AI Eats The World | Jeremy Howard, Vivienne Ming, Peter Diamandis | SU Global Summit

[Music] [Music] I think it is an amazing tool for affecting the world but I got to leave it somewhere in the five range because ultimately it takes people it's humans problems that need human solutions but as a tool there is whether we're talking about deep neural networks as they exist right now or next-generation technologies coming out there are problems that are not otherwise tackle at scale without alo and we'll come back to this no more and I'm I'm actually giving you guys an unfair request of putting it on one scale because it's probably two but Jeremy would you come out on that scale mmm some people some parts of the world - ten some parts of the world some people plus ten in fact a person you introduced me to Marshall brain wrote a great book about this called mana showing hell both both results can happen in the same world so it's interesting I'll quote my dear friend Neil Jacob Stein who is we'll be hearing from Neil Todd says it's not artificial intelligence I'm worried about its human stupidity yes yeah it's particularly human groups people stupidity it human groups keeping me pretty particularly when they're angry stressed and and in difficult situations so I want to come back to this a little bit more but let me take a second and I'd love you in a capsule each to talk about what are you how are using AI so let's begin you're a Berk you're a Berkeley now does my company and what you're doing yeah so again I started my life as a theoretical neuroscientist which is really where the deep neural networks emerges out of from our field but about ten years ago became an entrepreneur and started thinking about how I could take what I knew about brains and about machine learning and affect the world and it may sound crazy and wildly over-ambitious but I would distill down what I do across many different companies is I want to very literally make better people so my core company so coast is focused in education but very different than most people might think of it in line with some of your last slides it's education whose purpose is to produce happy healthy productive lives it doesn't the specifics factorizing a polynomial or knowing some particular fact about the world that's important it is are we producing people that are meaningfully impacting the world but I do personal work in health technology inventing treatments cures detectors I do work in really rethinking hard things like human capital you know I think both of us would probably agree that one of the biggest open questions about AI is how it's going to affect things like the future of work about professional and and non professional work and these are huge open questions and I get to be a mad scientist and just look and see how AI as augmentive intelligence really complain at all in that space and just to drill down one second for soco's your product there are you so in our with our product is jocose our core product is essentially a multi AI driven system a quick narrative pick my daughter up from school she's got so many pictures she's drawn that they all just very quickly end up in the compost bin so we built a little a I it analyzes that I see see her grandmother and at the same time it comes to muse what we call it and we get an analysis of her artwork it analyzes my son's speech patterns when he talks to me we combine that with a whole bunch of top-down knowledge about the things we know a predictive of life outcomes comes out of a very rich literature love to telephone about it and from that we the worst thing we could do is say here is what the crystal ball says about the life outcomes of your three-year-old as soon as you provide that that profile you've ruined everything positive or negative bad outcomes so instead we do something very simple and we can do it all via SMS we simply provide an activity every night for the parent to do with their kids the promise that it's the best way you're going to spend twenty minutes with them every night so I can go and sign up for so close you can go sign up we most of our accounts we give away philanthropically we always appreciate it for the parents that sign up for the full app version because pay for everyone in the rest of the world are being South Africa twice in coming weeks hoping to develop new programs there and the townships so when you can deliver take a fancy AI but put a front end that's just an SMS you can reach the people that actually need these interventions as long as you're pairing with people on the ground Thank You Jeremy I've watched you through a few incarnations the company I met you with Kaggle was recently acquired by Google where you were president actually you were you were sort of like if you guys know Kaggle it was a sort of group of smart AI experts I would compete against each other there's a half a million yeah to solve problems and Jeremy was like that like the top performer out of half a million and Kaggle sort of reached in and pulled him out and made him president of Cagle but what's happened since I well actually will mention the idea right at the start of this conference of using AI to diagnose disease so I actually chaired the company that kind of invented that it's called analytic the focus is at the moment on radiological data right the idea is that there's a shortage of about 10 X - 20 X the number of doctors we need in the world it's going to take according to the World Economic Forum 300 years to fill that gap through training so we're going to fill that gap through