On behalf of the school thank you for joining us today from wherever you are first I hope you're all well I know the ongoing pandemic has certainly been stressful for all of us and our friends and families but with the roll out of vaccines which is an impressive scientific and engineering accomplishment I’m also optimistic that we will gather again in person sooner rather than later our annual Distinguished Speaker Series was made possible through a generous gift from UCLA alumni Ronald and Valerie Sugar this ongoing series has brought leaders from some of the world's most prominent companies to our UCLA Samueli audience I believe Valerie is here with us today and maybe Ron as well thank you for joining us and thank you very much so very much for your support we're in for a real treat with today's program for UCLA students here today your first interaction with Facebook might have been your parents sharing much to your dismay probably your baby pictures and elementary school photos with their friends or videos from a family trip not even 20 years old Facebook now lies at the center of our interconnected social galaxy it has helped many of us stay in touch and connect with friends communities and organizations and no one knows all the technologies involved at Facebook better than our distinguished guest today Mike Schroepfer Facebook's Chief Technology Officer as CTO Mike leads the development of the technology and teams that connect billions of people around the world and advance the state of the art in fields like artificial intelligence and virtual reality Mike joined Facebook in 2008 previously he was Vice President of Engineering at Mozilla Corporation where he led the global and open product development process for the Firefox browser prior to that he was a Distinguished Engineer at Sun Microsystems which acquired a company he founded CenterRun in 2003 he began his career working at start-ups including a digital effects company where he developed software used in major motion pictures Mike holds bachelor's and master's degrees in computer science from Stanford University Mike welcome to UCLA thank you super excited to be here sorry I can't be there in person but I can't wait to hear the questions that's terrific before we hear from Mike before we hear from you though Mike let me also introduce our moderator today Professor John Villasenor Professor Villasenor holds UCLA faculty appointments in electrical engineering law public policy and management he is the director of the UCLA Institute for Technology Law and Policy which is affiliated with both the school of engineering and the law school his work at the intersection of these areas makes them the perfect person to lead this conversation we will have a chance for questions from the audience toward the end of the program so please use the Q&A function on your screen to submit your questions I will now hand this over to John take it away John well thank you very much Hal I really appreciate the introduction and let's let's get right to it here so I'll start out with some questions that I've got myself and then we've got a couple of engineering students who are going to be joining us and asking some questions and then we'll open it more open it up more generally to members of the audience so I guess the first question I have is just tell us a bit more about yourself and what inspired you growing up and did you always know you wanted to pursue a career in technology I didn't I mean I think it's a common myth that people sort of you know in my position often you know knew it from the very beginning I always liked science and technology and math and I wasn't exactly sure what I wanted to be when I grew up and that's part of why I you know went to a university that had lots of great departments and it really wasn't until my freshman year that I I and hopefully no offense to the electrical engineering department I I took the Intro EE and the Intro CS classes to decide what you know actually first I thought it was going to be physics and then I was like maybe it's EE or CS and I took both classes and sort of you know fell in love with computer science that freshman year and that's what sort of really embarked my my journey in technology but I think growing up you know fascination about the few what the future could be like whether it's through science fiction or just watching what's happening sort of in the world was definitely something that captivated me even if I wasn't totally sure exactly what my role in it would be and can you tell us a little bit about your decision to join Facebook I mean you joined Facebook in 2008 which was sort of a very long time ago in digital terms and you've obviously seen just an extraordinary amount of growth over the time period but what led you to join Facebook you know back then in the first place yeah I mean it's you know it's it's a it's as you say it is feels like forever ago in technology terms I mean this was a time when you know the iPhone had just come out and the app store was around the corner and you know there were less than 100 million people on Facebook you know not the not the billions we see today and you know the at the time I was at Mozilla and I was you know leading the development of the Firefox web browser and and other projects there and that was I mean a tremendous experience a lot of fun still have a lot of friendships from back then and I wasn't really you know looking for something active but when Facebook contacted me you know I said well I'll talk to them and see what it's like and I think two things really stood out for me you know one is the people I met were all really brilliant really hard working and was a crew of people I thought that I could really learn a lot from and then the second was you know there was just so much opportunity in the product if you just looked at how people were using it and and you asked the question you know is this filling a sort of a fundamental need in society or is it some niche specialized thing which is it's a challenge in early technologies and something I look for to this day and you say well you know everyone on the planet wants to keep in better touch with friends and family that is not a thing unique to college students or to Americans or young people it's sort of a universal need and that meant that there was tremendous opportunity and upside and where it could go I didn't I honestly can't say I would have predicted it is where it is today but I thought there was an opportunity for a lot of a lot of growth and you know for a lot of the students watching this today I think that's this often underlies the advice I give people which is you know look for opportunities with tremendous possibility and growth and look for colleagues and co-workers who will challenge and inspire you so much of your growth will come from the peers that you work with rather than the specifics of what you do and if you select a you know good natured and hardworking peer set that's going to do wonders for you in terms of your personal growth and what's it like working as the Chief Technology Officer for you know company as with you know is well known and impactful as Facebook like what is you know what's what is what's your typical day like look like what do you do yeah I don't really have a typical day and you know I think it's changed dramatically you know I've been there almost 13 years and I've I sort of led engineering from the beginning and and you know I think of it as in a couple of big phases you know in the early phases when I was first there a lot of it was just keeping the site scaling and running was sort of job number one you know millions of new people were signing up for the service the sort of underlying software