Stanford HAI 2019: Keynote with Bill Gates

now we're back to our afternoon program which is just as exciting as the morning if today's event is connected by a single idea it's that humanity is the key to building a positive future to AI but about five years ago I found myself worried about the direction a I was heading in this is a technology with the potential to change history for all of us but when I looked around I saw a very narrow group of people developing this technology for all of humanity to quote one noted philanthropist in tech a eyes researchers and developers were mostly consisted of guising hoodies around the same time oh gollu Sikorsky still a PhD student in my lab approached me with the idea of helping more young women get involved in AI this inspired an epiphany moment and connected to dots I had long been interested in that the first one is a ice lack of representation and the second is its need for a more human mission so together with olga and dr. rick summer director of Stanford's pre-collegiate Studies program we found it the Stanford AI for all summer camp in 2015 and invited the inaugural class of 24 high school freshmen girls to Stanford campus staffer AI for all is an intense two week program at the Stanford AI lab focusing on introducing AI technology alongside its human centered applications and topics more than 40 faculty and students in Stanford AI lab contributed to the curriculum teaching research projects and few trips to in the first year four summers later more than 100 women have graduated from the Stanford AI for all summer camp they have gone on to run middle school ai workshops star AI and robotics clubs in their own high school conducted AI research in university labs and connected AI with disciplines ranging from medicine to the environment and from dance to art over and over again their stories reinforce one powerful idea true brilliance can be found in every kind of human being our next presenter Amy Jean is a prime example amy attended the inaugural AI for all summer program at stanford in 2015 where she played a had a hands-on role in relay AI research around how computer vision can more accurately track hand hygiene practice in hospitals but those two weeks were only the beginning so despite the heavy workload waiting when Amy returned to school she stayed in my lab and joined a project with my then PhD student Serena young and our medical school collaborators dr. Jeffrey Joplin and dr. Arne Milstein mentored by Serena and Jeffrey Amy led the study of applying deep learning algorithms to keeping track of medical instruments during surgical procedures and even assisting even assessing the performance of the operation itself the work was presented at the machine learning for healthcare workshop at 2017 snips conference and many of you know that's one of the prime a I conference in the world alongside more than 150 research paper submissions amy's work received the best paper of word that's a career highlight for anyway in this field let alone for someone so young and I think unofficially she probably broke the aged record for best paper award Amy now attends Harvard University as a freshman where she continues her passion of applying computer science and AI to solving complex problems in creative ways she is a tremendous talent and I'm honored to introduce her to all of you please welcome Amy Jin Thank You professor family for the kind introduction it is such an honor to be here today and I'm excited to share a bit about my journey and AI so far in ninth grade I attended my schools and your research symposium a day-long event that mimics a mini research conference one of the keynote speakers that year was from IBM and he talked about the amazing feats IBM Watson had accomplished from beating the Jeopardy masters to becoming a chef inventing newfangled dishes by analyzing recipes ingredients and flavors I was blown away the concept of AI was still completely abstract to me it seemed like magic but my curiosity was piqued and I was eager to learn more that year I also happened to hear about AI for all through the women in STEM Club at my high school I was extremely lucky to have the opportunity to be a part of the inaugural class of AI for all at Stanford therefore my group project I worked on a computer vision based hand hygiene monitoring system to combat Hospital car defect infections I was struck by how interdisciplinary and how and the diverse array of real-world applications AI has for instance how natural natural language processing can be used to mine tweets for disaster relief and how by informatics could be used to illuminate cancer genetics at AI for all I wasn't just exposed to the power of technology but also to the idea that leveraging the power of technology for social good was at my fingertips I asked myself how can we make a world the world a better place for the use of human centered AI after the program ended I approached my project mentor Serena young asking how I could continue learning more and possibly get involved with doing more research under her guidance I studied Stanford's courses on computer vision and convolutional neural networks and later as we discussed project ideas the concept of using computer vision to evaluate surgical skill intrigued me annually 7 million patients after surgical complications were worldwide at least half of which are preventable as many are caused by poor performance by individual surgeons or surgical teams providing targeted feedback for surgeons could help improve patient outcomes however the current standard for surgical performance assessment requires expert supervision a manual process that is both time consuming and subjective but given that musicians and athletes have coaches to help them improve why don't surgeons benefit from the same type of feedback so