AI so not replacing the doctors but make them vastly more productive as I built that company I realized that the basic techniques we were using Vivian mentioned deep neural networks which is really what this is all about can be used for everything from disaster resilience to improving education to improving crop yields and I kind of you know also know this through my su work that there's all these grand challenges which we can help with this technology so since that time I've also built a company which is trying to educate everybody you know I mean everybody so that they can go back to their organizations and use this technology to help whatever they're doing whether it be counting cars from satellite images to make bundles of money in the hedge fund through to improving crop yields and yeah I don't care so we've had over a hundred thousand people now go through this course of all walks of life and they're applying this isn't so the name of you company is called fast AI it's a good name and anybody can go and do the course for free and literally people have gone through the course with you know we've only started coding two years ago there's one now is a Google brain resident if you've seen Silicon Valley the TV show that not hot dog app was built by one of our students lots of amazing so how many people have gone through fast at AI over a hundred thousand great amazing and so mi mi as an outside user looking at fast saw a I to educate myself and my employees or am I looking at it as a pool of people I can hire out of the the FOMA and I really think that it's better to educate you and your employees because you understand the problems you're solving you know you shouldn't be hiring in external vendors or relying on external software or hiring in some machine learning PhD much better to get the people in your organization who understand your strategy your constraints your data and upskill them that's that's our strong belief and what's the experience is it many video courses is it yeah it's an online course with these amazing interactive notebooks where you basically become an experimenter in AI you type stuff in and you see results and you learn through trying things out and neural nets or machine learning yeah neural nets which is a type of machine yeah as you know but neural nets specifically are the thing which is causing machine learning now to go exponential yes that wasn't happening before we were held back by some limitations of technical limitations the algorithms actually neural nets turned out to not only have no limitations but they scale exponentially with data and computation yeah the idea of having people do this internally because I think one of the real limitations right now with AI isn't on the technical side I mean there are some fundamental limitations there but we're not really pushing those boundaries it's really in the infrastructure which includes how people are implementing it inside their companies if you don't really understand what's possible then there was a Sloane paper that just came out recently saying listen you know largely you're just doing this very easy optimization let's say resource planning optimization or ad targeting when there are huge new opportunities existing inside your companies you just aren't seeing them because you're thinking how is this just a classic tool it's like a really smart spreadsheet and she's gonna solve my existing problems faster and better when really you need to think well how can this completely restructure my company where I don't have people pulling a bunch of levers I have people solving problems about the unknown leveraging this sort of tools that I exactly into Vivian's point I think it's it's you know my analogy here would be like the Internet squared right so if you were selling shoes in the 80s and 90s and you heard about the internet if you like shoes internet probably not relevant and now South must have come along and in your lunch and Amazon's eating every retailer's lunch the impact of deep learning to organizations is going to be much much greater and much much wider so not only do you need to understand the capabilities but also the limitations so for example people are increasingly looking to use machine learning and even neural networks for things like predictive policing and in the justice system it turns out that the data that they're training these things on is terribly biased due to all kinds of social and historical situations at least in the US and as a result it's quite possible for people to end up creating AI based systems which embed the bias that lives inside people so there are huge issues which is why it's really important you know again to Vivian's point to understand the technology so that you can use it but not misuse it how many folks here have actually done any kind of AI machine learning related coding can you raise your hand so it's relatively a small number probably 1% how long is it take for a person to go through your course it takes seven weeks but in the how much time each week are you putting it well let me answer slightly different quest because in the new version we're going to be having releasing in the end of October it'll be two lines of code to create a state-of-the-art image recognition system so the fall into anything or take you about 70 hours but within the first 20 minutes you'll know how to create a literally custom state-of-the-art image classification tool so I'm curious how many folks here would be open to investing 10 hours a week for seven weeks yeah or at least 20 minutes 20 minutes