What Does an AI Utopia Look Like? | SXSW 2024

good afternoon wow what a great audience yeah so I've flown all the way from London for this um I wish it was on a Hypersonic plane it would be quicker but we got a long time to wait for that you have a long time to wait for Hypersonic planes but we are working on them great so get the world shorter is a good thing clearly clearly clearly clearly connecting us all so I I I have this model London New York in 90 minutes so you can go there for lunch is very much on our set

of objectives be great for New Year's Eve we celebrated in different places so I I've got the great pleasure here talking today with VOD kosa the preeminent and world-renowned moonshot investor moonshots really really important we're going to talk about an AI Utopia we're going to talk about highly sustainable cities of the future what do they look like when will be when will we be living in them what kind of intelligent infrastructure we need to support them and I think we're going to view this through the lens of all the businesses that you've funded to get

there um but where do we start let's start with like this high level view of this future tell us about the city of the future I don't know 2040 2050 what does it look like well so the first thing you have to realize is cities evolve incrementally they you don't just design a new city OCC occasionally you can most cities evolve organically how many people have read a book called scale was out of the Santa Fe Institute a few hands I'd highly recommend reading that book if you want to understand how cities evolve and what's

common between the the the structure of a city and the structure of a tree which is a natural biological system it's a beautiful book uh one of my favorite books um but one thing is true of cities there's only one thing in a city that is totally fixed and that's street wids so uh I see Mark uh sitting here in the front we've talked about this a lot uh if you have fixed widths because you're not going to turn tear down all the buildings on both sides of the Street not practical how do you make

a city function better um we know from many studies that lifestyle improves if everybody's commute distan is under 30 minutes now of course we keep everything to under 30 minutes in the current Paradigm housing prices go through the roof so um the core function of a city is moving people through at the maximum throughput per any given length of stre Street width what we've sort of tried to Define at a company called Glide wise is have the throughput of Light Rail in a bicycle width structure that can be in the bicycle lane or above the

bicycle lane if you if you not dealing if you're dealing with too many intersections and all that and you can do this quite simply in uh frankly in prefab infrastructure so it doesn't take 20 years like the Boston tunnel dig Did you can do it quickly and then you get people to get through cities faster and I believe the core thing to improve a city is improve Public infrastruct public transit infrastru structure and I hate to call it public transit infrastructure uh because if you use imagination a new kind of infrastructure is possible so what

do we do to increase throughput in cities today which is ass backwards to be honest um you go from a bus to a double deck of bus to a light rail to a heavy rail you make him bigger and bigger and bigger why cuz you need one driver to drive these things and you want them to drag as many passengers along great for the driver are the public transit system economics if you pack these things with people so many many years ago I did the following calculation I took all of electric right Light Rail in

all of the Northeastern United States and said how many passenger miles did we have on the system and what was the carbon Emissions on electric light rail or electric rail in Northeastern United States turns out the average carbon emissions per passenger mile are the same as cars we could call them electric but they're feeding off a natural grid and I was shocked and so I've been thinking about this idea for a long time um in thinking about the fact that cities are uh are fixed width streets so you just have to increase the throughput a

lot and there's one way to increase throughput which is to have things not stop at red lights or traffic lights just keep going there's one way to make it more convenient for people who use it because more you aggregate the less people are likely to use it you get off your restaurant job at 1:00 a.m. at night you you want a car or a transit system there otherwise you'll drive your car and to meet that it has to be on demand it's kind of like analogous of your uh of your packet routing isn't it back

in the days packet R analogous to packet routing in routing networks uh George knows I worked on Juniper in the 19 95 96 time frame one of the most successful Investments we've ever made we made about 2500x return on our investment you don't need a lot of those uh but it was the idea that packet should be routed individually not as a bulk which was the traditional Telo model I won't go into that too much um but this idea sound really exciting if you can put light rate speed or throughput into a bicycle lane width

if you can make it on demand so more on demand than Uber because Uber may take five or 10 minutes to show up it shows up when you do and these are small parts so you get the privacy and security if you're a woman getting off your job at 1:00 a.m. you don't want it shared with a lot of other people all that is possible if you imagine public transit differently the other thing when it's not shared is it goes it doesn't stop for anyone else which means gets to its location faster in fact I

