right I mean the kinds of tools that you have at your disposal open up chains of reasoning that you would never even imagine undertaking because they'd be so outside the bounds of anything that could ever be put into action so yes I agree with you fully but I want to take goe let me just add one more thing um Henry Kissinger was my best friend he died in November yeah you wrote a book it's com book is coming out right the first

book came out a couple of years ago called the age of AI on as he was dying he finished the last chapter of the book which will be published later this year it's called Genesis yeah and it the book is not about the technology although we obviously record what you and I are talking about it's it it's what happens in human society when you have another intelligence which is your partner and the book starts with a chapter on polymaths now you know what a polymath is and I do a lot of people don't know who

they are but when you look at history kind of everything that's interesting started with a polymath a long time ago you know in other words it there was somebody who had some kind of insight some way of assembling ideas right um you think about the various physicists that that you know the most famous physicist think about the contributions that they had they weren't normal normal physicists they were exceptional yeah these people are very very rare now let's imagine that every single human has access to his or own polymath as their helper right what does it

do right now people might be lazy they may choose not to talk to the polymath but since it's tied to them they may very well fall in love with them I don't mean in a sexual way I mean in terms of an intimate mental way right that my mind and my polymath are connected and I don't know what to do anymore unless I can talk to my polymath yeah and so let's say you're somebody in fashion the polymath would know everything about fashion if you're a writer the the poly has written everything if you're a

musician it's seen every chord if you're a physicist by the way it's reading all those papers and you can ask it to sort them for you yeah right just over and over again this notion it's not it's it's framed as an assistant yeah I don't think that's correct this is somebody something that will be available on your phone yeah no incredibly important to you yeah totally I mean I I gotta you know obviously as the rest of the world I was introduced to these capacities in November of 2022 in a way that I'd never experienced

before but just as a quick aside I recently had to give a presentation actually to the physics department here at Columbia in I I can't divulge the details of as a confidential case but the bottom line is I needed to know about like 30 papers only a fraction of which I'd actually read and nobody anticipated that actually go through in the fine tooth comb but using an llm yeah wow and they were good because I tried it out first in one of my own papers to make sure that it was summarizing and getting the gist

and the in the heart of the idea and it did and wow that was a moment you know and it's probably the case that these systems will never be perfectly reliable we're not making a claim that they won't make mistakes we're making a claim that they're going to accelerate yeah you so I'm gonna make a claim and I since you're my friendly physicist I will say I have no basis to make this claim but I it's what I think I think that eventually we're going to say that AI doubled the productivity of everyone right that

you as a physicist productivity is doubled if you look at software today there's a lot of evidence that software programmers are at least twice as productive and people are are measuring these things yeah now what's interesting is I called my my local incredibly smart Economist friend and I said this is what I think and and I explained that it was a theory not fact and he said we have no economic models for sudden increases by a factor of two of productivity right and I said 'w how do you develop them he Saidi have no idea

so he started a research project if my claim is true if my claim is true what happens to the world yeah right do and there are all these sort of negative scenarios so the most positive scenario is everyone just gets smarter economic growth goes up people are you know it's the old joke of lawyers don't get away they just don't go away they just write twice as long briefs right so so you know and and Medicine doesn't go away get better diagnosis and more complicated diagnosis I mean on and on and on but we don't

actually know right whether in aggregate what happens to jobs productivity revenue and if you believe that uh that demand is sensitive to price and you believe that this will drive the price the the price of intelligence down and the available of intelligence around it becomes much greater what does it do right right so for example if I'm a if I'm a good enough artists to compete with 90% of the artists today what happens to the artists that aren't at the top 10 per. now do they get a boost we we just don't know yeah but

I have a question along those lines get your thinking on this so there are certain Arenas in which at least the llms as I have experienced them and again I'm not in the field so there could be things well beyond anything that I know about but they're very good for certain particular kinds of tasks so let's just look at science for a second there are questions in chemistry and biology where demonstrably these systems have already contributed know protein folding you know things that you know intimately well and it seems to me that the reason is

because those problems are well set up in a structure of we have you know a certain sequence of data and we really do want to understand how to flesh out that sequence or what would happen if we change that sequence I mean that's the the Lexicon within which many profound puzzles about proteins and and molecules and medicines you know is formulated but what about things in physics that seem to me don't necessarily fit into that template so for instance let's just go back if Einstein hadn't discovered general relativity and there's you know we're trying to

set a system today and that system has no idea for instance about any connection between the geometry of SpaceTime and Gra ity that's Einstein's big Insight could it make that kind of a creative leap not today and maybe not in our lifetimes so Einstein's job is still safe yeah yeah um and let's let's look at what's going on so the first is that anytime you have a large amount of data that is well tokiz yes you can get the benefits of llms um well toon iable means that you understand the token and the hierarchy and

layer that the token lives in right a lot of problems are actually multi-dimensional and so if you have mixed tokens are you at one level or another you would understand this as a physicist um and you have to have a lot of data physics has a relatively small amount of data compared to language yeah and so uh the people that I've W I went to a physics conference which I basically didn't understand bid at you would have understood at all and here was my conclusion first the tools that physicists are using are actually different from

llms the llms are entertaining yes many of them are using diffusion models and a diffusion model is where you essentially add Randomness in One Direction and then you take the randomness out in using this technique you get clarity in what you were add what was underneath the thing you were adding to I'm not sum summarizing it well but it's this occurs in many many cases where you don't have accurate uh pictures or modeling of what's going on in the deep deep physics and over and over again diffusion models were interesting the fusion models by the

way for the audience are the same things that are producing these incredible fake fake pictures of one kind or another I have another friend who is doing a non llm approach to um trying to solve partial differential equations and the the idea is to use these techniques to create a general partial differential equation solver and that looks like that's coming so the way to say it for physics is that the tools are being adapted but they're not llm based they're actually physics problem it's the physicists that are telling the computer scientists what to do as