working with my mentors dr. Serena young dr. Jeffrey Joplin and professor Faye Haley we developed a deep learning approach to facilitate the assessment of operative skill because analyzing surgical skill involves tracking tour usage and movement patterns we first collected a new dataset and trained a model to classify and localize the tools in gallbladder removal videos then we used our tour detection model over time to further characterize 2a movements we extracted several assessment metrics and found them to give rich insight into surgical skill making the assessment process faster and less taxing for example we generated toy usage timelines that show how frequently tours are switched back and forth heat maps that reflect the surgeons motion economy and torture directory maps that reveal how effectively a surgeon performed one crucial phases of the operation to validate our approach Jeff and a group of Stanford surgeons man a manually rated each test video their scores directly correlated with what we found from mark drafted metrics proving them to be effective performance indicators our work shows that by analyzing how tours are used in surgical settings we can start to better understand what is happening during surgery and ultimately help surgeons consistently improve this will make surgery safer and better for patients looking to the future I'm excited by how AI could serve as a toolkit that empowers us to devise innovate innovative effective solutions to everyday problems but I've also come to realize that my first impression of AI as a cure-all to the problems we face was limited as it advances at breakneck paces we must ensure chassé possible development and be wary of unintended consequences to do so we will need to enlist people of all disciplines to work together for instance working with and not working to replace surgeons and healthcare workers to bring AI technologies from bench side to bedside and bridging policymakers ethicists researchers and entrepreneurs together to ensure the safe and fair development of such applications AI exemplifies how magical things can happen when diverse minds strive towards the same goal of making the world a better place before I wrap up I wanted to thank my mentors Serena Jung and Jeffrey Jopling I could not have done it without their continuous support and guidance I'm also so grateful to Professor Feifei Lee dr. Arnold Milstein the Stanford AI lab Stanford medicine and after all for this research opportunity and thank you all so much for being here today thank you Amy so much for sharing with us her experience that she did all that in high school and got into Harvard so so I'd like to introduce you to another ai4 alum an equally impressive young woman Stephanie Tana Meza is a junior at Salinas High School and attended a Stanford AI for all in the summer of 2017 stephanie is a second generation of mexican-american from Salinas Valley here in California where she was raised against the backdrop of migrant farm work not a world we normally associate with in Silicon Valley she once told me that although she comes from a community that struggles with limited hot water and electricity she does not in her own words consider herself poor acknowledged having worked with her personally I can tell you she's right since discovering a passion for computer science and AI Stephanie has begun a project to take a data science approach to mapping the water talk cities toxicity in her town as part of a larger effort to correlate water quality with poverty people like Stephanie break the mold of technology innovation they bring not only ingenuity to the field but the new perspectives we need to broaden its horizons and they give me real hope for the future of leadership in AI please welcome Stephanie tanner Meza [Applause] Thank You professor Feifei for the introduction it is such an honor to be here today to share my story with you all we can all agree that inclusive diversity cultivates unique perspectives cultivating unique perspectives allows for the input of all people across the social spectrum which is essential in advancing society the question that now arises is how can we ensure that communities but they do with the diverse of individuals such as disadvantaged financial bearing historically underrepresented groups have educational opportunities available to them to be at the forefront of new tech careers I was born and raised in Salinas California also widely known as a salad bowl of the world for the vast production of strawberries broccoli and lettuce to name a few the majority of my community is home to many first-generation mexican-american children whose parents immigrated to land of opportunities and I am one of them as a low-income underrepresented minority I've experienced how growing up in an underserved community can hinder or feed one's intellectual curiosity by chance in middle school I stumbled across one of the very few free community coding clubs in my area called quarter dojo which planted a seed of curiosity within me this seed of curiosity grew and became the driving force behind my passion for CS I soon merged my passion for CS and for attempting to solve global environmental issues such as water contamination I got interested in the issue of water contamination because I have witnessed how chemical fertilizers and pesticides can end up in the water supply of our community rivers I started to use computing to explore this through science fair project in ninth grade I investigated the effects of agricultural and non-agricultural River water on aquatic life I was curious to understand whether agricultural chemical fertilizers or pesticides presented any harmful threats to aquatic life I continued my investigation to solve water contamination through the a Ifrah research fellowship program I worked with miss