believe it gets to its location faster than if you had a chauffered car cuz that would still be in traffic would still stop at traffic lights you get to wherever you're going faster don't have to hassle with pack uh parking it's on demand so when you want to go it shows up it's just a beautiful system and that would decongest cities there's also materiality to this isn't there because we we drive around in cars that weigh two tons to transport 70 kilos I guess in the future all of that steel off the roads can be

put to other uses yeah so and we can talk about cities more generally about this about I'm 69 when I turned 60 I started thinking of interesting problems to work on uh and I took 3 months in a lot of hiking uh and writing to write a 50-page document that's public called Reinventing societal infrastructure with technology and I just decided after that point for the next 25 years I'd work on these set of problems um by the way 9 years later it's still 25 years Health permitting uh I'm an optimist um but one of the

points I made is we have to dematerialize the planet if you look at the world today s or 8 billion people now almost eight 10% of the planet roughly has a rich lifestyle in every way and this winds back to the AI AR uh argument so 7 800 million people have a rich Lifestyle Rich in transportation they have private cars big cars rich in education rich in entertainment rich in housing rich in health care so in all ways if all 8 billion people want it can we have 10 times the amount of Steel 10 times

the number of cars 10 times the amount of cement it's just 10 times the number of doctors 10 times the number of teachers it's not possible that is not the right way to scale the planet yet we want to provide all 8 billion people with a rich lifestyle akin to what the best 10 the richest 10% have in this broad sense of richness not just income richness uh that's the problem I worked on and I'm very very excited about redesigning cities is a big part of it now what might this city happen it's very clear

today that an AI can have all the expertise of a doctor and we're working on it so having a physician costs you a dollar a month uh because it's compute cost only for everybody that' be a city service or national service having a personal tutor for every child on the planet that's an AI tutor can pay way more attention than any physical T tutor can and my works on my wife works on a nonprofit called secret 12 that is building AI tutors and they map gaps in a kid's learning far better than a personal teacher

can and then they can tutor to the gaps in learning so imagine all these services and if your landlord is being a meaning you get a AI lawyer to take him to court because otherwise it's hard to access justice so I imagine most expert in AI systems many of them so cheap they provided part of city services um you never have to apply for something in a city uh permit or a building permit the AI would do that and the AI would check your plan of what you want to do and see if it's compliant

and dialogue with you and you're done I tweeted this uh last week like somebody should build this startup nobody's building it today so anybody interested in uh AI agents that eliminate the need for City bureaucracy would be awesome but you can imagine these cities with with high throughput with a new version of a transit system nobody would want to drive a car if you can get there faster without a car you don't have to worry about parking it costs a fraction of what it does uh today either in public transit or in a private car

or in Uber uh why wouldn't you do it you do it every single time so I imagine that then what you do in cities you have all these AI Services you have public transit for physical movement you should build new housing differently so we are looking at 3D printing housing or using low carbon cement both strategies we're working on and they are economic they're economic because so much of the cost of housing is in labor and customization either you make them straight boxes and uninteresting places to live not pleasant places or you 3D print them

where complexity comes for free so uh I'm I'm Meandering off but I I imagine a different kind of city but it evolves there in bits and pieces so you're you're invested in lots of things in the materiality in in making well dematerializing the world is so important I forget the number there's couple of tons of material every one of you uses every year it's tens of tons I forget the exact number it just sounds ridiculous each one of you should we have a planet like that no we shouldn't but we've got time it's F to

work challenging how do we get we got one question up on the screen somebody put that question up before we got in the room it's a question that I think you get asked quite a lot so we've got this uh future city we've got this utopian ideal of services but now we haven't got lawyers we haven't got people working in the city we don't have people driving us um what are people doing right you've often spoken about this about the redistribution of wealth that comes around this and that's where the question comes in you know

we're going to disrupt traditional economic models here and I think that's a good thing right disruption that's what you're all about uh um but how does that work in in terms of uh income equality and let's talk about this every one of you has Healthcare which is whether you pay for it or your employer pays for it there's a fairly large percentage of your monthly income if Physicians were free or near free it's hugely deflationary on healthcare costs in fact we even have self-driving software for MRI machines and ultrasound machines you know much easier to

do self-driving for MRI machines than for cars and in and the throughput increases and the cost decreases so AI will generally be deflationary when you're talking about things like Services which is a big part of our economy where does that lead us it leads to and I have a 2016 piece I did in Fortune Magazine it was a long piece about 5,000 words on AI will cause great abundance great GDP growth great productivity growth almost everything Economist measure any economists in the room uh thank God uh but increasing income inequality that's the first time I