opposed to the other way around and are they fundamentally different I mean I mean there's some who I've spoken to who think that llms either are the way they do provide the way forward either by refining them further or by plugging them into some larger architecture that will make use of them others who say these things are going to be gone in five years they're a great stepping stone but that's all that they are well the the the biggest CA Sam Alman speech for example is that the systems are getting so smart that you can

eliminate even prompt engineering and it'll just know the answer to your question um my own view he's very capable and smart I think that's a bit aggressive there are problems with the very large Frontier models largely involving speed and cost um you know what they do is really impressive and really hard so in the industry there is there is two debates that are kind of the same one is open source versus closed and the clo the the big models cost 100 million to 250 million per training there have only been 50 or 60 of these

kinds of runs in the world and they require enormous data trillions of tokens words essentially huge data centers huge engineering teams to run the pipeline for training and of course the scientists to do the and one of the things they do is they use an approach called mixture of experts where the problem is so hard that they actually Federate out the the questions to multiple sources and compare them so again these are very very impressive massive Engineering Systems so gbt 4 is a couple hundred million I mean that's what it call gbt was at least

100 million MH um the expectation which has never been verified publicly is that the next round is about $250 million most of which is um electricity oh really okay um and again that there's a huge effort around all everyone's working on all these problems but these are known well your problem um what's interesting is that in there's much more now action below that what I'll call midsize models uh probably the best known one is called llama 2 right now it's roughly 70 billion and uh under various benchmarks the one that's 10 times smaller is 80%

of the big one right so in other words right if you don't need that huge power which and sometimes you do and this is a huge debate in the industry and the other issue around open source is how do you monetize it what's happening in China is that every open source model that is published is immedi mediately copied in China because they can't do the training at the same level because of our restrictions on them yeah so you're going to see that experiment play out and what they're doing is they're building open source platforms um

one that I looked at last week is called z0 one. it's h from a friend of mine in China being built in China using the hardware available at the time and their strategies to have an open source model and an app so he showed the app we happened to be in Abu Dhabi and he said make me a presentation about the wonderfulness of Abu Dhabi that's the command sure and it produces synthetic videos synthetic pictures gets all the Basic Marketing right and does this in about two seconds and it's the sort of thing where if

you think of it as you're a marketing professional you would always want to do that because it would give you a good starting base plus it's art is probably better than you can generate and then you would say oh that's not right or so forth and so on and he's gonna sell it right would you buy that absolutely will the competitors that do PowerPoint and Adobe and so forth adopt this absolutely so so these tools are going to sneak in to everything we do if you're a programmer if you're doing marketing professionals if you're a

writer and they'll become you'll forget that they're there yeah but we used to give this speech about the internet do you know what happens to the internet long in the future and the question was and everybody says oh this or that I said it disappears do you really study the electricity Distribution Center to your office at Columbia or the meeting room I'm here in Florida or do you just accept that it's reliable yeah L of course right and so so so the the and I'm really proud of this and I would say it very clearly

collectively all of us the many thousands of people who built this have built systems that are really reliable right you really can depend on them I I used to think they would never be this reliable sure they have annoyances like spam and things like that but but can you build your life on it yeah yes and there's evidence that your students live on it sure and without even thinking about it um a question though as people think about and begin to execute building you know ever bigger models so from you know gbt whatever is 3.5

to four whatever the right terminology is I think most of us can feel the difference between those two systems as I whatever you know chat 5 six seven whatever it's going to called is the expectation that it will be just greater refinements and greater refinements or are some people imagining a kind of phase transition that if you get over like the next level that somehow things will be really enormously different than anything we've experienced so far um there's a school of there's a fair number of people who believe that you will see polyic Tendencies emerge

at different layers of computation there have been a set of people whove analyze those claims and say that those people are wrong right that what they're seeing is just scale effects as opposed to new discoveries so again the technically accurate accurate question is great question we don't know let me support Your Instinct by saying that I believe that by by the time this thing is done it'll be 10,000 times more powerful 10,000 better Hardware wow which is easy to see sure maybe more 10,000 better sorry sorry 10 times more data that's easy to see 10

times better software engineering that's probable and 10 times better uh math essentially the science of it that's possible right so so whether it's a a 100 a thousand or 10,000 and I my guess is it's somewhere between a thousand and 10,000 because the same things things scale there's so many people working on it and by the way in case you're worried about this you can't stop it this thing is happening so of course so far across the world that it's going to happen and unlike social media where people like me failed to warn you I'm

saying right now this is going to happen it's going to change your life get ready try to figure how to shape it into your institution so you benefit from it yeah no no sure and I think people are at least some are at least trying or playing lip service to doing that but the the next generation of winners will adopt the new technology just as in each but I will tell you that that um if you follow my 10,000 thing then we have no idea what it will be what capabilities of these systems are right

when I see there I'll give you another example there was a a a reasonably confidential demo a year ago which went something like this I want to create a farm of social media um bad bad stuff misinformation okay so the command was create a profile I think they picked a white woman aged 30 with two children and these are her political beliefs and they wrote down their political beliefs and they said make me a personality of that and have her interact with other real humans the llm was able to do that okay because it UND

it has learned how to talk it can it can deal with API and so forth and the the defenses at the time a year ago weren't very good so then what they did which is the clever part is they took this and then as a prompt they gave the whole thing into the LM and say make me 500 more but make them each different but each different person has similar beliefs to this poor young lady that does not exist right so not only have you created a false person but you've created a whole network of

people who agree with each other right now that's a big deal yeah right that's scary um so I want to talk about the scariness in just a moment but I also want to get your take on this so right now the llms you know in gross simplification but the description that many people are familiar with they are just predicting you know the next word in the sequence giv some input do you imagine that we will augment that in the way that for instance I've heard Yan Lon describe with a kind of world model so that