and hardware technologies to build a social network weren't really well understood we built a lot of new things under the hood to make it work and so there was a lot of days where we were just trying to get through the week or get through the day meaning like keep the site running as new people signed up you know and then there was this massive shift in the platform we were primarily a web company when I joined almost everyone who used Facebook used it on the web and then you know what it felt like overnight everyone moved to a mobile mobile device and that completely changed the programming paradigm so instead of you know JavaScript HTML and PHP we were in Objective-C and Java on IOS and Android respectively instead of being able to show ads on the right-hand column there you know there was no right-hand column on mobile so we had to sort of rebuild that whole stack so as a business model and a technology sort of shift all at once that was sort of a big focus of our energy and time you know and then I think it was picking our heads up and looking towards the future and that's you know in 2013 era when we founded our Facebook AI Research Lab and started our efforts in AR augmented reality and virtual reality and then I think over the last many years it's been making sure we continue to push those forward but also sort of honestly deal with you know the challenges of you know the impacts of our products and society and I've spent a lot of the last few years managing abusive content and bad actors and sort of grappling with a lot of the challenges of now that these things are at scale fundamentally how do I make sure that they are you know bending society towards good not bad and how and you know to that point how do you I mean AI obviously is I would imagine central you know in the kind of technology picture of the kind of things you look at Facebook and how do you use you know AI at Facebook and how have you seen that change and just given that the sheer scale of content you have AI must be just an incredibly important technology that you look at can you share some thoughts about that yeah I mean I think the first thing I'll share is I think people it's really hard to have a good intuition about things at the scale we're operating at when I when I say you know there's billions of things happening a day on Facebook you know there's just not a good analog in people's day life to understand what that is like and how what seems like easy solutions to problems at 10 posts a day or 100 posts a day completely fall apart when you get to massive scale from technical from policy and et cetera so I think that's the first thing it's just it's really hard without being in it to understand that and the second thing I’d say is you know AI is not just central it's it's sort of really critical for us in terms of how to manage the rules of the platform at the scale we operate at you know I want to stress that these are the sort of what the rules are what the policies are what the norms are are fundamentally human and policy problems they're not technology problems they require experts in civil rights and hate speech and and and all sorts of other things to help us make sure that we produce policies that that achieve what we want which is to give people you know the ability to share openly you know it's a text box on your screen that you can type anything you want the ability to upload any video or image you want and that's where you sort of start and then we work with experts to figure out you know what are the harms the bad uses misinformation and how do we craft policies to outlaw those things on our platform and then where the technology comes in is on the enforcement side we say okay we want to enforce against Covid misinformation to make sure that people aren't sharing you know harmful misinformation about Covid or vaccines now we've got to find that thing and action it on our platform and that's where you know AI systems seven eight years ago didn't really understand anything about the content so they didn't understand the words they didn't understand the pictures the video they could really understand sort of just like things like click stream and data and other things you can you can analyze so if you wanted to ask an AI system you know eight years ago is this particular post hate speech or not it it was a useless tool for that now we're at a mode where 90 plus percent you know for hate speech just 97 percent of the content that we find that violates our rules are found first by an AI system before any end user reports if that or any expert on our on our platform reports it and it's sort of a relatively stunning progress in the last you know five to seven years and so by volume by percentage AI is by far and above the best tool we have to sort of say look once we have a rule and those rules change all the time Covid you know didn't exist 18 months ago how do we make sure that that happens at scale and as accurately as possible on the platform and we can there's lots we can talk about there so if you want to dive in I’m happy to go well I guess one really interesting follow-up is is you know with the caveat that you know predicting the future of any technology is pretty tough I mean how do you see the role of AI changing you know in the next let's say five to ten years I mean given the kind of trends that you've seen so far and then you extrapolate I mean any thoughts on that well I mean it has been really sort of stunning progress in the last five or six years so for people who aren't you know AI experts if you if you look at you know 2013 was really a a key turning point when the first sort of deep learning or trained supervised computer vision system Alexa.NET sort of won a computer vision challenge that to that date was most mostly dominated by sort of more bespoke handcrafted tools for this then and then that sort of took over as the way to build computer vision systems and in the last several years that's happened in language as well natural language processing so you know we're at a point now where these things are progressing very rapidly what I’d what I’d say in the next five or ten years is a few things are going to happen one is you know we're seeing dramatic advances year over year and the capability of these systems to understand content context and language so to know you know what are you talking about what are you interested in when you're when you're posting here to understand video to understand imagery we are also seeing sort of a rapid advance in the ability to push this technology sort of out of the data center on the server farm and on the devices so our Oculus Quest headset you know doesn't work without real-time computer vision running on locally on the device our portal video chat system you know uses AI in its critical path locally on device to process information and so I think you're going to see these big models running and the servers sort of embedding themselves everywhere and I think that's going to open up tremendous opportunity far outside of the core technology companies I mean I think if you talk to any of the big five technology companies they're all massively invested in AI using their business every day if you talk to the Fortune 500 that's probably not the case in five or ten years I will be shocked if that isn't the case everywhere and we're seeing glimmers of it now whether it be John Deere using it to improve their tractors whether it be you know early advances in medical imaging you're going to see this sort of proliferate throughout the world and really change the way AI deploys and affects people's lives so I have a cybersecurity question like you know I know for example UCLA it's a community of you know tens of thousands