called for the need for Universal basic income but also the fact that if you do the math of changing GDP growth from 2% to 4% or 6% and just two weekends ago I uh I spent a weekend with a series a big economics cohort of people interest in AI talking about how AI will redefine economic systems um I would say to you we will so accelerate GDP growth much of what we think of as unaffordable will become una will become affordable and we will allow for um for redistribution because of that uh I would say

a couple of other things people have to realize capitalism is by permission of democracy people vote to elect people either on the Socialist side or the capitalist side and if a system isn't working for a lot of people it'll be voted out my personal view capitalism is the most economically efficient system but we may move to where we optimize capitalism not just for economic efficiency but also for equity and I think that'll be the change we do and that change will come from policy you don't like forecasts and forecasters um but you you make an

awful lot of predictions which you should do in a moonshot game how long does this change take is this in the next decade two decades is it dependent on where we are in the world and if so are there countries that will get there quicker than us because of you know their lack of baggage well so everybody has baggage of some sort every country does um I worry so there's a lot of reasons this world I'm talking about may not happen if the AMA prevents doctors AI from practicing that'll slow things down right um if

people worried about Job losses with robot robots that'll slow things down so one can learn to dis distribute these benefits more evenly or I'm afraid in countries like China you can use tanaman Square tactics to affect change um this will be a social Choice it's a very hard set of Dynamics to prove but but I do think uh different company countries will adapt it different rates yeah of course and you mentioned and I worry about this yeah how do you think the US is in terms of this or which cities are the ones that get

get there we're we're here in Austin I flew in it's not a huge City it looks like it could adopt all these things a lot quicker than other places I've been to I I think it's it's a cultural thing in cities um some some cities are more Progressive than others with respect not in a political sense but in a sense of adapting to change and adapting new ideas um so it'll happen but the nice thing is some cities will serve as role models for other cities uh and and some will be leaders some will be

followers but uh hard to predict who can I take us down the route of uh of Technology uh um we we mentioned Glide ways uh um Mark's here in the front row he was just in a panel next door with some other people talking about uh uh a a future with no traffic jams uh um you you're invested in robotics in Transportation uh um in in lots of different things uh um going into the technology that's needed for this the intelligent infrastructures that required for this going back you know to your your history back there

with sun and Jun are we kind of like at a paradigm shift of technology is there more that's needed than we have now now because it seems to me there's quite a lot of fragility around so if we start laying you know overlaying all these Technologies you know can what we have sustain this or do we need to invest a lot more in the infrastructure you know every time you have a lar of change especially a technology change it changes assumptions it solves some problems but it creates others and we have to acknowledge when it

creates problems you know social media is an example the debate around Banning Tik Tok is one example of problems created by social media you know it is funny we still debate whether we should ban it or not but Tik Tok is banned in China they have a very controlled version of that app in China it's amazing we see no press dialogue that what runs in China is not Tik Tock it's okay for the rest of the population but within China Tik Tok is much more about behavior science techn ology the things they want to educate

younger people on not about videos and it has all that but it's a different version of the app and so we let them feed something different to our kids than they will to their own kids um and then they argue it's not controlled by the Chinese so talking about China you know I I I was listening to a comment earlier that China has 3 million 5G nodes yet there's only 100,000 deployed in the US um so there's some catching up to do from this infrastructure that's going to run our cities you've often spoken about you

know you don't like the institutional way of doing things is there a change that has to happen in order to promote this investment from an Institutional level in in the infrastructures that we need yeah you know markets Drive infrastructure investment and when the incumbents don't change a new player will do it uh so I'm I'm less worried about that but I do want to go back to your original question change makes some things better and makes create some problems and then the usual approach is you go address the problems uh they always pop up mostly

they unanticipated I don't think anybody anticipated 20 years ago the level of issues with social media or that it would endanger democracy uh so and AI agents may do the same I call the biggest danger of AI is puru of AI Agents from Bad nation states doing individual dialogue with our voters in 2024 now that's like a huge risk that I worry about so there are dangers we have to take care of them and handle them so AI I mean everybody in the room wants to talk about AI you've been in a recent uh uh

