Bay Area AI: David Silver Interview
Recording: Bay Area AI: David Silver Interview
hello I'm Alexa carproof the father and organizer by air AI and we have the first self-driving cars meet up here at galvanize which is data science education a place with the university program on data science and it's very fitting that we have with us here David silver team lead for audacity is open source of course on some of their cars thanks for having me here upload so this is super exciting I must tell you we just heard you talk which will be required to publish a function TV for everybody to see go free next to this Q&A it was super impressive who had a giant crowd mobile you forever after the middle right and I don't think we often see this rare so this is super popular topic so can you tell us a little bit about how you got into this and how it became so cool yeah yeah thanks and it was a really great event thank you for putting it on we had a lots and lots of folks who are just really excited about the topic and stayed for a lot longer than then I thought they might because it's a weeknight and folks are busy so I got involved in autonomous vehicles after working for many years as a software engineer most recently in ruby on rails as a web engineer and I decided I wanted to work on a newer more exciting topic place with a lot of growth and I got really really excited about self-driving cars and the potential for self-driving cars to to change the world and be just a really fun place to work to work in and the technology really captured my imagination and so I took a number of online courses at Udacity I'm before I worked at Udacity I was a student also at other online education companies that have other great classes and I did a bunch of independent projects I'd built my own little mini self-driving car I started a blog that I still write to this day about self-driving cars and and I knocked on a lot of doors and I was fortunate to be able to claw my way onto the autonomous vehicle team at Ford's research and innovation center in Palo Alto working for Ford Motor Company and Ford is a great place to work I just loved working for Ford it was amazing to work with all of these terrific engineers building autonomous vehicles and then Udacity came and asked me about how I might think of building a self-driving car training program to kind of help other students do what I had done myself to to go from being just a normal software engineer to being an autonomous vehicle engineer and so for the last five months now I've been working with Udacity to to help build out our self driving car engineer program awesome so I you know I am reading your video book console.write karzai because a lot of subscribers because all the current information so can I tell me you obviously work a lot on this where do you find all this pieces of knowledge and how do you decide what's most important to include there yeah yeah so there's just so much news in this field every day it's it's such an exploding area a lot of it is simply paying attention to like what's going on in the news both in kind of the startup ecosystem here in Silicon Valley but also in the major automotive centers of the world so Detroit Germany Japan there's a lot of really interesting stuff I'm going on that I I try and keep up to date on and a lot of it I I have thoughts about and I write about and then a fair amount of what we do now is is actually comes from our students we have a super energetic slack community at Udacity an online community using a tool called slack to a for our students to communicate with each other and they say all sorts of interesting things they have really smart opinions about things I hadn't even thought about in about areas that of the autonomous vehicle yuko system that that I wasn't as familiar with is our students we have you know 6,000 students in our slack channel right now and so I get a lot of good material for them yeah I read a lot but they give me a lot of good material on there a lot of funds are awesome tool sales maybe I should outsource myself to a too deep neural network yes so so we went 2-4 and I was really fascinated that you mentioned some tension sushi existence of arrancars another additional physical off kind of the classic online internet startup industry and one you mention that it's it's a it's kind of big tech companies vs karbonn facturers and then kind of detroit vs Silicon Valley and then imagine car suppliers you shouldn't even think about so can you talk a little bit about kind of this you know tensions and like who's gonna be at the power things who's gonna win the autonomous vehicle racing that's yeah that's probably an open question but there are a lot of interesting things going on in the autonomous vehicle world right now so you mentioned you know one big competitive point in the landscape is big tech companies yes Google Apple uber you know our building ride-sharing services and traditionally auto manufacturers like Ford where I work but also Toyota mercedes-benz BMW GM they have traditionally focused on building cars yes but they see a future where the world is about more than just buying cars or their world is about more than just buying cars where it's about mobility and transportation as a service so they are starting to move into the turf of companies like uber and lyft and potentially Google as google launches their ride sharing services so that's a really interesting dynamic and suppliers come into play there too because you know a company like Delphi which is a huge automotive supplier has announced that they're going to launch self-driving taxi service in Singapore and then you can imagine that might just be a precursor to to to other things they might do in the future right right as you know there's a lot of kind of you know turmoil in this ecosystem there's a lot of opportunity for companies to to move into different spheres of influence and folks who you know have competitive relationships in some areas may