of students and tens of thousands of faculty and staff and collectively cybersecurity is really hard with that many people but that's that's a tiny drop in the bucket compared to the 2.7 billion or more people that you have on the platform you know at Facebook how in the world do you address cybersecurity and manage it just given just the sheer size of the number of people the number amount of content how do you do that yeah I mean I think that security in general is very challenging because you're you know the analogy is you have to make sure every door is locked and the attacker just has to have to find the one thing you forgot and so I think defense is always is always a tough game anywhere you go I’d say in many ways I am happy about our scale and as a consumer I often when I'm looking at other services to trust you know the bigger scale they are generally I have more confidence in them because you can end up hiring the dedicated teams and technologists needed to really you know combat the bad actors so we have you know thousands of people working on these problems and they are everything from analysts and operations experts that understand how foreign intelligence services may work to you know software engineers with deep expertise in running static analysis to make sure that we don't introduce bugs in the software as we're building it and we can just invest massively on sort of every point of the stack and it doesn't mean that it's perfect but it means that we can have a very robust defense against these things and use the information we have to try to detect when they're when they're issues but it is something you just have to invest sort of in every area of expertise and in particular on the software and engineering side you know one of the challenges in cybersecurity is there's often this divide between sort of the analysts and the engineers and trying to make sure that we you know have all of those disciplines in the company working together is really critical to try to protect protect people on the platform and how much do you think that education you know obviously technology is vital to ensuring cybersecurity but then so is education just awareness I mean is that a big part of the effort as well or how does that fit into the overall approach are you thinking about of the average person using products or anything yeah exactly just that you know just you know educating users regarding you know you know your job is easier to the extent that users engage in behaviors that you know enhance their own cybersecurity yeah I mean I think you know I'm always a fan of education especially when it what I’d say is this is like I think education is good because I think people should be empowered particularly when it comes to technology and I have a deep belief that anyone really wants to understand how something works can figure it out and I’m always pretty annoyed whenever technologists try to be too you know high and mighty and not not engage with people to help educate them however I think out from a cybersecurity practice relying on every one of your users to be properly educated in order to be safe on your platform is a losing game it is our job to try to build tools that are default secure and so easy to use that the average person can use them you know without fear of compromise or problem so I actually think much more of the onus is on us and you have to be careful like education is a great supplement something we should do but you have to asse it's sort of our responsibility to protect people not that they will figure it out that's it's a great point and you know you know you just do the math right if 90 percent of your users are you know very attentive to cybesecurity security that means you've got 270 million people who aren't right so that's that's that's a really big number so here's another question about obviously the pandemic has you know forced us all to be far more kind of engaged you know online than we perhaps might have otherwise been and that raises this issue of kind of you know will our online experiences become even more immersive you know through things like virtual and augmented reality I know that's a space that Facebook has some interesting interest in is that you know to the extent that you're able to discuss that is that something that you can tell us a little bit about your plans for yeah I mean I think this is a place where you know I think there are two fields that your students will you know in the first 10 years of their career see dramatic change and I actually think that the 2020s are going to be some of the most exciting you know decade you know ever in humanity and you know you we talked about one already which is AI which I think is just going to infuse everything we do in a way that people can't even comprehend right now and then the second is is virtual reality and you know everyone watching this today is familiar with sort of VC fatigue or Zoom fatigue right you spend all day fixed in one rectangle looking at you know a face on that rectangle no 3D information I have no spatialized audio as brilliant as remote video conferencing is today it is a poor substitute for in-person communication what VR really has the promise to do you know in the next decade not in decades to come is to break down almost all those fundamental limitations I don't need to stay fixed on you I can sort of look around the room you're going to be in 3D not in 2D I spatialized audio so if we're talking to multiple people you know when someone's on my left speaks the same time someone on my right speaks I can sort of look at them and by looking at them they know they should you know that they should talk now the person should wait all of the like wait trying to jump into a VC thing goes away and I will guess it is a whole lot less fatiguing you know than your current VC setup so I bet you in 10 years it'll be a whole lot more common to sort of you know effectively teleport in via a VR system than it will be to get on an airplane or or use a VC system you know in person will be the gold standard but VR is going to be 90 95 percent of that rather than sort of the 30 percent or 50 percent that video conferencing is today I think it will be a fairly big transformation wow that there that happens that would be an enormous transformation with impacts you know far beyond our how we engage with people but just for you know for travel and for a lot of other things as well so here's a question sort of more generally what what are you excited about at Facebook and there's like what's next and you know other than of course the things that you just mentioned what are some of the other things that just get you excited about where you see things going yeah well I mean I'd say the first thing is you know I think a lot about what we do and how it is similar or different than other companies out there and I’d say that the the thing that excites me large about Facebook is I think we are the company that invests the most in bringing technology to the widest possible audience and that means cheap free easy open source open and it this goes up and down our ethos so everything from the amount of engineering time we spend to make sure our applications on Android work on older lower end devices this doesn't affect anyone in the USA but it affects hundreds of millions or billions of people all around the world we're building a fiber cable to literally ring the entire continent of Africa that will 3x the entire capacity or multiple increase the entire capacity of the of the continent in terms of a backbone so we're investing heavily in bringing technology to people wherever we can and even when you talk about advanced technologies like virtual reality you know I mentioned