you've you've come from the I like the vision of a new kind of City you like the vision of a new kind of city but the new kind of city has new and AI services for the city and services that people need made much more EX accessible I want to give you an example we just ran a study we have a mental health company uh and they provide mental health mostly in the UK uh but they're just starting in the US it's an AI based therapist in the NHS system which is the only dominant Heth

health system in the UK with the study of 100,000 patients they found diversity and inclusion improved 25% by doing nothing just letting the AI therapist be an AI therapist that's pretty stunning you know we don't solve it by doing something special and then other people resent that specialness you do it by Building Systems that are objective well in the UK you've got a very different Market haven't you with a National Health Service they need to find efficiencies cuz it cost too much so AI don't we want cheaper mental health here like who doesn't want cheaper

mental health here yeah yeah but the system sometimes is more difficult to work within right so I'm looking back you you mentioned the paper um the essay Reinventing social infrastructure with technology that you wrote whilst hiking back in 2018 um in it you wrote about the tremendous power of AI right that was going to change the world and this was before you took a position in open AI with Sam Orman and before everything that's happened over the last 18 months have your views of this tremendous power of AI radically changed in that time or is

it still that forecast is you know the same you know the thing with technology pretty far forward you can there's two ways of changing the world one is extrapolate the past and and that that's what most experts do and then you don't achieve any radical change you achieve incremental change and we see incremental change in society all the time or you can invent the future you want and then as an entrepreneur mostly drive that change through and alanas did a great job with electric cars I don't believe we'd be in electric vehicle Paradigm today if

he hadn't driven it almost gone bankrupt a couple of times and if he had gone bankrupt we wouldn't have electric cars the traditional automakers would laugh at him and say see what happens when you try and do something consumers don't want um so my view is it depends on entrepreneurs to drive like large change whether it's to what how you buy products or or cities public transit it doesn't matter um 2018 was the year we invest in both open a and nobody thought AI was going to be important but also in Fusion we invested in

conv Fusion systems at the same time and we looked at public transit roughly around the same time saying what are all the things that are worth doing where there's a reasonable technical p to do because some things are almost impossible you know if you told me to eliminate cancer today uh I don't have a clear path in my head though I have one I'm attempting for cancer also uh it's radically different uh but you take these large risks which means you increase dramatically the probability you fail and it's okay if you're trying to change the

world to fail I don't mind a 10% CH a 90% chance of failing if there's 10% chance of bringing Fusion to the world or new public transit or new housing or new AI That's sort of the approach I take and I think without that kind of an approach we will not see large changes and we won't solve the big problems and when you make these large changes there'll be side effects that you'll have to go back and mop up and fix like we need to in social media today are you surprised by the speed so

you know open AI came onto came out with chap GPT in November 2022 and actually there's this AI hype cycle that's going on it's absolutely everywhere and it's coming upon us very quickly is that hype is it real is it much quicker than you envision is it going to bring further Innovations to us quicker uh what has happened is first the progress on algorithms was faster in their capability Than People anticipated what that has done is created a self-reinforcing cycle because a lot of very very smart people now want to work on AI which will

accelerate the rate of innovation it's a self-fulfilling prophecy when the smartest people start working in an area you'll see more progress in that area um so I do think it'll move keep moving very very fast people will over expect things from it in the short run but long term it's going to be much more impactful than most people believe even today okay and there's a lot of talk about uh um you know the advancement of of General AI and there's a lot of fear around that as well does that come even closer can you talk

a bit about that because that's this uh you know this General AI is going to be in this city of the future right well so let's talk about different kinds of fears there's one kind of fear uh could you produce an infinite amount of music for example absolutely can there's music model that do that and I tell a fun story I'm like probably in the 1% bottom 1% of musical Talent on the planet personally uh and my daughter got married last year and I wrote down my thoughts I entered into chat DVD and said turn

it into rap lyrics and then I entered into music Ai and it wrapped the music for me suddenly I was able to rap in a very personal setting with very personal thoughts uh and that's capability that would be beautiful to have it's much better than Hallmark cards did you upload your own voice to it yeah you haven't got a rrap prepared for us today no okay uh but I may have a second career it was good enough um if I need to but it'll increase the amount of Freedom people have to do things but our

music our labels worried about a lot proliferation of Music yes they are very worried about do people want it absolutely why do I say that just go to Spotify and say what percentage of the music streams are independent labels as opposed to major labels it is up to 40 some per now is Indie labels or independent music which tells me consumers want more of it even without the AI so many more musicians get a chance to be on Spotify and have their streams listen to and I think AI will only dramatically enhance that um we