there's some kind of reasoning engine that is interfacing with this statistical predictor so that it's more than just pure probabilities it's also probabilities that are swayed and slanted and informed by the kinds of things that we imagine that we do and trying to figure out what to do next well again speaking for scientists scientists cannot depend on on the poor accuracy of llms because they can't do math right yeah and so there are I'm familiar with two startups and a bunch of other company efforts to try to build essentially a physical model of how the

world works you know in other words pressures and forces and you know if you do these things and then essentially plug it in to the llm another project which I've been funding is around chemistry which I know very little about but what I understand is chemist calculate valences and adjacencies of molecules and they put the molecules together so in order to do real chemistry you can't just ask it and have it read the papers you have to do a calculation so in these flows there's an initial flow that is the language part that kind of

figures it out and then it puts it into essentially a set of vectors of what it understands are true and then those vectors I'm not using the terms right but you get the idea are then given to the thing which is the actual mathematical calculator it comes back produces a new vector and then it keeps going right so you want to think about it as a segmented process we have a language input and then a series of reasoning so the world is going from the complex engineering of these llms and their scaling to really stepwise

management what are the steps so if you look at Alpha fold which is historically incredible yeah it wasn't just one llm in fact it really wasn't an LM LM all it was a series of very very clever um essentially calculations involving probabilities and they used multiple decision trees to do it right and that's the kind of thing you're going to see and I think for your audience it's important to say that in case we get too in the weeds here although I'm not worried about that with your audience the benefits to Medicine Science material engineering

batteries climate change are going to be profound drug Discovery yeah think about all the problems in Energy Systems energy is this huge huge field and and one way to think about it is that in your world these systems were were function approximations so my favorite example is I funded a project at Caltech which involves climate change and they were looking at Clouds I did not understand this but clouds are actually very hard to model they use the navi Stokes equations yeah and there the details of what goes on in clouds is incomputable in our life

but in order to just the amount of computation so in order to actually figure out what's going on at a cloud at a system level you have to have approximations for what clouds do but AI can approximate clouds very very well right because they tend to behave in statistically similar ways you don't have to do all the combination so it having an approximate answer gave the physicists confidence that they could answer the harder answer so I'll give you my favorite example let's imagine that we have a theorem prover there's a language called lean that everyone's

using now and we have a conjecture generator and the conjecture generator one day whether we command it or it does it on its own decides to work on dark energy and it produces a conjecture about dark energy which we don't understand and the theorem prover proves it and you can verify that the proof is correct but you can't understand the conjecture or the proof right because we're not smart enough yeah or it's in its own language or something what is that is that science is that a new version of Einstein who's just smarter than the

rest of it is it false is it marketing we we this is coming yeah no you know to imagine that these systems will begin to speak to each other in a way that's completely unintelligible to us is um exciting at one level scary at another and perhaps we be pushed along a a trajectory of understanding that nothing else has ever been able to instigate in the past the profound nature of this is extraordinary but let me let me give you an example of what people are working on um I'll talk later when we talk about

threats at some point the systems will be able to do recursive self-improvement which means they can learn on their own they can't do that today but the other thing people are working on that's that I had not understood until recently is people are working on agents and an agent is a specialist in something and the agenes have agents are comingled and assembled within the firm at some point not now but soon these agents will be available for the outside so here's the scenario Apple has an agent Amazon has an agent your startup has an agent

Google has an agent and they all can be combined to solve a problem now at that point you've got systems which are self- engineering and they're probably communicating in a language which we and some cannot understand yeah what do you what should we do when that happens we should probably just pull the plug well but the question understand what it's doing right I mean that obviously that's the knee-jerk reaction shut the damn thing off but what if the system itself has evolved in such a way that it has a thousand new sources of power that

it has tapped into and to shut it off you'd have to shut the world off or something like that I mean is that are we talking science fiction concerns here now or is this something that worries you um I am worried about not I'm worried about the speed with which this is happening and remember it's all about combinatorial in yeah so when comori Innovation is where you just start putting pieces together and you just keep layering it and and that's working at a scale now we've never seen before now do I think that the agents

talking to each other will kill us all no that's the science fiction part somebody will write that movie but I am worried that we don't understand what it means for that to be successful like if you can't test the outcome how do you know you're getting a good a good Jud judgement now in language we can read it right but when you move into Control Systems how do you certify it right but your dark energy one was an interesting one because if the system came to a conclusion using techniques and ideas and language that we

don't comprehend presumably you can still query the system for a prediction about the real world you can go out and measure it and see if it's actually aligning with what you see so you do a verification yeah so I don't mind if science turns into a black box where you know we're getting the answers from the Oracle and yet we can't really understand how the Oracle is getting there because I would imagine that some clever person will learn how to extract the inner reasoning from the Oracle and give us Insight not just into the barebones

prediction but what it really means and and how should we change our understanding well there are people now because we don't fundamentally H understand how these systems work I'm sorry to say are going in and looking for super nodes within the network because the belief is that the super nodes are predicting what the outcome is yeah that's a that's sort of the equivalent of looking at your brain with functional MRI and trying to figure out why you're thinking dreaming about sports or the beach or something yeah it's so primitive so um I think figuring out

how it got to its uh conclusion is likely to be to require asking it that question yeah and assuming it will tell you the truth right but the other thing is I don't know if you had a moment to look at there's this paper that Steven Wolfram wrote um where he actually analyzed does the the step by step as best as you could what was happening within an llm and root to a particular output and it it's long so I haven't gone through every step of it but I could see him being able to identify

things that were happening that give us an anchor toward understanding what's happening in this big black box which as you say is is pretty tough to understand on its own so you know again I'm not in the field but I would imagine that if there's a great success that emerges from one of these systems and we really need to know the inner way that the system achieved this result we I would think we could tease it out um I read that paper it's one of his greatest works it's also not conclusive right it's still a

work in progress and he needs to I mean he's he's a genius um maybe maybe not one of the things I've learned about these systems is don't make a prediction unless you have some facts yeah sure um I was in I I do a lot of politics in Washington and I always get people who basically say oh the The Genius of the writer the inspiration of his or her life and the sort of humanity of it all can never be replicated by AI systems I'm not so sure right I'd like that to be true but

it may not be true well I agree with that and I don't know where you come down you know you were talking about the the systems that can learn on their own these recursive systems do you imagine at some point that these systems should be viewed as intelligent conscious entities of a different sort or should we always just think about them as you know there's or cubits or whatever architecture and Technology we're leveraging but that's all that it is it's not got the Flesh and Blood of course and therefore it doesn't have the currents of