have cooperative relationships in this new area or vice versa all right you know its interest because in after mentioned Delphi and I can realize that this big question for suppliers right so it sounds like if your supplier you kind of need to get to think about control of the brain controls the brain controls the body and so then I connected to Bosh right I'm going to become connected world and every just connected vehicle so actual exchanged speakers so I'm going to speak it connected the world and bring some AI spirit of the valley there and they're going to feel the speaker two separate cars conference which without doing in in in March and we just emotion thrown the CEO fetus is the keynote all right and actually so thanks to you by the way it's kind of business utilized a lot of different things right thank you Aereo outcome and we'll have a little bunch of more things and kind of you know we really hope to see you at this conference right but but I've kind of it got me thinking right it's almost sounds like multiple comment is not realized that whoever controls smart technology the brains will control the whole industry how much ado of the thing is true right like basically is it key for borscht or Delphi to master the eye we should all senses of their cars or you know because they do hardware no it is not gonna be metal and plastic and right and rubber like you still somebody has still cos gonna do this how do how much do you think the balance will shift into computer versus kind of sensors parts you know the kind of material think we should go into the car yeah that's a great question i think it's it has shifted a lot in the last 10 years and that's only going to accelerate in the future from from hardware to software so if you think about you know how the automotive supply chain has traditionally worked there are auto manufacturers and they a large part of what they do is assemble components that they've sourced from what are called tier one supplier suppliers who supply directly to auto manufacturers and the tier 1 suppliers a lot of what they do is assemble components that have been sourced into your to suppliers and so everyone's kind of assembling things all the way up the chain and what companies are understanding more and more now is how valuable the data is because deep learning and machine learning rely so much on training these big models with lots of data over time and so leaving that hugely valuable part of the technical stack to a supplier really leaves you at the mercy of of you know somebody else's machine learning algorithm and it takes a long time to ramp up your own and so you're seeing a lot of tier 1 suppliers bring their a software in-house you're seeing a lot of me auto manufacturers start to bring their software and houses they figure out who's going to be where a lot of companies kind of look to the mobile phone world is kind of how the past played out so you know mobile phone manufacturers like Samsung and an LG you have have seen their business become a little bit difficult because of how much of that ecosystem Google controls through Android and a lot of companies in the auto space are nervous about simply becoming hardware manufacturers that someone else's software controls yes you know there are done struck me so good friend of mine won't work for NVIDIA and they have a partnership Mercedes I'll eventually have their arrangement jersey transit or whether right which is interesting so it's really struck me that this guy seems to didn't get the memo that Google of the top dog they didn't seem to like here so what's really struck me like if you're a small startup in Silicon Valley you're gonna try to do web scale search now you're not even trying to do you know deplorable images to the extent that Google does it seems like this current guise of fearless they just don't care or they think they shouldn't care they just for Chicago that machine learning so in VT is making huge progress and you know according to kind of rumors they can leapfrog the bunch of stuff and you know you see forward and out and all these guys just do it on their own and I think like 100 Z mention that they should because it's kind of who controls the brain but it seems like they really are not afraid of these top dogs in machine learning then you know the facebook and Google's and Microsoft's why do you think is this is this because I just don't didn't hear about them but they don't have a choice and they just have to compete with them fat all yeah I mean New Jersey's a long way from California they may not have gotten the memo no I mean the nvidia hooks are great they're a huge partner for our program and they really are fearless and at the at the forefront of machine learning I mean I think what you see in terms of little startups like auto and bigger companies like like Ford competing with tech companies like Google what you see is there's not really a winner in this industry yet um there are folks who are further ahead than others but there's nobody who's so far out front that it just seems like they've got the game wrapped up and there's a lot of value at stake a lot of customers at stake a lot of jobs at stake and so this isn't something where I think most companies are willing to shrug and say we're too far behind we're not going to give it a shot I think you know the games in play and and different people are entering the arena and it's not too late if you want to start your own little startup you know there's a big market out there she's really cool vol that's politically right yeah you know Lisbon to the coma hey I question right so it's like it's been fantastic an amazing company right then this is like here's the guy who correct the iphone you can or anything like it sounds like he's garage he like hooks up some wires of the collar he can control it and then k kind of receives this note from the government and seems that this rocks of opens oh so what do you think is kind of their all the government so