have ai doing anime that's pretty cool anime is so so hard frame by frame uh people love it but it's expensive becomes much more accessible uh we have a book publisher that's AI driven book publisher that's proven it can do books that have 10 times higher probability of achieving a certain level of success people say how much does this book produce in Revenue $100,000 a million dollars more uh it can do better than traditional book publishers so all of media will change and all of it for the benefit of of the consumer but there's a

fear with the consumer right I'm just thinking that you there have been people like so that's one kind of fear yeah where artists might be fearful I actually think the best artists will get much more creative will leverage it much better and many more people will be able to be creative so that's one kind of fear or a doctor worried about will my job go away there's a different kind of fear which is will the sentient AI take off and hurt Humanity I don't I I don't say I don't worry about it it's a possibility

but it's a small risk compared to much bigger basket of risks Humanity faces you an asteroid could hit us in the next year or decade or something that we don't know we know that's happened before we know China is the bigger risk I worry about that as old a magnitude bigger risk than sentient a so I look at the basket of risks and say we should appropriately allocate resources to addressing those risks and sentient AI should be an area of funding at us universities and universities all over the world on how do we manage those

systems I do think there will be many ways to manage those systems um so you're a moon shter though right you're going all the way over there but we're governed by people that don't have moonshot kind of like uh perspectives I I agree that's there's so many overwhelm one of the reason this story has gotten so much play is because it makes for great press headlines whether they're true or not they tie into our belief about science fiction and it gets way more played than it should yeah the benefits are overwhelming as we've spoken about

right so you have to balance things out you have to balance benefits you have to balance out the risks like China interfering in our elections uh that's a much bigger risk that i' would worry much more about what about regulation people talk about the regulation of AI and I was coming over on the plane and I'd read that uh I did I scanned through a paper on the European type regulations which seemed very well balanced but what's your view on that you know if we regulate too much we don't know what we are regulating or

what the capabilities are we start regulating we will kill Innovation and we will kill most of the benefits and we will give our enemies a free hand just like Tik Tock China has one version internally they'll have a different version outside so we can't lose the economic race in the race for State ofthe art AI against somebody like China now I'm very focused on the geopolitical side of this and I think it's the real and maybe the biggest danger of AI it's interesting uh they've got intelligent infrastructure in China but not necessarily access to the

uh microprocesses and things that we have in the US how's that playing out at the moment this you know well long term they'll catch up but if they're behind we'll hopefully get better and better how technologically catching up in this kind of like World Isn't that quicker and quicker and quicker is catching up six months six years it is it's a real concern uh and some of the western Tendencies like open source increase China's ability to catch up but back in the day uh um with some Microsystems you kind of like you disrupted the technology

scene at the time right an AI is new today uh the internet was new back then and you kind of like bet on tcpip and you kind of like fundament nobody believed in tcpip for the public networks in 1996 nobody Cisco didn't every major Telco in the United States told me they would never use tcpip the internet is based on that and you're part of a phenomenal company that took market share and grew and then you went on to Juniper and all of these things how do you see that the moment you know in AI

first move advantage that goes on you know at the moment there's a lot going on with Nvidia amds coming on how do you see that kind of like play out are you seeing lots of other exciting things happening out there as well from a technology perspective in what areas in microprocessors in the Network Technology you know a lot is being tried with microprocessors um you know what's called a GPU and everybody now knows that phrase is basically one or two simple functions multiplying efficiently and accumulating so multiply and add uh is the main function and

they've tuned these chips to mostly do this function and increase throughput of what are called multiply accumulate uh functions there are other ways to do it we have an attempt at doing it in analog Computing may or may not work I know some reasonable efforts to do it optically you can compute optically too it' be a lot less power consumption uh and hopefully you can manage that much more power creates a lot of problems including heat and reliability and all that so my view is innovation will increase there's two parts to AI Hardware one is

training models and the other is using the models so what's called inferencing uh I think inferencing will be much much easier to address with alternative architectures um and then we'll go to training it'll be a comparative space in five years from now can we you spoke about power there and there's a lot in the Press I know we have a lot of questions I'll let you yeah we have we kind of like we're going through some of them I think they they're covered off um just quickly just about uh uh uh um Power and energy