a day what lays 20 minutes to get started or absolute 70 hours there's an appetite for that in the room here just to get a sense so well you know what to come fast I you know yeah there you go it's free from our free marking pitch for real product I think if you if you think about this in the context the goal here isn't for you to be writing production code of complex AI systems it's rather think of what may be the future of medicine could be so I developed a system that was mentioned earlier it predicts my son's blood glucose levels about an hour into the future I developed another that predicts manic episodes and bipolar sufferers weeks into the future in both of those cases it doesn't replace endocrinologists and and neuropsychologists but of course it totally transforms what their jobs might be they need to understand where systems like this might actually fail where they would need to intervene but at the same time it phenomenally empowers them and to becoming some hybrid of a medical doctor and a data scientist and really then their job becomes the human side appart a truly personalized medical plan which just is not how it gets handled right but but these like incredibly advanced algorithms like although until recently they've taken hundreds of thousands of lines of code to produce and it's more likely you'll screw it up then get it right because there's so many details to get right it is getting a lot easier you know and as I say we've got it down to two lines of code now and I think it looks a lot like anybody who used the Internet in the early 90s or a member editing PPTP dot-com files and setting up sliff and like you have to be an expert and write all this code to connect to the internet now you know I think deep learning and AI is going to more like you know a mobile phone internet connection it'll just be something that we use anyway one of the things that was interesting is I came when I was in China just two weeks ago with my my a 360 community I was meeting with a number of entrepreneurs there's probably more AI related to startups coming out of China than any place on the planet now one of the other things was - Lee who that's one of the investors and my planetary resources you'll hear about later and was the president of Google China before it got shut down has been investing in in chipsets a kid you not that are basically five dollar chipsets to enable anything to have machine learning capability on it so that we're heading towards a world in the near future where everything has some level of AI I mean every kid's toy will learn yeah it looks like the next Apple iPhone based on some leaks seems likely to have a a custom processor embedded in it and apples also recently released some libraries that allow Google's tensorflow AI system to be built into Apple iPhone apps so that's happening very quickly so what is it what is it well that what will that look like when everything starts to be AI enabled just you know extrapolate on the words if you want so I you know Google's making some interesting progress on this also with what it calls federated learning so imagine with these sorts of chips built into your phones it does all of this deep neural network learning there and then only transmit the models back so it's not necessarily sharing all of your personal data but I think thinking about a smartphone or smart car even smart lightbulbs while interesting is not really seeing the big picture of where we're taking this which is really thinking about all this devices as a big distributed robot exploring the space so if I'm Google or Apple or 10 cent or Baidu I'm not thinking about owning a platform with a whole bunch of devices on it I'm thinking about a single entity that's exploring the world that gets to every one of them is a separate sent on now we're back to a ie to the world good well we're getting there you know I don't want to overplay what's capable so don't think that this is I'm describing an intelligent thing on in our sense but I am talking about a very explicit and intentional system that if you are simply thinking what can this sensor collect on this phone and report it back to me for analytics as opposed to an active what is called a reinforcement learning model for example it's actively probing the world and trying to learn and every one of you then become part of this sensor system actually we're we're building something very much like that I'm helping build a company called dr. day I actually dr. AI gets clattering with Deloitte actually yeah and we're actually shortly to release a blockchain based system the dream of just what you see with medical data where you can join up with other parents whose kids have illnesses like yours share your medical data in an encrypted way through the blockchain and then that can then be accompanied with a offer of a reward to data scientists to solve that and so it's all done on the edge devices on the on the phones is basically where your medical data is held and so we're teaming up with a number of big medical organizations to have make it possible for their patients to suck their data into their phones and then through this blockchain system connect it to everybody else who has similar medical issues and then use these kind of edge devices in that way it's really cool it's exciting I have to say when I did this with my son I had to hike all of his devices and it turned out break several federal laws in doing it which I understand they wouldn't want it to be easy for people to hack medical devices but he essentially is a cyborg now