the Oculus System now you know anyone watching this can log on to Amazon and buy for 299 dollars a full functioning VR system no PC required no custom special room no sensors you know that's new as of the last you know 18 months thanks thanks to our investments in technology and having been there as we built all these technologies we had this fork in the road which is we have this really awesome VR technology if we make these headsets a thousand dollars or 1500 dollars we can pack them full of features and if you have a PC you can get bigger rendering and that'll be great for the hundreds of thousands of people the low millions of people who can afford to spend thousands of dollars on a VR rig and instead we took all of our energy and said let's bring this technology make it simpler make it come out of the box and make it cheaper 300 dollars total system cost versus at the time it was about a you know 1500 dollars or more total systems cost and so that that approach of like how do I take this technology and like not just give it to you know the rich westerners but really just push it to the world I think it's that the approach that I’m really exciting and then even when you go into AI you know this is why we published 1700 papers in the last you know two years alone we have many dozens of GitHub repositories that are active we build things like PyTorch which is now the premier framework for sort of doing AI research and we do this stuff in the open because I think it can benefit so many people so I’m very excited about where our AI is going to go I’m very excited about as we talked about where VR is going to go and we didn't even talk about augmented reality but sort of more than any of those things I’m excited about you know spending my time and energy driving down cost and increasing access so that these things can help the most number of people all around the world yeah and if history's any lesson the cost declines that you just mentioned you know from fifteen hundred dollars down to you know three hundred dollars so we're not done right in other words that by assing that technology you know three four five years from now will be even lower cost and therefore accessible to an even broader range of people so it's it's a it's a great example of the sort of democratization of access of these of these revolutionary technologies I guess maybe one more question before I move on to some questions from students in the audience do you have any advice for today's engineering students you know the the people like you were back when you first started your career when you were in college but you know those people today in terms of applying their problem solving skills to solving the you know the tech challenges that are going to be coming down you know at us in the next you know you know couple of you know decades yeah I mean I so the first thing I'd say is you know the specific technologies change you know I learned a program in C++ which is still around but isn't you know the dominant you know technology out there these days the web didn't exist when I you know started in college certainly the iPhone didn't exist and you know back when I was in school out you know AI was mostly about search trees and you know alpha beta pruning and that sort of thing and I didn't find that particularly interesting at the time so I worked on systems and graphics instead and now it's all very different and so you know I think the most useful skill you can do as a student is sort of learn the fundamentals because abstraction caching good API design those things transcend programming language and problem you know decomposition of problems etc so there are a bunch of fundamentals you can learn that really help you and go as deep as you can like I took assembly language classes and it's helpful to know how the machine works and then the second is never get tired of learning I mean this is the best part of my job is was you know when we're in an office walking in a room now it's on a you know a conference call it's like wait I don't understand what you just said can you can you tell me how that works and then you just get a I get a lecture from an expert in a field that's that's emerging and that's like my best day ever because I get to learn something new and so I think if people can learn how to learn and never let their egos get in the way and say well I should know this or I'm super smart and just like you know I always approach every meeting every conversation every person with like they have something I can learn from what is it and how do I learn that thing and then the third piece of advice I’d say is you know probably the best benefit I have of being you know 20 plus years out of school is I can sort of watch all the decisions not just I made but you know all my peer group has made and the mistakes I've made in the past and the good choices I've made and I’d say that the biggest factor in a lot of these things is certainly don't choose things for sort of extrinsic rewards so doing things because you're you think it's cool your friends think it's cool it pays a lot of money it's a good brand that stuff gets old quick and after the sixth month of a crummy day at work you know the shiny title or your friends thinking you're cool isn't really gonna get you through the day what does get you through the day is intrinsic rewards what do you care about in the world what are you passionate about you know is it climate change is it mobile payments is it learning new things is it helping other people like figure out the thing that like when you do that it puts fuel in the tank and then find that and do that because that is the thing that will get you through all the crummy stuff that comes with even the best possible job so I have probably the best possible job on the planet there's lots of stuff I have to do I don't like but the reason I do it is because of the other stuff I do that really motivates me getting technology advanced and out and into people's hands and like that is the is the thing to seek and then I said this before the last thing I'll say is just choose a peer group that inspires you like that will make you better well thank you very much so now we have a few current engineering students with some questions so first I’d like to introduce Itohan Ero who is a computer science major at UCLA and has a question yes hi my name is Itohan Ero I'm a second year computer science student from the Class of 2023 and I’d like to ask you how does Facebook balance resources between feature requests and engineering maintenance that is a is a is a really good question and the honest answer is there isn't some simple like formula you can look up in a textbook to tell you how to do this it is quite context dependent based on the area and what you're doing and it's one of the problems that I put into this sort of like forever problem bucket which is something we're continually discussing and continually debating about you know have we spent enough time on maintenance and rewrites and and other things like that versus investing in new features and other things what I will say is there is no perfect answer but what I try to strive to do is to make sure people understand explicitly what trade-offs we're making like there are times when we're like we don't have time to wait we've just got to get this done and we know we're taking on tech debt you know as a very specific example we spent a lot of time preparing for the U.S. elections last year to make sure we're doing everything we could about misinformation and whatnot I would take on any tech debt that I'd have to pay down this year in order to try to make that down and I was very explicit with the teams like we've just got to do everything we