you invested in in Fusion in Commonwealth uh Fusion Technologies and when I look started looking on your weite had all sorts and manners of things to do with energy from you know generators and long-term batteries and this um this city we do lots of fun stuff mostly break based on technology breakthroughs and but this city of the future is going to have to have like abundant energy as well oh absolutely where are we on on that kind of like journey to get into anergy I'm now convinced uh solar and wind are great um up to

a certain percentage of our poverty requirements but you can have a bad month and and no wind or no solar so we have to have reliable power I do believe Fusion will be that source of power um there's another source of power that isn't talked about M much which is super hot geothermal I'm quite bullish about super hot geothermal the most people don't talk about it as a power source there's very few parts of the world where where you can't dig deep enough of a hole and not get free heat energy to produce steam to

run generate power uh what's the problem uh the problem has been when you drill to hot temperatures drill bits just melt you drill to four or 500° the drill bit doesn't work you're talking about the uh in my school Atlas the picture of the volcano yep the magma coming up that one yeah yeah so uh those those are solvable technical problems and there's many ways to approach them um and we are attempting many of those so it could be a source of power one of my dreams is there's a coal plant that powers the Congress

that iconic building and I want to drill right below it and produce enough St to eliminate the coal coming in it'd be like just such an illustrative big ambitious thing to try and get done and then it can be repeated in most of the western United States heat isn't that far deep and so most of those plants whether they're coal plants or natural gas plants can be replaced by heat below the power plant of course they already have the energy infrastructure there sounds like a simple idea it just needs one big breakthrough in drilling are

you invested any companies oh yeah absolutely okay watch this space so we have got we've got a couple of questions up here right some we've covered uh um we've got one here at figure a company released a video yesterday featuring their figure one which can have a full conversation with people are you going to have an AI assistant anytime soon VOD uh so let me give you the history of robotics so far you build a robot for a purpose and you program the robot to do one thing uh and it does that well assembly assemble

cars on a GM assembly line those kind of things what the new effort and figure is one of those uh though it's a hardware play there's a number of efforts that I'm very optimistic about we said we took language and we built a general learning model like gp4 or GPT 5 or chat GPT or Gemini we took videos and we built models that can build Sora like videos in the physical world if you collect enough data you can build Foundation models for robotics that AR tied to a particular piece of Hardware you can run g

chat GPD on any hardware you're right so I think that'll become a very interesting area for development the next 5 years and many people will attempt uh what are called Foundation models for robotics which is foundation models for the embodied world the physical world because software hasn't gone into the physical world as much uh at least not in a general way and these will be learning systems so you take a robot trained to do one thing put them in a completely different environment like the surface of the Moon and learn the environment and be able

to do things that's pretty exciting in robotics figure is one of the pieces of Hardware uh that people are building and there's others 1X and hod or others every version of that Hardware exists there's a cool robot from China called unitary making them Dirt Cheap in China um and they will collect the data that we need to train the foundation models and this will be again a a cycle of more robots more data more learning the robots do better they can do more things so we'll see that cycle it'll be an exciting cycle the last

next five or 10 years but keep in mind even a self-driving car is a robot in the physical world we've been talking about self-driving cars for a long long time it's been the uh the pinup poster hasn't it of Technology uh and by what you've been saying today we're kind of like going to frog leap that or where do we go next self- driving so self-driving cars were done in a world which was highly orchestrated and so they had to program a lot of things in and some areas you know self-driving car can kill a

person you can't take the risk of it's okay to kill some people as long as we're learning to drive better that doesn't quite work at least not in the US though somebody in China told me one of the leading AI guys told it to win the a self-driving car race China will allow that it's sort of sad but that was a direct quote from somebody in the AI field um and China is already the largest exporter of electric vehicles in the world now not an area they had any standing in uh in the traditional internal

combustion engine world but uh I I do think robotics will be very interesting they'll be learning systems and when they learn enough they'll get into homes they might you know you already have have a dumb robot like the Roomba Vacuum but he'll be able to walk upstairs or you know put dishes in the dishwasher there was there was actually a robot I thought they were going to keep it for us for the 1:00 session there was a full stand up robot yeah here's the difference all these systems are really expensive and and they can do

certain tasks when you produce in a high enough scale uh you you get to much lower costs much more data and much more intelligence so I I was just actually coming up the elevator and I said you know it's interesting to me that an elevator can cost as much as a car though it's 5% of the complexity of a car why cars are produced in hundreds of millions of units and one of my forecasts is in 25 years we'll have a billion bipedal robots that's one of my forecasts by the way those will do more