he's got live intelligent devices all over to him and what makes it really exciting is with the system we built they essentially communicate with one another so this is something that's also different than a traditional platform plays there's no black box server this stuff all has to go to because they have local processing capacity and they're exchanging information their own little mesh network then suddenly this this little thing which is my son is smarter it's smart unto itself when you combine other families and there are a number of different plays and specific illness groups who are talking about or in education you know what truly works best for a kid like this right now who has these strengths in these guys I mean imagine for example your your kid has an illness at an early stage another family has contributed data from the same illness a year later after some intervention enough of this and you can actually figure out what intervention is going to work for you or your family the one thing about it though is particularly in a world like this it being really easy to think that the right intervention is the one we learned in the Bay Area and then you take that to some kid in Nairobi and it fails miserably for a hundred different reasons because it's a different people in a different place and you have essentially inherited the bias without realizing it I think that can't be true although you know with deep neural networks as long as you've collected the right data you should be I'm not saying it's a fundamental problem it's a human problem so let me take you back to three news items the past two weeks making sort of sensational headline news the end I'll ask you for comments first is Facebook shuts down AI because it learns its own language okay second one is you know open a eyes system beats top dota 2 video game players and and third one is AI is more risky than North Korean nuclear war so so in order yeah please kind of okay thank you uh well I lovely I'd love to add some honest particularly that first one about the chat can we just so I wanted to hit this for you because as as as an su representative you in one sense our representatives out there to the world and there is such a level of amygdala ignited I hype about AI and we fear that which we don't understand can I give you four names please there are only four people I'm aware of you should listen to who write about AI so I'm going to tell you who they are one is Dave Kirsch gone who writes for courts one is from array owl who writes for Forbes and top BOTS one is Jack Clark who recently joined open AI and has a great newsletter and then Tom Semin ate that wired those four did not not only do they not echo that's article but they all quickly wrote Corrections so the problem is most journalists have not no ability to spot the but here are four people who who can so follow them when you look at some of the I mean there are some modestly successful AI companies out there whose whole strategy is on o generates a stories to sway people's political opinions now if that's a profitable business I wouldn't be surprised if just writing that I mean just writing the headlines to grab your attention is is also a really profitable business and it's not surprising then that they took this route but it is I saw headline just you know Vivian Ming has cyborg son that just came out yes of course it's more dangerous than North Korea yes my son is Skynet and I find that fun now but you know the thing is like you don't hear their attractions right so Peter mentioned this dota thing so for those who don't know Dodos like the biggest computer game in the world or one of them multi-million dollar prize money and open AI which is his overhyped over Research Lab did a pretty cool piece of research where they created this very very very specific spot that played this particular version of this particular game in a particular way with a number of particular kind of cheats and beat somebody good and suddenly that was in the news and the alum mask you know blah blah blah the next day 50 people beat it now nobody wrote about the 50 people that beta let alone the fact it wasn't this incredibly one ever heard that part yeah heard that so that's you know what's the point I want to make is that AI is extraordinarily powerful it's going to transform every aspect of our lives it's something we need to be aware of if you're not incorporating some version of AI into your company you're not going to survive in the long run because you're not going to compete in the long run but at the same time it's understanding the appropriate way to appropriately use it and not to create a fear-mongering about it I think either direction sort of techne utopian vision that if you just add enough processing power human problems just magically disappear human problems take humans to solve them ai is just an amazing tool to augment that power I will also add a little bit of additional nuance with regard to the Facebook story the vast majority of machine learning models that have ever been trained have been thrown away because they don't do what you want them to do these things for months to get them to work right and they don't and you scream at the command prompt and you get rid of them because it turns out they incredibly fragile and nuanced creatures the most robust problems are the easiest to solve so so in this case it isn't shocking that in the course of trying to a bunch of bots and let's speed up the training process because there's a shortage of data well have them talk to each other right now