Fireside Chat with Andrew Ng (Landing AI)

hello everybody hey nice to see all of you all right well thank you so much for joining us you seem like an awfully busy guy you know there are a few organizations a few courses a few startups that all kind of have your name with them so I'm curious starting things off where are you most operationally involved like what's your day-to-day look like um the rise of AI means that a lot of things we need to do to help and reach his full potential so these days I'm building spending most of his time at three organizations landing AI is helping large companies jumpstart our adoption so if you come from a large company and you want help freaking on AI that's when landing AI does AI fund is a start-up studio that creates new startups from scratch everything from coming out here it's a validation MVP hiring the initial team capitalizing along your own and then deep learning on AI is our educational home so a lot of individuals if you want to learn about AI learn the techniques of deep learning or if you're non-technical and want to understand what a is and how it may affect your business deep learning on AI produces HTML content often working of Coursera in order to teach people about AI I hope that with these three teams landing ai ai fund and deep learning AI we could do many pieces of the ecosystem we need to help AI reaches full potential so of those three orgs like when you wake up and you're looking at your to-do list which I imagine might be kind of long which of those organizations has the longest list a day after day you feel like um I'm pretty spending most of my time sharing a lot of brain cycles on landing AI right now which actually is one of AI funds portfolio companies but one of the things I'm excited about is I guess you know my background long ago sometimes I got build the Google brain team starting but the Google brain team which how Google become very good at AI it and then also built AI group that by do which hope I do become arguably China's greatest AI company but if you look at what we as my what milta a alchemy has done were transformed software internet companies companies like Google and Baidu and many others you know Facebook Microsoft and and another software companies I think the next phase for AI is to go and transform all of the other industries as well everything from manufacturing to agriculture to retail to healthcare and so that's what landing AI hopes to do so you left by do two years ago in that time what have the trends or kind of changes in the enterprise AI scene looked like I think it's all been up into there right okay if you look at the most recent um a I Index great great report um I think the number of jobs in deep learning increase I think 35 fold over a two-year period um and this is a reflection of companies wanting to hire more people in AI or deep learning and we see this in man as well with the deep learning courses we offer a 30 learned on Coursera or if you look at maybe a more questionable metric CV insights at this wonderful chart showing the number of mentions of AI by CEOs on earnings calls and the chart kind of looks like that and and I have no idea if the number of times cos mentioned AI on earning calls is indicative of anything could be a negative trend buzzwords toss them out there but I think maybe the one thing it does indicate is the rise in enterprise a man okay I services so you know you mentioned a few industries that landing I was focused on why is manufacturing such a big early focus um we started in manufacturing and I think you know one for example we do a lot of work in visual inspection where rather than using human eyes to check if a smart phone has a scratch or a compressed canister has a leak you could use computer vision and a learning algorithm to do that at greater reliability and this is hoping factories inspect many different things in fact objects with scratches or leaks and so on and improve using quality but beyond manufacturing hum doing a lot of work also in agriculture and healthcare and as one example I'm doing a lot of work of a agriculture machinery company to take traditional agricultural machines mill they were they were there's no not very small agricultural machines to try to put sensors on to them and optimize how they perform using machine learning so want to be cool oh and this is what we're working on if the same farmer same field but put in a smarter machine can optimize the planting and use of the same farm I think this can help drive farmers productivity you hope you know farmers that otherwise may not make that much money help them get a better living and also optimize the MDT the environmental effects of how we produce our culture so I mean they're obviously a lot of AI SAS companies out there that have been you know getting started why did you see the space and feel like you needed to you know Co found one yourself you know I am yeah yes one of the thing thing about AI is I don't think of AI as a lot of applications of AI can cannot be implemented as a SAS service where you swipe your credit card and then use it and then your company is now a high enable in fact one of the biggest misconceptions one of the biggest misunderstandings I see for a lot of seals for AI still is that the impact of AI on many industries is strategic rather than tactical and because they are still so new maybe I'll explain this using an older example the internet instead um there were a lot of SEOs that thought hey the Internet is a big thing those many years ago right five five years ago 10 years ago and ministers hey I got a web site I got this internet thing covered um but that's just was not true um if you look at companies that learn to use the internet properly for example UPS and FedEx is not that they build a website is that they realized a lot of the logistics data flow using the Internet and this is fundamental to the way they do business um or companies like Microsoft and Apple Microsoft and Apple were not internet companies they were found it operated made tons of money before there was even such a thing as a you know modern internet but they transformed the Internet company so the Internet has had a strategic impact on these companies ranging from you know things that FedEx UPS Apple Microsoft and I think AI will also have just have a strategic impact on many industries and what I mean is that the way you compete the way you create value in the future will be different than it is today to use 1:1 to an example look what uber did to the taxi industry the internet enabled that internet in mobile and so it wasn't that if you're a taxi company you build a website then you've got the internet covered no what happened was the Internet and mobile re-architected the nature of the industry the way you compete how you create value how you believe defensible business and I'm seeing early phases of that now with AI also coming in to have a strategic and Pat well when we architect in the nature of different businesses companies that embrace it they have a correct vision of the future and though the pieces in E will survive or even thrive and