can to show up as best we can here we'll sort of you know re-plumb the systems later as need be right and making sure you make those decisions explicitly so people don't second guess or misunderstand what they're doing and I think that's one of the big secrets to how to operate well is just making sure people are clear on what decisions we're making what the context is and how they can do their best work and that's really a lot of my job is to just make sure if anyone's confused like it's it's kind of my fault like I should have made it more clear to people because when people are clear then they can do awesome work and they don't really need me does that help yes that answers the question okay thank you very much and our next question is from Christoph Charles who is an electrical engineering major at UCLA Engineering hi I’m Christoph Class of 2024 and I like to ask you are there currently any limitations that you're facing in developing deep learning for Facebook I could fill up the rest of this session with the limitations we have but that's the good news because limitations are opportunities they are opportunities for change and development there's a lot we don't know about AI and how it works we don't even have fairly good formal theories on why these networks learn the way they do there's a lot of experiments happening on how big you can make these models how much data can they take in before they stop learning and I’d say one of the biggest challenges in artificial intelligence right now is this concept called you know either generalization on one hand or transfer learning on another so if I train a system to learn one thing you know how easy is it to learn something else so you know the early systems I could play games you know could play Go but couldn't play chess for example and then there's a breakthrough that says okay any turn-based game where you can see the whole board state this AI system can do great so it can do chess it can do go but it couldn't play StarCraft it's an entirely different system that can do that because it's a partially observed board in real time and so we're making incremental advances but a generalized system that can learn like the han brain can we are fairly far away from in AI and so that means that there's a lot of fundamental breakthroughs sort of left to be done but this is not demotivating to me this is exciting you know I think that the human brain is a existence proof of the power of sort of generalized intelligence and so that means that there's a lot more we can do with computing and the ability to do it over the last few years the last thing I’d say is you know even in the last few years we've made this sort of stunning advancement from you know what was what we call supervised learning where you had to sort of exquisitely hand train an AI system so if I wanted to teach a system to recognize cats versus dogs the first step was building this very detailed hand labeled data set with cats labeled as cats and dogs labeled as dogs and dogs that look like cats labeled dogs etc and then you use this and you train the AI system and that was how pretty much everything was run until about two or three years ago now we have this concept called self-supervision where you kind of just have the system to learn the way people do which is they look at a lot of images or they look at a lot of text and you build these things called pretext tasks which are a test you can have the system administer on itself so it takes a sentence and it hides one word and tries to guess what that word is and you sort of have the answer and the data you have to begin with and if you do something like that you can take billions of sentences off the internet and train an AI system that ends up being better than the sort of hand trained systems that's self-supervision that moves from research to like we have systems in production right now relying on self-supervision and so that's a that's a stunning breakthrough and there's a lot more to go there in the in the coming years so I'll stop there to leave other questions because you'll get me talking for the rest of the time but it's a great question I know this is these are these are really great questions and great answers so here's a question the question is how should students think about how they might rise up through the ranks in a software company I think most students understand that if they want to get a job most engineers students if they want to get a job at a place like Facebook you know they want to try to do as well as they can their classes you know be good at software programming you know impress the interviewers and so you know it may not be easy to do all that well enough to get a get a job but at least there's sort of a road map right and the question is you know if they want to rise up through the ranks besides say software programming skills what are some of the other skills that you think would be or are really helpful to students to you know to aspire not only to join this industry but also to sort of you know rise in it and take on positions of greater and greater responsibility I think the most effective technical leaders I know are really a combination of two sets of skills you know one is it is very useful to have deep technical skills and have real experience I think that you know having you know joining a company and starting as a software engineer and writing production code it is just different than the university environment I think you know university educations are critical and are wonderful but you know operating at a billion user scale just teaches you some lessons that you otherwise wouldn't learn so you sort of start there but I think the other thing that is often undervalued or under looked past is soft skills it's people skills it's communication great you can do great technical work can you communicate it to anyone can you communicate it concisely and clearly and accurately can you work with others can you inspire them to do their best work anything interesting I mean name an interesting technology out there that has a single person working on it one person like I can't think of anything right now absolutely everything that has built that's interesting is a team effort and that means that you have to get people to work together and that's honestly often the hardest part the technology sometimes the easy part is just getting everyone sort of working together in a line so you know practice your writing practice your public speaking you know practice team projects and working with others and realizing that it's not about you and being right it's about getting the job done and getting everyone to do their best work and you know those people who can live in both worlds and can geek out on the technology and then flip over and say like oh wow you know this team is struggling this person is struggling what do they need how do I help them like those are the people that I think have no upper limit in their in their career that's that's great advice so another question from the audience is and by the way those of you in the audience you're more than welcome to continue to submit questions we've got quite a few but I’m always looking for for more good questions how one of the biggest concerns with online content generally including of course in social media companies is you know misinformation fake news and things like that and it's it's such a hard problem right because the volume is high and you know if you're overly aggressive then you know you kind of make the wrong decision if you're overly aggressive and filtering it then you filter out stuff that actually isn't fake and you're not aggressive enough then you let you know stuff that is fake through so there's a false