that they were trying to get it to speak English and it failed to speak English so they went oh we screwed it up and they stuff exactly so you flush it cuz it didn't do what you want to do not because it was about to get the launch codes but here's another slightly nuanced one I will admit this is purely this sort of inventors arrogance thing here one of the things that annoys me most about the dota story was the first headline I read about it Elon Musk's AI beats world's best players and I gotta say if everyone that worked on that model read that a headline and thought I don't remember Elon being there working on this till 3:00 in the morning every night you know it's nice when the funders and I mean the same thing when I hear DARPA invents or any of this stuff in fact a woman invented that at open AI and she is not out there going about it yeah so it's you know there's a lot it is hard work making these things actually do anything useful once they're able to they can have truly amazing impacts on things but I think it's appropriate to appreciate their limitations but also to I don't know why the fear-mongering is the the thing that is really scratching this sort of itch in our nucleus accumbens this kind of reward signal here the writer you should be anymore afraid of an AI in the hands of to current world leaders or sorry atomic weapons in the hands of to current world leaders that are duking it out in words versus an AI doing it once AI one of my companies we actually predicted who people should hire recruiters give you five seconds they look at your name your school in your last job our system look at fifty five thousand variables based on one hundred twenty two million people and a report of the New York Times said what would you which would you prefer for your kids and I said my system isn't meant to hire people it's meant to augment recruiters but if I had to choose I would choose mine as that first pass versus all of the known by in the original system but you did but you did mention a very real threat earlier which we should be scared about which is the future of jobs one like okay killer robots isn't happening at least not for a long time but killer people might well and like we do now have very direct evidence that the automation of industries is is destroying jobs and they are not coming back yeah and like people say like oh well with all these AI things surely you're going to need lots of data scientists know I am very much in the process of building automatic data scientists we don't need lot of data scientists either we don't need lots of people to manage the automatic truck fleets you know maybe one for 10 to 100 trucks so as these jobs disappear it should be abundance right it should be great only if we find ways to allocate resources in a way that's not based on labor or capital input so this this could dispute you and I have had this conversation those you who follow my work you know I write about this politically that it's not the Terminator sequence and AI and robotics I'm concerned about it is the near-term loss of jobs it's not even the loss of jobs we we can predict which jobs we're going to lose we cannot predict which jobs we're going to create number one number two it's the rate of job loss that is the issue and so when I talk about you know building a bridge to abundance it's how do we how do we properly think about and span the next 20 years right because it's this period of time so I'm actually partnering with Tony Robbins and every January at a 360 we're doing a half-day program on we calling it Bridge to abundance the future of work and I know s you through Gary and Bowles are gonna be doing a lot of work on this cuz I think it's one of the one of the signature issues of our day that every one of us need to be thinking about and understanding how do we deal with that because it's going to be very real and there's going to be a public outcry I call it you know people people firebombing the Google buses because they want to get their anger at somebody we're taking their weight we're already seeing the tip of the iceberg of what anger driven politics looks like you know when people get scared and they get displaced and they don't feel able to do anything about it they get angry and there are 300 million guns in this country largely held by the people who are going to get the most angry so like it this is a very real issue is not can you imagine the utopia where we don't have to work and we can do whatever we like Peters already done that for us but I don't think any of us believe that we've figured out that this bridge it's the bridge to get there yeah yeah so I I have a book coming out called how to rub up proof your kids say again how to rub up proof your kids and great and it's really in three sections one is a discussion of what AI is in the practical terms of how it will impact us and the other is the problem statement of what happens right we have to completely rethink how we think about education and jobs I don't mean what skills should kids know now or how do we rescale a factory worker that's that's not even the right way to think about this at all we need different people Gallup estimates there's maybe 130 million people worldwide that are actually engaged with their work they find meaning and value in it they care about it now I'm it's wrong to think that AI takes jobs it's much more than it does specific tasks but as those tasks get automated there are a lot of jobs out there that might be easier to hire a lab tech floor instead of a doctor to hire essentially just