companies that think all I need is I need to build a website and now maybe my website needs a little you know tens of little plugins I can have whatever those companies many of those companies I think will perish um so I think I'm starting to see more CEOs get smart about this um but how we are we are protecting different industries is still being sorted out and I have a clear view of some industries in others if you're not you know a company that's investing in research heavily or you're not a tech Titan but you're a medium to large sized enterprise like is it worth building out your own AI engineering team or should you just be using one of these you know startup services yeah you know um for the launch enterprises is clear I think every large enterprise probably has the resources and must spend those resources to figure out AI from a strategic on the tactical point of view um the for the medium-sized companies it is it is a bit harder but I think that to the extent that AI could have a transformative in impact on how your industry is structured I think of this incumbent even for yo for many medium-sized companies CEOs and then by I'm not sure what medium-sized means I think of you a billion dollars and up you better figure this out if your dollars I know you maybe have only a little bit of resources to think through what is a I um you know recently met a company with about 2,000 employees and they told me that they were getting might have been overkill but they told me they were getting every single one of the 2,000 employees to take AI for everyone which teba entire office on Coursera because they wanted every one of the two thousand employees including you know everyone engineers product managers reception is moppets our recruiter to know enough to think through this might be overcome maybe not every company needs to do this but I thought wow this will make this company incredibly thoughtful about a hundred navigate into this future just think okay you know I think I think there have been a number of trends recently it seems like some of the AR startups today are relying pretty heavily on human labor as you know to train some of their systems do you think that this is a trend that's you know going to continue for the next few years just employing thousands of people that train us data is like is being a part-time human in the loop going to be like a career for a while yeah I think that yeah until you've been on the insides of an AI team it's hard to understand how insatiable our stupid learning algorithms are for data um so I think the need for massive numbers of Labor's will continue and and then quite likely even grow for the foreseeable future as AI continues to infect more industries where there isn't as much data um you know in terms the bones of a I want one example I think um oh the software community we spend a long time figuring out how to manage code right so we have you know really clever ideas like version control and it took us a long time to evolve from well is it CVS subversion to get um and we have solved a lot of software engineering processes like code reviews and scrum and agile I think the process is the managing data are so much less mature so how do you version J's are how do you share data how do you manage you know what do you keep in the cloud what do you run your labs are whatever your son mutators oh this is one of the things that you never do in academia but you do all the time from chance we just edit your test set right in academia you download the test set you suppose evaluate against it and publish the paper you don't mess with it but on practical production projects we find all the time that sometimes the tested is mislabeled or wrong we just go and edit the test set but how do you deal with that data and well the systematic processes for editing the test set um I think that the the majority of our tools for managing data is much lower than a majority of tools for managing and editing a code so I think that's another important direction that I think um if you go we'll go in where where do you feel like the next deep learning breakthrough is going to happen closely relate to the data one subset of deep learning I'm very excited about is the rise of small data or sometimes also called low data but a lot of deep learning was driven by the software consumer Internet companies you know companies like Google Baidu Facebook and a few others and consumer Internet companies have tons of data you know hundreds of millions or maybe billions of users and the general law data so the previous wave or the current wave of deep learning long the algorithms were designed for when you have massive amounts of data if you look in other industries for example agriculture machinery can be working with you know we could collect data from a dozen farms easily but there are not a hundred million farms then we could go to to collect data from so what is a learning algorithm that can work even if you have only data from you know a dozen or twenty iid farms um or manufacturing if you're trying to tell if a smartphone has a stretch you do not have a million pictures of stretch smartphones because fortunately no factory has manufactured a million scratch smartphones that need to be thrown away so can you get your algorithms to work only with ten images so one of the cutting edges of deep learning is almost as well I'm most excited about is arise a small data and due to advances in deep learning things like transfer learning one-shot learning future learning the implementation data generation I'm seeing I actually also sell supervised learning I'm seeing exciting progress on on this and one of the things that my teams at landing I've done is spent a lot of time thinking about small data because the ability to gather deep learning algorithm to work even with 10 training examples that allows us to break into a lot of new industries where you just do not have you know a million old you know heaven forbid 100 million data points are you gonna end up having to rely on again on human contractors kind of parsing through some of that stuff when you have even less training data um let's see we do do some labeling ourselves ok with some of contractors and some in-house of our own team ok oh and I think we do a lot of things that and and I think what you know one example we do a fair amount of transfer learning where we use various data centers learn representations and then transfer it but does it but often is the is the details of those algorithms that they make a huge difference to the final performance mm-hm so switching gears pretty heavily here um you worked at Google in the US and Baidu in China I'm curious of your thoughts on kind of the AI race between the two countries and what advantages you think that China might have um why are you asking me about China Vantage it's not American evangelist royal waffle we'll get there soon you know I think the I raises is a misnomer much as the rise of electricity benefits in all nations and I think today I'm very happy that Singapore has a great electric grid and the fact that people in Singapore kepta you know can