negative false positive trade-off so can you talk to us a little bit about some of the technology challenges that that you're looking at and how do you think about that problem at the scale that you have to encounter it yeah it is a challenging problem as you say and this is the way sort of I approach it which is you know there are sort of I think a couple of different ways to break down the problem the first is look we know we have clear expertise in an area e.g. there are world experts in this health experts that understand you know the safety of vaccines or the efficacy of masks or whatever it may be and they are acknowledged to experts you know whether it's the CDC or others that you can sort of rely on to say yes no this is accurate and so then the problem is basically a detection problem there's this piece of misinformation it's very clearly you know wrong and a third party fact checker or someone else or a partner can sort of you know attest to that and then you're just trying to find it at scale because people will sort of you know mess around with the wording or say other things and that's a place where technology can play a huge role because we can you know find small changes I mean what you see is people do you know years ago what people used to do if they were trying to evade systems is you know you take a screen shot of a photo or you crop it a little bit or you blur out part of it or you add a word or whatever it may be to sort of make it slightly different so that things that are looking for exact duplicates word for word pixel for pixel duplicates would be fooled by these systems and what I can do is it can move into semantic similarities so I can say well these things aren't pixel for pixel accurate but like they end up having all the same words or in a natural language processing model these two things actually mean the same thing even though it's a slightly different arrangement of words and so maybe that either gets flagged or gets sent to a third-party fact checker or gets sent to an internal system so again I think it's fundamentally about you know a tool for scale you know when you're when you're approaching emerging things and you know Covid just first happens it's harder and that's where you need people and this is why like this is never a people versus AI thing I always think about it as as AI augmented people you know you know I try to use the analogy it's like if I was gonna dig a basement you know if I give you a shovel it's gonna take you a while if I drive up in a backhoe we're done in the afternoon right and to some degree that's what I’m in the business of doing is providing sort of power tools augmentation so world experts and people all around the world can sort of amplify their work and make sure it works well at scale so thank you very much so here's a question about data science and you know I think data science is obviously fundamentally important but it's also a little bit different from sort of software from programming right there are there are data scientists who also are amazing programmers but there are amazing programmers who really aren't experts in data science and they're experts in data science who aren't necessarily great programmers and so if you look specifically data science what kind of role does that play at Facebook do you see that expanding in the future you need to talk maybe if you can talk to us a little bit about how you see data science that would be really interesting I mean I think it's a great question and I’m gonna expand it a bit because I think that people often have too narrow a review on the skill sets required to build anything interesting and I think what I've learned or what we've learned over the last 13 years is that the best teams actually require multiple disciplines in order to work well and what I mean by that is like everyone understands what a software engineer is but there's also the role of the product manager who can sort of help make sure that we're building the right products there's the role of the data scientist who can sort of help do the rigorous analysis to make sure we understand you know the metrics and what's happening and how different features or products function well there's the user researcher who is you know running user tests and understanding people understand how this product works do they understand what the controls do and how do we improve them to make it easier for them to do it and so really sort of understanding this sort of grouping of specialists that's engineers product managers data scientists user researchers and sort of getting all the disciplines together and I’m not even naming all of them in a project is how you build something great so you know I think one of the biggest changes from you know Facebook circa 2008 to now is how much more cross-disciplinary we are in our product teams you know as a rule rather than maybe a software engineer sort of driving everything back in the day and I think that's been a great benefit to the products we build you know especially the scale we're operating at you know we can't just build products that I want to use right I don't represent anywhere close to the to the user base of Facebook and so if I just like well I like that it's not really a great reason to ship something and this is where sort of user research and data science really can play a role because we can go and talk to all the communities who are using our products and ask them what they want how they react to different things and run sort of studies with different user interface and we can look at the data and say like okay we're doing experiments and how do people use these things and are they using it in a way that we think is is best and that's the way you operate certain products at the scale we're operating out it's less about me and more about how do we discover what the community really needs and then go build it for them now that answer really resonates with me because as you I’m sure you know engineering undergraduate curricula tend to be not frankly as multi-disciplinary as they perhaps ought to be and I think that's an area that you know you know schools recognize that there's an opportunity for improvement but I think we're not quite there yet and so I think the more we can do to broaden those curricula the better and the better prepared people will be to join enterprises like like yours so there's a question I’m sorry I was just going to add one more thing if that's okay because I think this is the other big change that's happening and I think it's really going to affect most of your students and I think it's a good change but it is a change which is you know I think it's really important for people to understand sort of the impacts the technology we're building has on society both good and bad and I don't think you can sort of outsource that to another team we you know people use different words for we call it responsible innovation but you sort of ask yourself the questions as you're building the product you know what would a bad actor do what would someone abuse what would side effects be of this product and as I’m building it what are the sort of mitigations I can build into the product to eliminate or reduce those harms the sort of you know the days of sort of just shipping something and then seeing what happens you know I think have passed and so so I think from a cross-disciplinary standpoint also engineers being open and willing to engage with and talk to experts in civil rights in ethics in lots of other things that you can't be an expert in all of them but you have to be willing to work with those experts in order to understand and say like hey when you do this here's the impact this is going to have on this community