that that you know test already at a pilot kill the person you know if this is just a fault or not we don't know fully but basically i think the expectations the guard will come into this marketplace the carnival regulate and kind of the como a story shows that it's already starting to do this how much do you think we should kind of as an industry we should care about the government and how do you think this is going to affect all these developments you know how quickly do you think they start off to be regulated the garments s and how how was it a changed industry yeah I think you I think the government is unavoidable in this industry I think you know there are folks who would rather just be able to do all of this without any government intervention and whether that's right or wrong I think it's probably not realistic we're going to have to everyone who works in the industry is going to have to work with regulators and already is working with regulators I think the things that folks working in this industry have going for them are that a lot of people are dying on the roads right now I mean the real problem with self-driving cars is not that you know they crash a little bit too much it's that human drivers are really awful um the biggest accidents with self-driving cars have come from humans hitting self-driving cars correct and so it sounds in the way yeah and so I think there's a lot of room for improvement in saving lives and in the government at least at the federal level seems to be Foursquare behind that and really want to push this there also some I think really important groups that will benefit from autonomous vehicles disabled people the elderly this will be a huge game changer for the way they live their lives and I think that's a big reason for us as a society to really get behind autonomous vehicles as a way to improve people's lives so I think the government will be there um but i think you know we'll were on the side of right and and i think most governmental entities recognize that i think maybe the biggest challenge right now is simply in the u.s. getting all of the various governmental entities on the same page so we have 50 states plus a federal government and in state some states there are multiple entities that regulate vehicles at the federal level there are a few different entities that regulate vehicles and so it can be easy to kind of died a death of a thousand cuts from all the different regulatory agencies you have to comply with and I think trying to get everybody on to a unified system the way if I go to the DMV I get a driver's license in California that is valid anywhere in the US i think if we can do that for autonomous vehicles it'll be a big wins actually next to my fellow patient I didn't think about the stage about living sorel because I was thinking about the lifter right like measuring a lot of us are fabulous and you know your wife yourself we drop off your kids and your kind of go to the same route and you have a left turn right and it's kind of one day you're not alert in connection so I was thinking you know one thing question we didn't talk much today about this but you know its self-driving cars took to each other mm-hmm second negotiate right so I think that's actually a very easy way to solve the problem because if car spending course it who goes first because of left turn and prolific term problem very easily because you know it's kind of major source of accidents but you know the cars will have to negotiate so maybe at some point Monica's cars will have transponders of some sort and talk to each other right but the dimension of the states and different sets of different left turn rules right so so how far do you think is the situation where basically the car is equipped with transponders can can be on the road right because that will definitely tell other cars that this car is here and you can negotiate laughter is you can negotiate there's traffic merges right basically it will simplify the whole thing much much better yeah i mean that's that's kind of the dream right is to be able to drive at 300 miles an hour and a self-driving car that never has to stop at stop lights because it can communicate with all the other vehicles around it um I think that's probably quite a ways off on well I wouldn't say unfortunately but I think self-driving cars autonomous vehicles need to learn to navigate the world the way it exists we're not gonna yank every other driver off the road simply to make room for autonomous vehicles hopefully eventually they become ubiquitous enough that that they're kind of the de facto standard if only because there's so much safer yes um we should we save oughta wash thread the blades have to have to be equipped with responders right and you know the collision was couple a very rare but I think that's industry figured out that that's the way to do it yeah it's injured there's a lot for autonomous vehicle engine is to learn from how aeronautical engineering and and aircraft manufacturing has developed an auto pilots and planes I think you we will see some vehicle to vehicle communication you know within the next five years it will probably start out between vehicles of the same manufacturer for relatively limited functionality and it will I would imagine a de facto standard will emerge I can see a world maybe in which there are self-driving car lanes kind of the way we have HOV lanes or the way we have bike paths for bikes um some people think that that one ever happened maybe not it's hard to get infrastructure built in the u.s. right now but I can see that world happening but I think a world in which you walk out into the street in san francisco and there aren't any human drivers on the road I think that's a long way yes and so I guess the wall professional desk is Udacity is obviously live by sebastian thrun who was a pioneer in the cell drive cars and in other multiple other points like Coursera or election right but but that was the first to do the