a pair of legs and an arm to walk an AI around the room rather than a professional and that's going to have profound impacts if you're not above that line where you are doing engaged as I say frequently sort of creative problem-solving then I don't really need you that much not as a professional that is this I need decision makers I need problem identifiers I don't need people doing rote work yeah and unfortunately 130 million people may sound like a but in the global scheme of things that's a rounding error great great points so I 30 seconds to close each Jeremy I think that the main thing I want to focus on is what you guys can and should do about it although behind this I mean in fact behind the scenes these algorithms that complex they are a bit faster yes to get right but we're in the process of or you know optimizing or automating that but for you guys to use it in your organizations in your jobs if you're in government use it in policy it's not that hard you know and if if you don't actually engage with learning this now you're gonna be like the folks who ignored the internet until it was too late I mean oh wow you've taught me this Jeremy I mean Jeremy sat down one day and spent a day with me I don't know some part of a decade ago going through and teaching me about the basics and it's like every one of you have extraordinary amount of data companies don't realize what you have is assets your assets you know include your cash your customers your your and ultimately what people don't realize is your data is amazing asset you have right and I thought so many people think Oh we'll just outsource it to some big vendor and you can't add source such a fundamental piece of strategy like think about Google right Google are basically a data products company can you imagine if they were outsourcing data strategy they wouldn't exist so you need to obscure yourselves and obscure your people and believe in your people ái skills are not born on trees they're developed in also it's a it's a way of thinking it's a way of we can figure out in that data what's the algorithm what's the gold Maya and promote and retain people based on their ability to understand data not just on their ability to understand your domain vivianne a 30-second quip and then I'm gonna go to yeah I'll try cut it even less than that I strongly agree but I say to nuance it even further it's that knowledge base that you have about your business in your problem space that you can't currently at least outsource that to an AI even if you wanted to and you certainly don't want to outsource it to a big player so you you need to bring something to bear on these problems I have yet to start a company in which I didn't show up feeling phenomenally ignorant instead of applying neural networks from day one I spent months and months learning everything I could about the market and the problem space yes you're already experts don't sell that out to companies that already want to eat your lunch yeah you are an expert in in some problem and it's that expertise in that problem that differentiates you ok so if we bring the lights up a little bit and and we have mic runners we have mics what do we have i Peter mark Donahue from abundance good to see you good to see you too a question about where you've seen really innovative models of human interaction with what I call and Peter does intelligence assistance versus a ice instead of replacing the human how do we accelerate them and make them merge with technology to the greatest potential yeah so so many examples I'll throw out one which Jeremy knows about my company human longevity right we look at full-body MRIs and it just takes you know it most of the imaging 99.9% is normal and you're looking for that 0.1% what you really want to do is have the AI look at and find the issues and then have the human look at the you know the 1% or sub 1 percent so medicine a lot of that is medicine it's also how analytic works in exactly that way and also we're shortly going to be releasing a beta of another company called platform but AI which is specifically designed to you get a lot of good done AI names yeah yeah I invested well in them basically it's a platform to allow humans to kind of interact with the machine learning algorithm in a way that the machine learning algorithm tells it tells them what it needs to know and the human tells the machine learning algorithm what what they can provide and you try to get this kind of optimal combination one example I'm sure there many others I'm pushing into the true far end of this my lone remaining areas of purely academic research are neural prosthetics particularly cognitive neuroprosthetics and so literally increasing people's working memory span I have a company in New York that is working on something that very few others are which is synchronous team with EEG and other neuro indicators where we're actively looking at the indicators of trust safety of communication quality so that we can short-circuit the amount of time it takes a diverse team to become a high-performance team so I can a group into flow together yes exactly but and really leveraging that one of the co-founders was the head of DARPA for their neurofeedback sniper training program awesome next question who's got the microphone hello hi hi Peter yeah I'm right here okay gotcha hi my name is cos teeka's PAHO I'm not Greek I'm from Albania I came like your father 20 years ago with my family I have a question I've been dying to ask you it's a it's a two-part question so if our selves right they always replicate