and we need to change the way this product works and that definitely is something that I’m not sure is well understood in most undergrad CS curriculums now that's a great point and I guess I'd add that you know while we certainly want a a flow of graduating engineers going into for example the technology industry I think you know there's not a lot of engineers for example who are members of congress on a percentage basis and I think we also need you know more people with technical training and some of these other aspects of society as well I think that would help so here's a more technical question you know there's the algorithm with respect to like machine learning computer vision things like that there's the algorithms and then there's the hardware that the algorithms run on and what do you see as the kind of bottleneck in the next you know five years do you I mean obviously this would have related because sometimes people develop algorithms with you know with keeping in mind what the hardware can do but where do you see the bottle like if you had to pick you know between those two the algorithms and the underlying hardware capabilities and you know I often think of it as a triad which is sort of the hardware capabilities the sort of the fundamental algorithms and then also the sort of the data sets and pretext tasks that you can use and and I think you know it's never one thing and what you find is you sort of always discover the next bottleneck once you fix it in one of the areas so if we have a new neural net architecture that's able to ingest more data before it starts trailing off in its sort of learning rate then all of a sudden you're like great now I can use 10x the amount of data which is going to stress my data set and they're like well great that's going to take me two months to train so now I need to figure out something different on the hardware you then go fix those problems and say cool I've 10x the data I've got 10x faster training I can do it two days now now this network is sort of trailing off and it's not learning anymore so now I need to go back and sort of look at that so it's a sort of a constant process when you're circling through all three of these but to me this is exciting because if we were just working on one of these things I would be worried it's very obvious to me that sort of we don't yet have sort of core network architectures and algorithms and AI that are what we need to sort of push us forward and that's exciting that's really exciting for your students at the same time the hardware isn't static you know there's been a lot of concern about the sort of the ending of Moore's Law and that is mostly true when it comes to sort of general purpose CPUs but the explosion of AI has changed the way compute works you know this it's not a new concept but CMD or sort of vector processing you know lots of data a few instructions has been around for a long time but AI really pushes the limits of this is why you have GPUs which are intended for graphics being re-purposed to train AI algorithms because they're basically monster you know matrix multipliers and and massive FPU throughput you know which is a lot of what you need at the training side and we're seeing this now on the inference side too where even mobile devices are getting dedicated inference hardware in their chips so your phones will have hardware specialized for running AI models in them at very high performance very low power and if you talk to you know putting stuff in hardware is often a 10x opportunity if I take something in a general you know CPU and I move it to specialized hardware I have at least a 10x probably power or performance boost coming for me and so the fact that we're just deploying a lot of these things means that we've got sort of a bunch of future advancements and then again because of this circle we may say well I've got this hardware that runs this network and then our researchers show up and say like oh we have a new kind of network and it requires this sort of operator you're like ah that isn't in my hardware it's like okay next spin of the hardware and that new piece of hardware now runs that thing 10 times fast and like this is where we go for the next 10 years and this is why I like I can't stress enough how exciting it is right now in terms of you know the ability to make future advancements we are like nowhere close to out of ideas yeah I can imagine that's like you know I've been at this a long time but man is it exciting right now wow can you talk a little bit about Facebook's efforts to encourage diversity in engineering teams I mean that's obviously an industry-wide you know topic and I’m sure Facebook has been very actively thinking about that as well any thoughts on how you're going about doing that yeah I mean this is a one of the most disappointing things to me personally about the current tech industry is how much better we need to do on diversity inclusion sort of across the board and it's it's especially striking when you read the history of computing you know for anyone out there to go back and sort of look at the ENIAC and the early programming that happened you know the concept of loops and subroutines you know were invented by Ada Lovelace before she could even work on a system and you know Grace Hopper is credited with the first compiler or linker depending on how you want to describe it and the invention of cobalt and so the early days of computing were actually you know driven by a much more diverse cast of characters than than we have in current day society and yet technology is impacting so many people around the world that it really does need to be built by everyone and so you know this is a problem that all the tech companies need to be engaged on and I think it you know boils down to making sure that people have the right educational and job opportunities that working in tech is a is an optimal choice for everyone with every background and that there are plenty of opportunities for people to advance and grow and so you know what this often requires is just a lot of work to make sure all those things are true day-to-day for every person you know in the world you know this example we announced today the Raise Program which is basically a two-year program to bring people into the field of AI from other technical specialties AI is a hot area of technology everyone wants to do it not everyone can go off and get a PhD in AI or machine learning and what we're offering is basically an opportunity for people to join our AI team and train in person with them as a full-time employee for two years to sort of jump start your career in AI things like this where we to me the best way we can do it is by taking the scientists the researchers the engineers we have and using their time and energy to help sort of mentor bring along sponsor and advocate for you know more people to enter the field and join the company I wish I wish there's an easy simple answer but this is one of these things that we just got to work harder at every day to to make better well thank you very much another question is how does Facebook how do you in practice get people from very different sort of fields for example neuroscientists and cognitive scientists and computer programs I mean it's it's one thing to hire them which Facebook clearly does and you know do you just do how do you get them to sort of exchange ideas and sort of mutually reinforce their kind of creative output if you just put them in the same virtual office and they sort of spontaneously talk to each other do you sort of program it what's the sort of secret for actually getting the interdisciplinary output from the interdisciplinary team yeah I mean this is a I mean now you're describing a