self-driving car open source which is really amazing so let's say you know I want to take another course as you guys have an auntie currently I why should I take AI emotional courses at Udacity if I want to be sober kerosene versus other guys what what does them go how's it going for sale yeah yeah so i can say something i think relatively free of bias which is that you dasa t has the best online course platform and i can say this free bias because i was a student not all that long ago in taking courses on all the different courts platforms and there are great courses on all the different platforms but you dasa d somewhat uniquely um builds all of its own content and Udacity has a really strong focus on pedagogy on how to teach that something that that I didn't really appreciate until I became to ask the employee was just how much work goes into systematizing the way you'd ask you develop score um I knew nasty does a really great job helping people not just watch videos but then actually do code so are our founder Sebastian Thrun who's both a luminary in the educational world and the autonomous vehicle world sebastian has the saying that you don't lose weight by watching other people exercise unfortunately one for me to take to heart and and so what you know what we do is we put students through a lot of quizzes a lot of programming projects and that boat helps them learn and internalize what we're teaching and also bill gives them a great portfolio of projects to take to employers and and ultimately transition to a job in the machine learning world the artificial intelligence world or the autonomous vehicle self-driving car world so I know you guys have actually a car for a course right we should applaud kuya photo and I hope to get this guy's a lot of stuff to saw regards conference so when I was measuring to a friend of mine he basically said you know if your bicycle have the code which is controlling the car and can kill you maybe you should only be able to deploy it to the car when you are in the car this is true but can you cannot tell me this is true or not and basically you really have a car on the road should should be kind of a good rule of thumb right like you know if you if you won't really like be able to crash car into things if this should be the cause who gives me in the car well one thing is at the end of the course as you mentioned students will be able to put their code in a real car that we have students will do this in teams of about 30 students each they'll all work together to get a car to drive around the test track so we'd have to fit 30 students in the car um that that's tight it's a it's a lincoln MKZ which is a really nice spacious car but 30 is a lot so I don't know we'll get all the students in the car at the same time but it will be on a test track we were very safety-conscious there are really uh well defined guidelines in the state of California um that Sebastian Thrun actually laid out when he was running Google's self-driving car project so we're kind of adhering to Sebastian's Google rules now that you'd asset E and so safety is paramount here we're not going to do anything for educational purposes that puts anybody's safety in jeopardy so the only put the guy on the God who committed the call may be just the commit or maybe the project manager well this is great though this is great this is this is really exciting you know and I have seen people on Twitter tweeting about being in this car so I really want to see this car and I you know that amount of stuff I invited me to see so I'm really really excited don't think there's been any course like this or you deploy to go to the actual car right i think that's that's a game changer yeah we talked about it is in only at Udacity experience we are really focusing at Udacity on the type of educational experiences that only a company like Udacity can offer and this is something that as you said isn't available anywhere else we we went to a lot of work to try and make this happen for students and we're really really excited about it cool so one thing I will ask you them kind of to wrap up so we're super fast changing industry heuer and deep inside our bill double this so I'll put on the spot you know completely about what where will you be in the year right so like a little so next to get one wise use of this myth up and I think stock of what happened right what do you think we'll see in November 2017 November 2017 man predictions are hard especially about the future um so I think what we'll see in the next year is a fair amount of progress in very select cities on self-driving cars so uber has self-driving cars driving around in Pittsburgh right now picking up passengers google has made some announcements about operating a ride-sharing service in San Francisco whether or not that will be done with self-driving cars i think is is not totally clear but i wouldn't be shocked given how ubiquitous their self driving cars are in Mountain View couple of companies Delphi an autonomy have announced launching self-driving car ride sharing services in in Singapore and so I think we're going to see self-driving cars than the next year kind of from the realm of purely internal projects that are only accessible to people who work at these companies to being in limited public or release where if you live in the right city and you you get lucky when you call for a car you'll get a self-driving one I think by next year we may not see self-driving cars all over the world but I think you will be able to ride in one if you really want to just go to the right city and you'll be able to do it so we're going to lucky we'll go to palo alto and some guy will go to Pittsburgh and they'll hopefully be picked up my car that's my dream for the next year all right the super exciting so we'll definitely look at me but I can probably use the Foley and we'll compare notes and also happens alright thanks so much for having me thanks Darrell appreciate this is you're welcome