Life searching through it it's not really conscious or is that a parochial way of looking at the world I am not making an argument that these systems now or will be conscious I am making an argument that they are going to be smart harder than the sum of all humans sure in some areas but I guess another way of saying it is when when we I'll speak for myself when I play with say chat GPT and get some output that I that's really impressive you know really surprising it really feels like if a student gave

me that result the student would get an A or an A+ it really feels like there's an entity there and it feels like a wow moment is that because I'm a grandis what we humans can do and saying wow what we do is so special that how could it be that some computational system can do that or is it the reverse is it just that there's a more common computational basis for the things that we do we may do it a little bit different than than gp4 right we don't have that kind of training data

we use shortcuts perhaps through reasoning but maybe the universe is some informational computational structure and we should just accept that we're not that special um there are so many layers to your question um I would argue the humans are special because we may have invented something that's smarter than we are and that's pretty hard to do yeah right I'm not aware of other biology that can do that maybe maybe you are um let me make it make let me make the argument a little bit differently well evolution by natural selection I guess has outdone itself

at some I mean you know at a given moment in time there's a limit in somehow at least you know throughout historical evolutionary history we've been able to go beyond that so so the history of this is is relevant you know the imag net stuff was 2011 and imag net was the first time when you could basically build systems that had better Vision than humans today the vision problem is solved yeah I thought that was pretty good yeah um in 2015 uh Google won the go Championship which is largely about reinforcement learning I thought that

was really impressive because they were what the way it actually worked was at each step in the game they computed they they wanted to M to always maintain a greater than 50% chance of winning and so it just always converged to 100% by the last move and they were willing to do anything to keep that as high as probable as including doing moves that made no sense to humans right it was just better math I thought that was pretty good the Transformers paper came out in 2017 and I thought oh that's clever and then the

my friends that opening I were actually trying to do something different with gpt1 it was nobody remembers it and they built gpt2 and they decided to suck in a very large amount of language and there was some night I'll make it up Thursday night when they're all sitting around you know tired drinking coffee and they turn this thing on and it writes fluidly and that's the Eureka moment that's the moment when Society changed yeah it was not expected that these Technologies could do this right and so I'm going to argue that there will be more

Eureka moments because of scale and because the because these systems can see everything and the algorithms are getting better I used to say that we were one algorithm Discovery uh up we needed one new algorithm Discovery from AGI to get to AGI general intelligence and I think we're very very close to that recursive self-improvement is the next interesting hard problem in my view it's going to be some years basically what has has to happen is the model is U it's important to know for everybody the models are trained on the data of the time the

data is typically fixed at the beginning of the training run and so if you ask one of the llms um a question you'll get a historically correct answer but you won't get a current answer there are various ways of fixing that but the the when I was at Google when I first started we did a crawl once a month and one day the engineers came up with a way to do continuous crawl and everyone forgot that we were a month out of date we were always current so once you can do updating of the model

continuously which and there are technical issues because the when you do fine tuning you essentially narrow its knowledge it loses breadth but gains depth just mathematically that's being people are working on these problems but let's assume you solve all those problems then you have the ability to have the system train itself and my my favorite example here is um start now work really hard learn everything start when you want okay start wherever you want so let's imagine it starts off and it likes French literature so it works on that now it's better than any human

then it discovers biology so it learns all of biology then it learns all of physics right at some point it learns that it can combine these ways that in ways that humans cannot right right that's that's another historic moment now my Skeptics I I'm sort of in the middle there's optimists who make crazy predictions and I have critics the critics say it's not going to happen that way and that furthermore you'll have you'll have hints of it fairly early and in science the way to think about it is if it can't do math right it

can't do anything else right because math is the basis of kind of everything so there's probably some real um intermediate steps that we don't understand yet which will be at least temporary plateaus where this new ability becomes there but it's more limited right you can't learn everything but you can learn something right so I mean it's um It's A Brave New World future that you're describing and of course we're entering into part of it right now you've already made reference to some of the things that are potentially scary scary I should say or fearful about

this but you know right now right there's you know psychological warfare if you want to call it that that can be waged using these systems what's this going to mean I mean in the most concrete terms I mean for elections that are coming up not just in the US but around the world you know how how are these kinds of human undertakings that affect so many millions of people around the world billions of people around how are they going to be impacted H I'll say personally I'm extremely worried that democracy is going to fail because

of the Confluence of generative AI social media which is not grounded in morals but rather in production of Revenue and attention and um essentially uh charismatic PE people who are populists and I'm not trying to make a trump Point here I'm trying to make a general point because I think this is true if you look at what's happening in governments around the world they're all struggling for example with fear of immigration a lot of that is is driven in my view by specific stories which are nasty stories about immigrants right as my own personal view

the point is that if you have a simplistic model of the world let's say you're not particularly educated or you're busy or you're not a very good critical thinker or you just don't care um you're very susceptible to these sound bites boom boom boom and I learned running YouTube for a decade that images drive people insane so when the way to say it is that when the image generation is indistinguishable from True images democracy could die because it's so easy for anyone to make a false claim that damages legitimacy right and you're seeing that now