keep the lights on all you know all the time it doesn't make us in the US any worse often maybe we can learn from how they run their electric grid they could learn from us I can always learn from you but actually I feel really privileged to be able to learn from many communities around the world I mean that very sincerely I think the China interesting companies are very close Amaka and very good at going to markets I think America still has an edge in basic research but I would say ready to people of both nations or maybe people of all nations I think we're at a point where all nations from all nations by the way fun fact um Jeff Dean tweeted out several months ago that we're now at 100 archive papers per day on AI and this volume of research is not generated by any one country is generated by the global community and all of us just learn from you know the whole world so when we're talking about learning you know from from everyone obviously China has a national strategy in terms of how its rolling out AI should you know other countries have a strategy like this I think that many countries should have a thoughtful national strategy and one things about AI is it's still so immature that there's no room for many nations to have a big role in this future AI Empire world that we're building frankly one of the mistakes we made with the last technological disruption the rising internet was um I think we created tremendous wealth but we also contributed to wealth inequality and all the wealth we could create it was concentrated you know in Silicon Valley and in Beijing one of the things I hope to see where the rise of AI is for um we worked really long well that's very very clear but for this wealth to be fairly shared and I think part of that means making sure that we have many centers of AI around the world rather than you concentrated primarily in two places so I hope that many nations around the world will have thoughtful a AI strategies so that they can build something don't powerfully our capabilities focusing on maybe solving the national problems where I think many countries can be competitive when can can can do unique work that are harder to do is looking Guardian pitching it is supporting their own local industries where are you looking to invest I know that you were in Colombia recently um so in addition to our California team we also have a team in Colombia as one of the places that I think has a lot of momentum in AI and I'm optimistic about building there but more broadly I was in Taiwan last week where I was visiting a nonprofit AI lab and seeing new models of innovation and and I was in London back in April and I think I think that there'll be many opportunities for a lot of countries to build out their teams I'm very excited by the Colombia team but I would encourage all you know all all nations all nations with some resources to think through their strategy and how to invest so a quick quick aside on the AI fund are you actually investing through that you know oh yes so the piece of ecosystem right lani I house large companies become good at ai ai fund those AI startups from scratch so we're sort of studio and right now we are building 10 companies in parallel to to our publicly now which are landing a I was portfolio VI funds as well as a warp our digital mental healthcare company run by my friend Allison Darcy plus eight others that haven't been announced yet but that AI fund we on our team and our friends and partners brainstorm many startup ideas and then our associates team will select ideas to work on we're very good at falsifying ideas with them doing ideas where you know we were working on it for two weeks and then code the ideas they're very efficient at falsifying ideas and is usually only after the idea has reached relatively high level of conviction where we've validated customer demand um kind of maybe built an MVP validated that it's technically feasible that we will then bring on a permanent team founding team you know CEO and and then also in jak capital we will fund the company in order to have it have it go forward one of the myths we tell in Silicon Valley um is that whenever there's a disruptive technology is always the startups that win and that's just not true we're the rising internet there were startups like Google Facebook Amazon back in the day that then well they're also incumbents like Microsoft and Apple that did very well so I think these two pieces AI fund and landing AI tried to do these two complementary things where we take advantage of this very disruptive technology to both go after opportunities that are more efficiently done as a start-up and go after opportunities that are more freshly done by hoping large enterprises transform to transform to come I am how it interesting so wrapping stuff up you know I'm personally interested in this obviously us consumers and Chinese consumers kind of have different expectations of user privacy and you know after after running an AI program in China you know how did your thoughts on privacy shift um I think that the gap in expectations is narrowing I thought that uh recently you know things in China changed so quickly is wonderful and sometimes very terrifying but recently um there were a few data breaches in China that were publicized and this is actually causing many Chinese consumers have a greater desire for privacy so I think some things we see in the news it changes just in a matter of you know weeks or months I think the gap in expectation is privacy while there is still a gap has been narrowing somewhat do you think of myself between working at Google and Baidu your you know thoughts on your own privacy pretty much maintained oh I yes I boy not maybe that is I am very careful with my personal data of privacy okay oh and I do many things that I know many other people not know many other people do not do with my own data are you are you hiding your faces from the facial recognition cameras actually dad I don't care about that what do you care about um I think that without naming a compass I think I put a lot of thought into which other companies that have accessed my data and make sure to you know use encryption in all the places where it seems prudent to do so my laptop does have that little you know sticker thing over the webcam and and and I think I think yeah I think I try to be very thoughtful about you know what are the digital devices in my environment and what if they get what if something happens to one of them and how to make sure all that reasonable amounts of security and privacy is made um yeah right now okay great hey well thanks so much for the conversation appreciate it thank you sir thank you guys [Applause]

Channel: TechCrunch

Views: 17,587

Keywords: tech, techcrunch, technology, newest technology, hottest technology, brand new tech, gadgets, technology gadgets, hottest gadgets 2019, 2019 tech picks, tech top picks, current news, hacker news, latest technology news, cool gadgets, enterprise, enterprise products, techcrunch enterprise, tcenterprise2019, enterprise19

Id: ZsHJlHj_-Io

Length: 22min 18sec (1338 seconds)

Published: Fri Sep 06 2019