Raquel Munoz a senior data scientist at medium on a research project that used a I technique techniques to predict whether parts of the Colorado River a river that serves the lives of 36 million people and endangered wildlife presented good or bad water quality the point here is that none of my projects could have been possible if it weren't for opportunities like Puerto dojo and AI for all who exposed me to the computing field at an early age knowing how impactful learning about computing at an early age Wester me I set out to do the same for students in my hometown I founded a computer science and artificial intelligence Club at my formal middle school for students who are eager to learn about AI and CS my club serves as a resource for middle school students to gain knowledge and exposure at an early age about computing something that is uncommon in our community most importantly my club plants a seed of curiosity within students or at least I hope over half of the students our future first-generation college students come from socially disadvantaged backgrounds and for the most part this is their first exposure to the computing fields since the beginning of the club I have become motivated to learn about these fields outside of the classroom Michael a current club member said he is quote interested in a computer science and thinks it is very amazing that seed of curiosity I had instilled with me inspired me to create change within my community in addition statistics have shown that less than 20% of females occupy tech jobs in the u.s. thus to help alleviate the gender disparity in CS I founded a girls who code club at my high school to help bridge the gender gap both academic programs contribute to my ultimate goal of ensuring that communities with diverse set of individuals have educational opportunities that of a to them moving forward what most excites me about the future of AI will be seeing how it is used within different fields such as environmental health discipline to end off I strongly believe inclusive diversity is important and human centered AI because everyone should be part of solutions to complex problems that are meant to address society it is important to have your presentation of all of all types of groups across the social spectrum to cultivate different perspectives about issues I would like to thank the entire AI for all community for their continuous support throughout the years thank you so much Thank You Stephanie thank you that was just so amazing but as impressive as Amy and Stephanie are what I find most inspiring is that there are many students like them in our AI for all programs but so far I've only told you half of the AI for our story the part about how it started here at Stanford by 2016 that demand for our Stanford camp far exceeded all expectations with students flying in from Ohio New Jersey Connecticut or even China we realized staffer alone couldn't keep up as luck would have it I crossed paths with the true ally in the fall of 2016 Melinda Gates and I began a series of conversation about our concerns and sense of urgency to make AI ntek more inclusive and diverse starting from students in classrooms and reaching to workplaces in the industry not only did she support and encourage our vision for AI for all but she coordinated an early funding round to help us grow and that quote I used about guys in hoodies earlier it was hers what began as a Stanford program became a national nonprofit in early 2017 co-founded by Olga dr. Rick summer myself and a great team of staff led by tests poster board members and advisors that year we expanded to a second campus at UC Berkeley now in 2019 I'm very proud to say we'll be on ten campuses this summer from Boston to Pittsburgh from New York City to Phoenix Arizona and many more are in the planned dr. Olga rusik offski now graduated an assistant professor at Princeton founded a chapter there as well our students expanded to become even more diverse to become an even more diverse group with an emphasis on women people of color and students from low-income families and rural communities this simply wouldn't have been possible without the generous support of Melinda Gates among many others now I have heard that behind every great woman there is a great man and I think it's actually true in this case Melinda Gates and her husband Bill Gates co-founded and run the Bill and Melinda Gates Foundation which has dedicated itself to expanding opportunity to the world's most disadvantaged people and communities they've done truly amazing work on global issues like health care education and income inequality and are perhaps most famous for their aggressive mission to eradicate malaria bill of course has his own legacy in tech which began by writing a basic language interpreter in 1975 he co-founded Microsoft and led the company to become the worldwide leader in in business and personal computing software and services after decades spent in the software industry which he arguably helped create he shifted his focus to working at the Bill and Melinda Gates Foundation in 2008 as a tech leader turned philanthropist Bill is a valuable ally in our quest to use technology to make the entire world a better place so to joy Amy and Stephanie in a conversation about the future of AI please welcome mr. Bill Gates [Applause] thank you so much for joining us today mr. gates I guess just to jump right in to start off on a positive note what excites you about AI and its potential to benefit humanity well there's so many things that are deeply mysterious the ones that I get to focus on have to do with health in developing countries you know 95 percent of the children to die under the age of 5 or in these countries where we have almost no doctors and we don't have the skills to bring the kind of interventions that we take