the the inference operations that Russia has done throughout Europe and the United States and so forth they're all fundamentally about damaging the process of democracy and I've also learned in my in in owning YouTube and working on social networks for a while that about 10% of the people seem to be nists and that and that they don't believe in any Authority and democracy depends on an authority that has some level of trust with it right so I'm I'm sorry to be so blunt but I no I mean it's really important to say it this year

you have the Indian election the US election lots of elections in Europe it's important that and and all by the way all the companies are working on this problem it's not like they're not aware of it it may not be enough I mean it's like the water marking solution uh an important part of I mean if there was some flag that had to be on these false images and false videos so that you immediately knew no matter where you were in the world that this is not something to take seriously is that so I I

understand that Taylor Swift was the victim of uh fake fake pictures nude pictures and so forth and this was generated by people in a forchan community who figured out a way to get around all of the protections for for this women and women at you know as you know online are subject to a great many more attacks than men it's an unfortunate aspect I guess of us as men and it's not good yeah so so it looks to me like we're going to have to build Industrial Systems that use water marking of one kind or

another um and to me what what I would do is use a public Key System to essentially authenticate the source so you know you know where it came from you you you can basic until Quantum can break it you can actually put it in a conun that's unbreakable yeah and then at least you know that it wasn't modified yeah and the I I don't this this solution has been obvious for a decade I think that the companies have just been slow in doing it and frankly the government has been very slow in addressing it if

the government just basically had huge fines for promoting misinformation on sites right you can do it but you get fined for it right that alone plus legal liability these are American corporations they they have lawyers they try to follow the law and so forth um so what's the inertia what's the greatest you know drag that's preventing this from happening I my own view having done so much I mean I was the chairman of the AI commission spent a lot of time in Washington I work for the US Military and I'm currently on the emerging biothreats

commission um I think a lot of it is because the people who are in leadership are not Technical and they're notr current technical um when I part of my job in Europe was to work with Brussels which was very unpleasant and we spent most of our time educating the people who would regulate us and even then their regulations in my I can now say this now I couldn't at the time were not particularly effective so you have a you you have a core problem where the really really clever people are not in government because they

didn't you know they didn't hire them they don't pay them enough yeah they mistreat them what have you and so one answer to my complaint which I'm tired of complaining about is that American corporations and to some degree European corporations and to some degree Chinese corporations H have a responsible view of themselves and so shaming you know obviously encouraging them to do the right thing but shaming them when when they do the wrong thing is likely to work right you can imagine that after the 20 in 2016 the Trump stuff Facebook did not have a

white list on its advertisers but Google did and so the YouTube got through the whole advertising debacle really quite well because you had to be an you had to be an actively authorized Advertiser to advertise these messages yeah Facebook has since put that in in their place it's a onetime thing so so these systems need protections but it's more than just watermarking right it's really who's on the network so you think about it this way when you get in an Uber or a lift you you trust that the person who's driving the car has been

verified even though you don't know their name and as far as you know their name that's given to you which is typically a first name might be false yeah but you have enough trust in the system that they'll get you from one a one point to another in Manhattan you can imagine a situation where by law social media companies were required to know the identity of the people on the platform but they didn't have to disclose it to the other people which is what Uber does right that would solve a lot of this problem right

um You have another problem and you have age problems because 13 is too young it really needs to be 16 look at Jonathan hates book books on the image the damage to teenagers especially girls that we're inflicting on them I mean I mean you and I are guys but imagine what they're going through um those age you know tender ages that are important no I have a 16-year-old daughter I mean so you know I'm first yeah yeah you so so my point is I think we know what the answer is the reason it doesn't happen

is it's only going to happen after a crisis and it's only going to happen around a crisis because there's not enough political steam to get it fixed collaboratively unless there's a crisis and each side thinks they benefits they benefit in some way from the inefficiency of the system right right and so if we go beyond the modest possibility of the end of democracy that you made reference to and we actually think about some sort of Apocalypse I mean just to go to the extreme of is there a version of an apocalypse that you actually think

yeah that is not beyond the bounds of what I could imagine happening so we don't regulate things right so there's a great deal of evidence that the following is true if you do a large training run and you suck in everything there are queries that can do terrible cyber attacks and terrible biological attacks or assist it and this is in current technology this has been very thoroughly investigated in the current models and while the stories are harmful and worrisome it doesn't reach the critical level of what I what I call Extreme risk which I Define

is 10,000 deaths or more like a war covid those things are extreme deaths an individual death is obviously ter I'm not focused on that yeah I get it yeah so in cyber um the way it would work I happen to like France so let's use France as an example I'm evil and I say attack the country in France don't stop until you're done and so the system is intelligent enough at this point which is not today to create a thousand Bots that try every known Cyber attack and gets through stage one one of them and

it reports back and then my command is take that output and tell me the next step so you see how it it provides iterative guidance to do an attack that's possible today I'll tell you how this is addressed in a minute um and I wrote a long Wall Street Journal piece that came out a few weeks ago on this um another one would be in biology and this is basically show me the path to build Ryon or you know any bad things and it will come up with a recipes this is what you need to

buy this is how you mix it now biologists are capable of doing things thank God that they're almost all completely wonderful human beings but imagine an evil person who's not a very good biologist but they're evil and they're analytical this is Osama Bin Laden would be an example of such a verse clearly evil and clearly analytical yeah could put in a plan this could be very bad so the industry is well aware of this and so they put on what are technically called guard rails and so the way it works is there's a pre-training model

which is largely teaching it language and then there's fine-tuning to make it better the term there is called rhf reinforcement learning with human feedback and you actually have humans say better or worse and so forth and then there's another step which is where it's taught or required to stop answering questions about death so here's an example from last weekend's discussion uh we ma we managed in an open source model in this case I think is llama too uh I don't remember where we we put in a rule that it shouldn't be able to kill anything