little bit my day job which is you know I think it's it's the hard part is is getting everyone to sort of work together and see the world from from each other's perspectives from a skill set perspective and so I do think that it at the end of the day it requires a few things you know one is getting people close together so you know in the real world that means like literally sitting next to each other I would joke that the first solution every problem is to move people's desks next to each other because you know just like getting more physical presence is helpful because a lot of misunderstandings stem from people just not like starting from a basis of trust to basically like wait I disagree with you maybe I’m wrong and like you can do that when you trust the other person and when there's less trust you just think the other person's wrong and and so you start with sort of building that relationship the second is people work better when there's a shared goal you know are you and I you may be a neuroscientist and I’m an engineer but we're both passionate about brain computer interfaces for example and if we work together we can make that thing happen and so figuring out how to get people aligned to say like hey we need all of you and as I said before it's cross-disciplinary it's not who's more important or who's better who's key to the project it's like it's a team and this thing doesn't ship without all of us and we all want this thing to ship so here's how we get aligned and that to me is and we talked about people skills I think the other thing that is you know management is often a bad word and I think you know when you think of management as a thing that's out for itself like yeah that's a bad word but what we think of as a good manager a good leader at the company is someone who is empowering their team to do their best work and figuring out like how do we get all these puzzle pieces to fit together and get this team of different points of view to like make one plus one plus one plus one equal a much bigger sum than any one of the parts and when it happens I describe it like a conductor and a symphony you know if you have the world's best in every single instrument playing a different song it's still going to be awful right even though you've got the best team on the you know ever when you have a conductor that figures out how to get everyone to be their best in one direction it gets you goosebumps I mean it's the day when I was like wow look at that team go and that's we need more people in the world who are doing that work and just like saying I’m here to make them do their best work and when that happens like wow stand back and so that's sort of my thoughts on it well thank you so maybe one final question just going back to your career Hal gave a brief overview of how you ended up as the CTO at Facebook you know that's obviously a extraordinary position in the technology world was there a I don't know a pivotal moment or a big break or a point in time you know that sort of led you I’m sure it was many things but is there are there any particular things that you can remember along the course of your career that helped really put you in the position that you're in now yeah I mean I don't really subscribe to the like one single moment sort of philosophy of people's lives or of history what I do think is when I reflect back on the sort of the decisions I've made and I and I think you know I'll tell one story from my past which is you know you said I started a company 2000 it was the dot-com bust it was a crazy time to start a company but it was great it was a lot of fun we ended up selling it to Sun Microsystems at the end it was sort of a medium success in terms of it you know it wasn't a company anyone has heard of right and they're not going to hear of and that's that's totally fine and and here's one of the big mistakes I made which is this is why I talk a lot about intrinsic versus extrinsic which was I had a chip on my shoulder I wanted to prove to everyone that I wasn't lucky I was good and I wanted to start another company but it wasn't because I was passionate about the problem or any of these things I just wanted to start a company number two to prove to people that I was just that good and I basically wasted a year of my life trying to like come up with the best opportunity an idea and nothing was good enough or a billion user opportunity and it was because I was doing it for the wrong reasons and that was the time when Mozilla called me and said you know I've heard this Firefox thing just launched you know we're like a non-profit open source thing we kind of need some help scaling our engineering team and I remember my those 17 years ago my thought process was like well that'll be interesting I'll learn a lot like it's a non-profit I might not become famous or anything but like that'll be interesting and I met all the people like wow you're I don't know how open source works like I’m not an expert in browsers but I know how to build engineering teams like I'll help you build an engineering team if you teach me stuff about this and that was probably one of the best decisions I made because it was like it was amazingly fun and I learned a lot and that was sort of similar to how I approached the Facebook job which was once again like I don't know if this thing's gonna work or not at the time in 2008 there was a whole lot of concern that the company wasn't actually gonna make it like the thesis on the ground was like social networking companies don't make money MySpace you know had a lot of trouble with their issues you know email never made money et cetera et cetera and so a lot of my friends were telling me like don't join there they're not they're gonna crash and burn and I was like maybe not I don't know but like I’m gonna learn a lot and I think I can have an impact and again this is like so every time I sort of went there it was better and that's why that's the advice I give to people it's just you know because that's what work for me and that's what everyone does is tell you what worked for them so you know your mileage may vary but that would be the way I would sort of interpret how I've made some of these decisions great thank you so much so we're about out of time let me just say a couple of very brief closing remarks one is that we will be sharing the video on the UCLA Engineering website once we get it processed secondly I think you know it's very difficult to know what the future of technology will hold but you know today's students you know can be reasonable can reasonably expect to be in professional practice if you do the math through the 2060s yeah just incredible right yeah and like we can't possibly know what technology is going to be like 2060s but we know it's going to be exciting right the change is between now and then so all of you who are undergraduates or graduate students right now who are entering the technology workforce or will be in the next couple of years you know you have before you just a set of incredible opportunities and I’m sure you'll see some incredible things so let's close with a big virtual hand for Mike really appreciate you taking time to speak with we really appreciate it and best of luck with your work and thank you again so much for spending the time with us thank you all thank you to the students really great questions and I just would agree that I think technology really holds the promise for a better more prosperous world and man if I think about what the 2060s are going to look like I hope it's I hope it's amazing and great and I think you all are going to be a huge part of that so don't lose your hope and optimism get out there change the world do some good stuff great thanks to all
2021-03-14