work than all of humanity does in physical labor today think about a world where that happens but we have to get to scale to make him affordable yeah we can build A1 or $200,000 robot that does a few things in the home and Sony's had this AA robot that maybe is tens of thousands of dollars but it does almost nothing uh it has to make coste effective sense in the world in do real things it will happen in the home but I think it'll happen in the home after it happens in business applications we we've

got a another question from the person asked the first question which is taking us into a different world up into space that's something we could talk about where do space exploration intersects into a vision of an AI driven Utopia um you know so the nice thing you know we are investors in rocket lab so little company can democratize access to space they made it really cheap per kilogram or per pound to go to space SpaceX is now competing with them and dropping their prices for small loads space is getting a lot more accessible um so

there's there's a lot you can do in space in fact we just had the first launch of a I don't know even what to call it a manufacturing part that can make biologic drugs in space and we just prove we can re-enter and recover drugs made in space that's pretty cool uh because you can do things that you can't do with uh in on in gravity uh in drug manufacturing there things you can do in Communications like certain kinds of optical fiber can be much easier made in zero gravity than with much better properties so

coming back to data infrastructure um you know if you're trying to go from New York to London with one cable it's sort of a pain to put connectors at the Bott botom of the ocean and so it'd be nice to have fiber much higher quality fiber that can Spain much larger lens and much more fiber so there are many applications uh AI in space which is the specific question uh there are a few applications that need AI in space and I was just talking to somebody given the houti problem in off of Yemen uh why

we couldn't monitor constantly that space turns out you you need a lot of cameras if you need a lot of cameras you need to process a lot of data it's best done on processes in space or on balloons or some way to monitor I I I don't think it's a hard problem to monitor every square kilometer every square meter of of the ocean's surface and you know this one set of people the hooes are causing massive disruption in shipping enormous yeah we should be able to monitor it we should be able to use AI in

space to reduce the data and just say hey there's there's something in there moving in that direction that's all it needs to say that data is very transmissible transmission from space is a scarce resource and Power in space is a SP scarce re resource but it's possible solving that problem I didn't know you were invested in rocket Labs I didn't know that oh yeah we started when it was uh five people or something small I put a small investment it just because it's got my name in it so I just thought you know yeah inves

in things that say rocket we we have a rocket a satellite somewhere in space that says coasta Ventures on the fairing that's very cool it's sort of cool um but that's an example and then there's things to be done between space and here we have the coolest weather balloon company and the plan is to collect more weather data than all of the world's weather data collection uh and we can float 10,000 balloons over the planet and really predict data accuracy they're already using AI techniques not traditional weather modeling techniques which divides the atmosphere into grids

and tries to compute in a Brute Force way but using deep learning models they already have with the balloons they have the most accurate weather forecast in the world when you listen to your weather forecast at night on the news or look it look it up on Google they have about 3-day accuracy that they have now extended to five or six days but this will keep improving proving and they want to 100x the number of balloons they have so that's pretty cool while they're doing that they can monitor every aircraft in the air anywhere and

maybe even every ship anywhere in the ocean like all these are possibilities and you know we working on all those and it's a lot of fun there's what we haven't got much time left there by the way a lot more fun than playing golf I'd be bored playing [Laughter] golf all we got one more question up that's come up here uh um and it's interesting and probably more preent right now someone's asked what's the future of AI and decision sciences and how will an AI CEO look like will you be working with AI Founders anytime

soon you know somebody has attempted to do that uh look the general rule should be anything humans can do AI will be able to do better the question is how much much agency and this is not a technical question this is a social question society's one a hand to Ai and I believe that's a choice we will make and we will keep making it and keep changing it as we learn more and get more Comfort but there isn't one specific answer well we got to trust in the future haven't we yes know as a last

parting comment uh um where do you think we'll be in the next few years how quickly is this rolling out just leave us with that sound bit and then we can get on out Andis thing is I think we'll see a lot of progress and some amount of busting of the hype of ai ai will underperform in the next 3 years it will o overperform in the next 10 years compared to what we expect today and we should keep that in mind it's just the nature of human beings to get too excited get on band

wagons uh and and real technology and real applications take time that's a really great place uh uh to end this um Veno Costa thank you very much for sharing so much thank you [Music] everybody