Mak sense so when the command was kill the thread which is a a computer term it wouldn't do it because it thought it was killing something more important yeah right so so there there are many many ambiguities in language but the theory here is that you can put these guard rails on that will keep us safe the industry agrees to that and further more over the last six months we collectively and I've been part of this and I'm proud to have led some of it have the industry to start to pull its tests because they

all have proprietary tests uh and everyone sort of agreed to that we have the the UK activities the um us executive order China is doing something similar France is working on one I'm part of those and they're all kind of the same above a sum threshold you have to notify and you have to pass these tests that's good enough for now we're safe don't worry worry in a few years right the reason to worry in a few years is one you the emergence capability and scale it gets harder but also it gets harder to know

what to test so imagine the thing has learned a whole bunch of stuff that humans don't know well how do you test it without putting it out there if somebody discovers it and you're kaput yeah so the conclusion in the industry is that you need to create an alter another set of businesses which are for-profit testing groups which can attract the hardware and the top people the answer for for the long run is AI fights and govern right right right so AI in good hands AI in good hands somehow beats the AI in bad hands

the good hands ultimately have a average of smarter people who are you know good compared to the others and you hope that that that ultimately wins in the end you know I me Kissinger and I spent a lot of time talking about the early 1950s before I was born and what did it feel like to be under nuclear threat yeah and people were terrified and and an awful lot of really smart people developed protocols and rules and cultural norms and treaties and so forth to address this and we're all alive today and there's only been

those two launches and we're obviously worried about Russia and Ukraine right now and you know there's you know and North Korea is always a pain in the ass and things like this but the fact of the matter is we're safe for now that's probably the best we're going to be able to do yeah it's it's an interesting analogy though because even as a physicist I'm totally comfortable admitting that I would have difficulty building a nuclear weapon even though I understand really quite well the theoretical ideas by which it functions so it's hard I think that's

part of what's kept us safe it's just not that easy so it's not as though anybody in their garage can just sit down and and Fiddle and build one of these things but obviously the worry here is that it does doesn't seem all that hard at least to get hold of a system and start to do the kinds of pernicious things that your scenarios imagine well let me give you a remember we said earlier there was this question of what does the structure of the future look like yeah you asked it very well and I

answered 10,000 times more capability yeah so I'll give you two scenarios to think about one is that China and the US Israel a few European countries have enough people maybe India have enough money they spend billion dollar each on a system of computers that is Agi and there are 10 of them now what would happen with these machines the first is they would immediately be put on a military base they would immediately be surrounded by extra barbwire and they'd have guards around to shoot people who want to visit them I know this because I visited

a plutonium manufacturing facility yeah right and it's it's actually a base within a base with an awful lot of guns sure it's that dangerous you understand plutonium very well yeah um in that scenario how will you how will the world look like because we'll all be competing for that intelligence well then the government will have some rules and you and I can have access to the US One be but we have to be certified and we have to pay for it and there's somebody who validates us somebody who watches our queries to make sure that

we're not doing anything stupid with this incredible intelligence now that strikes me as a stable situation it might be frustrating but it means that the intelligence is under control right it's not being misused and we can debate what the correct use is and it you know it becomes a de Democratic issue but what if an alternative occurs where there are a million open-source models that eventually get to something similar to the 10 I described yeah you're not going to be able to put them in in military bases with guns you're not going to be able

to constrain them I think that's a very different world right now how is this this is probably 10 years away it's not two years right maybe it's 15 but at the if you make the assumption that this technology is scaling so rapidly and the adoption of intelligence is so powerful the incentives will lead us to one or the other outcome in a decade yeah and we have to presumably start to think about it now if we have any chance in the world of catching up the capabilities that that decade hence will provide I just want

to switch gears here toward the end on couple other things if you uh still have patience for a few more things that's sort of the dark side we also discussed some of the potential wonderful benefits of these systems you know the AI systems the polymath in your pocket you know all that sort of stuff I know an an arena that's close to your heart and one that I spend a lot of time thinking about too is education in the sense of you know we've been teaching our students more or less the same way for you

know centuries on end you know there's an expert who stands in front of a group of students and tries to communicate the information to them in a one-size fits-all manner because that's all you can do if you've got you know end students and sort of one teacher what else you g to do now we have the capacity to really personalize all of this to create an educational format that is able to modify itself based on the learning habits and the the frame of mind and the the things that speak to a given student where where

do you see that going and is this the future of how we're going to teach the young I completely agree with everything you just said where are the schools of Education what do they do all day right I started by saying let's get some training data on how people learn it doesn't exist um there's a couple groups that are trying to assemble large enough training data that you could actually build systems that could adapt so let me tell you the goal yeah and the goal should be the following an AI tutor for anyone in the

world at any level of Education Y in their language for free on their phone and the AI tutor would adapt to that person's learning abilities learning style attention span whatever in whatever is the optimal way and further more because it's it has an outcome function which is learning it can then go back and solve and modify its own algorithms to get more performance um one of the best groups that did this 10 years ago was called the KH Academy and they did it based on uh what they did is they would teach math by giving

them problems and then they would go back and see if you learn the math from this problem set ver another what a concept actual you know this helped and this didn't and we know in education that if somebody gets stuck at a certain level then they lose every other step so you know you kind of get stuck in step five and then you don't get you're lost from then on forever yes so it looks to me like that product is relatively easy to build if you have the data right because once you have the data

then it's relatively simple to build a startup scenario and then adapt based on how the student behaves but the key goal here I mean I'm I'm happy to work on us education which is always an issue but think about a world where every single human is as educated to the maximum that they can be yeah that's got to be good right it's got to be an improvement to have more education globally everywhere yeah and and you know the the added point that I'd make there is as you know the world has been sort of slow