and information goes from cell to cell it's a form of a copying consciousness so if consciousness is copied to artificial intelligence isn't that the same at the end of the day isn't it still consciousness whether it's a machine a robot isn't it like a human transcending into that and it doesn't really matter what happens to us eventually and then the second one is if we could have a matrix like world simulation where you can tell the difference between the real world and that and you can bring anyone in there the you want would you guys accept that so to the first question can I get the lights up in the audience so I couldn't see ya so to the first question I think the starting point is and I say this with some recognition of the research in the field nobody is actually invented artificial general intelligence a lot of people researching it it is not a thing that exists in the world and we have noble reason to believe it's about to be yeah I only don't think it will grow directly out of deep neural networks for example it's the wrong sort of technology concept for this so the questions of I love the science fiction sort of idea of thinking and preparing for a world in which we might need to make decisions about that but I think it's better for us to focus on what the technology is now there is a massive untapped spread of that technology so it's truly fascinating ideas but it's not I don't want anyone a mistake that artificial general intelligence is a thing and that in the near future it will do any of the things we might imagine all of this right now is a distraction from the bigger societal issues the vivian's talking about I think thank you next question and we're on the verge of quantum computing how is that going to affect AI and how do you see the future of AI based on the quantum computing coming up is a convergent in any way okay so quantum computing yeah no one really knows the current quantum computers that kind of exists on quantum computers at all the ones from d-wave do something basically called simulated annealing really quickly we don't know at all that they can do anything faster than computers or that they ever will there are some companies that are working on true quantum computers but on the whole the ability of quantum computers to speed things up in general is massively over-promised we actually don't have any reason to believe that they're going to make a huge difference to things like deep learning algorithms they might but at this stage there's certainly no reason to assume that I think there might be a sort of a metaphorical angle here which is people think quantum acuity computing and maybe that has some additional almost magical element that it adds into it really the whole deal here is how long does it take to train a model right when I did my dissertation it took me a month to get in trivial models to converge now you can do these amazingly complex systems relatively fast on GPUs and big scale systems that may not sound like much but it's a transformation if you get to test new ideas in the market over the course of a couple of hours of thinking about a problem instead of waiting months for a single model to converge and give you its its output totally transforms the kind of market there's a lot of cloud where improvements around but this one that's commonly said that if you add one cubit you double the amount of computing performance it actually turns out for various technical reasons that's not remotely true so so I'll just I'll repeat what my friend Chad Righetti I don't know if you know if you know Chad and Righetti computing I had him on stage with me last year at 8360 talking about computational power and particularly quantum computing and he is he does believe that that he's building what is a I agree with the d-wave is not a true quantum computer but what he is building right now he believes is a real quantum computer and he does believe it will have immediate implications for machine learning so I guess the question is will we'll find out projection either from Google or IBM or Righetti computing or whatever's going on in China which is not that's actually not a company that's doing something with accelerating deep learning algorithms by using optical directly so rather than having like silicon chips with electricity going through it's literally speed of light optical stuff and it does it using basically photonics mm-hmm there's a lot of interesting areas going on the interesting to see which one but the thing that's interesting to close this session out on is I hope you're taking away from this there is so much going on and a lot of these things are influencing each other and one one tipping point in one area you know it may be that you know deep neural networks enable us to design a better quantum computer and then that quantum Peter enables do so it's it's the uncertainty and it's the multitude of things that are converging in unexpected ways over this next decade that make it really an amazing time to be alive please welcome and thank Vivian and Jeremy [Applause] [Music] [Music] you

Channel: Singularity University Summits

Views: 15,901

Rating: 4.7241378 out of 5

Keywords: SU, Global Summit, Singularity, Singularity University, San Francisco, Peter Diamandis, Jeremy Howard, Vivienne Ming, Silicon Valley, tech, technology, future, humanity, prosperity, abundance, machine learning, ai, robotics, blockchain, virtual reality, earth, planet, nature, performance, artificial intelligence

Id: dhlvfIzhwHs

Length: 42min 13sec (2533 seconds)

Published: Thu Oct 05 2017