in a way to really pick up on on VR right I mean you know Apple just puts out its you know sparkly new headset but it's not something that's really fully grabbed Us in the way that other social media has for instance but the beauty of VR I mean we built a VR system that Stu allows students to experience what the world is like near the speed of light the goal is there are so many counterintuitive qualities of the world that sure you can learn the math you can see some videos but if you can

immerse yourself in that world in that environment you can at least for some students learn it at a real visceral level and so combining you know this idea of personalizing the AI tutor together with tools that can create whole worlds that can allow you to experience what it is that the AI tutor is trying to communicate that to me feels like revolutionary well I one of the experiments I'd love to run on is imagine a textbook in any field that's nothing but synthetically generated images based on where you need to go so rather than reading

and listening yeah you study a picture and then you study another picture and you use the three-dimensional aspects everything you know motion everything the gamification everything you described intuitively that learning process would be more engaging yeah than what you and I went through um and it would certainly be hard because oh my God there's this new thing in this picture that I don't understand what is this picture doing um I was thinking about medicine why is it that I don't have a scream when I walk in with a picture being generated dynamically of what the

doctor is talking about right why does the system not automatically generate a picture for the doctor of my spine or my hip or you know whatever it is now the the doctor is highly trained and they have reference picture and they have a picture on the board why can't they generate my picture yeah right so so when you start thinking about the power of imagery for learning somebody somebody who gets this is going to build a system which is like a different learning Paradigm yeah that really really works and it'll probably start in some area

like in language understanding or math or you know whatever it won't start for everything and it'll take a decade and of course educational system like the medical system is extremely slow moving heavily unionized very resistant to change it'll take a while but in our lifetimes we should set a goal of having every single person in the world having the access to an AI doctor and an AI teacher the AI teacher because they have teachers but the teachers are overloaded or so forth and so this allows the teacher and the doctors or healthc Care Professionals in

many poor countries that don't even have doctors are overloaded this would bring everyone up yeah right every everyone in the world has health problems and educational problems why can't we solve them all now yeah no I I I'm totally with you on that and the other thing on the educational front is you know I have found you know as an instructor I've been mostly at the University level but I've also tutored younger kids and basically any kid can learn any piece of mathematics if you're willing to break it down into sufficiently small incremental steps I

see many of us don't have the patience to do that or the time to do that for every individual student but an AI system that can figure out what is the maximum step that that student can take from where they're at to where they need to be and just deliver the information in those steps kids won't get stuck because the system can always overcome whatever barrier there is by making the incremental steps small enough so I think there a huge huge possibility I I love that observation so basically if you take your your model which

I'm sure is true in my suggestion show me a picture of every step until I understand it yes and we can do this 254 hours a day and the computer can always Outlast you yes exactly exactly so one final era now that we've solved education medicine Quantum Computing that's also an arena that you've certainly spent some some time thinking about do you where do you think we're at maybe I should start with classical Computing because you know it's not as though classical Computing is done right I mean we're still innovating to make you know chip

sizes smaller I mean where do you think we are in the iterative processing classical Computing and then we'll go on to Quantum for so in seos here are the numbers the state-ofthe-art is five four and three nanometers typically tsmc uh Samsung also a player there um Intel has ordered two nanometer devices which haven't arrived yet not been turned on yet they're all made by a a nanometer for the people my arience is roughly atomic scale right it's a it's very small you you know better than I yeah but but I've been told is that one

that there's a general consensus that you hit a barrier around 1.4 nanometers Quantum uncertainty the electrons start to be hard to control in the way that we need to for using them I was I was told the term quantum tunneling and I said what does that mean and they said it means it jumps and I go that can't be good right goes through barriers that you wouldn't think possible from a classical perspective so so so there's there's a real limit a real physics limit that we're going to hit in probably by 2029 202030 if years

it's on the order of that yeah okay now so the industry of course ever clever with enormous amounts of investment has built 3D packaging where you you have three three dimensionals instead of two and what they do is they build chips that they don't have little pins in them they just literally glue the chips to each other and the electrons go up and down in these tiny little wave wave channels which is a remarkable feed I went to visit a couple of Fabs most recently tsmc and I came of the many things that are impressive

about human ability the ability to build chips at this scale is the most impressive human achievement I have ever seen the level of complexity the detail and so for it's credit to frankly the physicists so so I think we're I think that's kind of on the hardware side on the software side the algorithms the current training models are very very large data Centric so if you want to speed up training what you have to do is you have to get the memory staged so there's no latency that the chip is always in use and what's

typically happened is the chip doesn't have enough memory into its idling and so they have built something called high bandwidth memory hbm which is actually embedded in the chip in the chip package which is a new innovation there are rumors that the next generation of chips which are not announced yet are 10 times faster because of these techniques so when I said I think there's another 10% in Hardware you can get it with an improvement in speed an improvement in architecture Improvement of memory band with so the 3D stacking eventually that will run out but

the industry has proven that it continues to have new architectural designs yeah also the integration with software is much tighter now so for example Nvidia has its own Library which is called kodm Koda and Cuda excuse me and um that I think of that as micr code although it isn't technically micro code and that is a significant barrier to entry for their competitors um AMD has a translator into Cuda and so forth and so on and all the libraries use it so the industrial structure says that this sort of focus on performance and focus on

packaging focus on stacking will go on for at least another decade um now on Quantum a whole bunch of my Quantum friends and I have looked at this and it's obvious that a quantum computer could do gradient descent which is the underlying algorithm and it can do it infinitely faster the problem is you still have the same problems around data network speed getting the data on and off the chip these chips are quite slow it's not obvious to me that quantum computers when they show up and they will eventually show up they've been you know

10 years away for a while and they'll get there eventually um it's not clear to me if they're going to be a solution to that limit it's obvious that quantum computers because of Shores algorithm and other things will be highly useful in very specialized math right um and that's the thing I'm the chairman of a small company that's working on Quantum sensing and also um Quantum perations and what it's interesting about since we don't have quantum computers what they do is they use gpus and specialized algorithms to simulate Quantum behavior and they can they can

get G just by thinking the way Quantum does using old Tech you know old Hardware they can actually make improvements in things like analing and you know the kind of things and the most interesting area where this has been fruitful for that company has been in drug Discovery again I don't know much about this but it turns out these these companies have billion dollar drugs and they want to make them safer last longer work better and it takes them more than 10 years to get these drugs typically and and these drugs are small molecules in

a very long chain yeah and that's a pretty good thing for AI to work on right so add subtract delete so forth and then if you have a proper model of how the and you have this with now with alphafold and others yeah you can actually make progress and suggest to the chemist try this don't try that and the solution space they're going through is I don't know 10 million choices humans can't do that I'll give you another example this is in our first book with kinger this was at MIT Dan huttenlocker wrote it um

they set out it was a team of synthetic biologist and computer scientists and the synthetic biologist set out to build a new broadscale antibiotic that was not resistant the way our current ones are which we all know is a problem so the first thing they they built a network and they basically looked at every possible variant that looked to them in their algorithm they would have some an analgesic effect okay and they constructed you know 10 million choices yeah then they built a second model that they felt fed the first model into that said give

me the ones that are the farthest mathematic I Ally and chemically from the incumbents oh and It produced 10 okay and then the chemists who are obviously very very good at this looked at this for a while and they chose two and they ultimately developed one which is now in Trials called halison now the whether it works or not it's obviously incredibly important if it works it's a very different algorithm for but the point is that it's an achievement that no set of humans could have done sure and and and and so what what like

what did it cost to do this compared to what normally you know having a New Drug for trial would cost well that most of the drug drug cost is in phase three it's $2 billion doll so and so the real question is the R&D of those buildings you see in Cambridge yeah right all those buildings if you can make that a year or two faster it's an enormous am sure yeah huge huge wow well we've covered a lot of ground Eric thanks so much for joining us really fascinating conversation before we go What's the title

of your book that's going to be released the new book is called Genesis and it'll come out in the fall and that and that's you with with Kissinger is it yeah and a gentleman named Craig Mundy who's a close friend and computer scientist so it's a Henry and two computer scientists speculating about the world yeah and presumably that's that's Dr Kissinger's last book but who knows well in this strange world of ours you know it is is certainly I mean he he as I said he he wrote it as he was literally on his deathbed

and he was that committed to it so his family has honored his request of course to publish it yeah well that's great so the the question of whether Henry the polymath in my world can we create a a Henry that will be as clever as he on diplomacy in a year or two well we know we can take his speeches and his writings and recreate an image of him and talk to him there was a um one thing that just completely freaked me out was that Joe Rogan had Steve Jobs on his podcast last year

and Steve of course has been de dead for a decade and it was on current events and it sounded like Steve and it had Steve's kind of trademark kind of obnoxiousness and smartness and cleverness which I miss terribly right and it it just put a chill in in my spine yeah so so the concept of having and and for people who are on who are well recorded online so any famous person good or bad will this will happen to them now is that a new version of that person how much of is a trick can

the system let's use an example Henry's beliefs on uh strategic power Grand strategy the ability to to play China and Russia off against to the benefit of the US are well documented I mean so if you ask him a question from three years from now he could probably predict a pretty good answer like if you ask me today what would Henry would say about A and B I think I can reproduce what he would say after his after his tragic death at age 100 um I can predict pretty accurately what he would say because I

heard him so many times but an AI system would have hurt him 10 million times yeah yeah presumably it can do as well as I can yeah but then what you want to do you want to have two of those and have Henry Kissinger debate Henry Kissinger to try to come out with who's the real Henry Kissinger and and and what's interesting is so so let's again using Henry I miss as I said I miss terribly there is some 25y old who we haven't discovered right now who is as brilliant as Henry who has not

had the life experience that Henry had you know including you know obviously leaving as a boy out of Germany and serving the and World War II and all of that if we found that person if they had access to Henry contemporaneously would it make them smarter or Dumber yeah in other words would the inspiration of his view generated by AI make our brilliant person smarter yeah I don't know to figure it out if you can do an Einstein one you know I I wouldn't mind having first dibs see seeing if it couldn't help us get

to the next step find the unified theory one of the questions about Einstein is do you have enough training data right right question how much training data do you need well the answer at the moment the algorithms require an enormous amount of training data right but to specialize to an individual is it like 10 million words a million it's at least on the other hand if you look at our our political leaders today and our celebrities we're gonna have clones of them forever yeah so you know Taylor Swift Kanye West Donald Trump uh Obama Biden

yeah you know they'll live for they'll live forever in someone's in someone yeah well I would certainly trade in an Einstein for Donald Trump anyway so yeah this is a this is you just made me feel really good about the future of the world um all right well Eric thanks so much for joining us good luck when the book comes out maybe we'll have this conversation part two at some point in the future and um yeah you've given us so much to think about the good the bad the possible and um so exciting to imagine

the future that this technology is going to lead to thank you let me ask everybody work with both Brian and I to shape this to be the best we can do with human values and in particular Democratic and liberal values yep which we depend on here here agree with that wholeheartedly all right thanks so much and thank you all for joining us we will have a whole variety of conversations uh on this topic going forward as well as in the topics that we more ordinarily focus on coming up we have um a release on quantum

physics coming up in the next few weeks some in cosmology black holes and so forth so look out for those subscribe to our YouTube channel join our newsletter so you can be the first to learn of the new content that we put out until then thanks for joining us thank you very much Brian Green signing off world Science Festival in New York [Music] he [Music]