for granted here so the idea that we can take AI and understand for example why prematurity rates are so high and understand the nutritional deficits that take place of the kids in these very poor countries up to 20% of them die before the age of 5 and 40% of the remainder will never develop physically or mentally to their full capacity so they are deeply malnourished during their early years and so their ability to learn and contribute is permanently damaged we've always known that there's various dietary influences that the microbiome affects both the prematurity and and these nutritional outcomes but it's only with AI including partnerships with the Mark Davis lab immunology lab here at Stanford that we're taking all that data and using AI to understand okay what is it about proteins or pathogens and some really low-cost interventions are now emerging to help us intervene and dramatically reduce prematurity in this this malnutrition so it's when I see it applied to something that without AI it's just too complex we never would have seen how that system works that I feel like wow that is say a very good thing moreover what are some actionable items one can take to ensure the responsible and ethical development of human centered AI well the world hasn't had that many technologies that are both promising and dangerous you know we had nuclear energy and nuclear weapons and so far so good although memories seem to be fading on that and you know recent behavior certainly is is deeply concerning on that front with AI the you know the power of it is so incredible it will change society in some very deep ways so it's great that Stanford stepping up one of the early pictures in there was actually of shakey the robot over at SR I and I was 13 years old when I saw that video of shakey and it's funny to think how over optimistic we were like okay shakey is stacking up the blocks now you know let's get it out in the factory tomorrow and this is going to be really easy to solve these problems and so for a long time a I when I and when I started Microsoft I literally wrote a note to my parents and I said okay I may miss a bunch of breakthroughs in AI and that'll be what I give up to create this company but oh well but for about 20 years I didn't miss much more recently there's amazing things going on and fortunately Microsoft has gotten to a size that it along with you know Google and and many others gets to participate but the you know the fact that the technology is moving so quickly and the policies and understanding around it even something just as simple as okay face recognition you know what what sort of awareness and use case should there be for that even that and you have these are not issues that confine themselves to nation state boundaries in a simple way like a lot of previous technologies so it you know it is concerning that someday Stanford won't want to brag about how it was a pioneer in AI you know unless we we do a good job managing it yeah I guess along the same lines as we see that problems are becoming more and more complex and require collaboration across disciplines how would you encourage this cross-disciplinary collaboration that is central to the development of human centered AI well the their potential collaboration than other area our foundation works and a lot is the u.s. education system and they're the very basic questions about why are some teacher so good why are some students not very well motivated and other students are very well motivated unfortunately with deep correlations with socioeconomic factors we are really at the very beginning of that you know the state of the art is such that everything we've learned about education last hundred years you could not say that the best teacher the most inspirational excellent teacher lived 100 years ago that's how much we've learned about education now doctors it's a little better you wouldn't say that the best you know cancer doctor or eye doctor was one that somebody went went 200 years in the case of the u.s. that the dropout rates have not improved the overall academic achievement has not improved even as we've doubled the percentage of GDP that goes into the field so the opportunity here to take and get out of endless debates but to really look into okay what are those good teachers doing what is the nature of that motivation which interventions can really change that that would be a very profound thing you know education is is sort of playing and and yeah if you look in the Rd percentage that society assigns to education you know where do the smartest people go in do they go into educational research how does the educational research budget compare to say the NIH research budget you know what is the what is the equivalent of Bert in the world of educational research where somebody has something profound and everybody goes oh ah that's so fantastic there is no equivalent it's a a a kind of a desert so anyway I think it is a chance given the incredibly general purpose nature of these technologies to find patterns and insights it's a chance to do something in terms of social science policy particularly education policy also you know health care quality health care cost it's a chance to take systems that are inherently complex in nature and that just individuals kind of trying to troll to the data can only find weird correlations like okay Minneapolis spends half as much as Texas but okay how do you intervene what is the next step are kids growing up in a certain location seem to do better income and race independent than other locations that's the kind of thing a human might spot but these systems should help us look not just a correlation but try interventions and see causation as well so you know it's a chance to supercharge the social sciences with the most important by far being education itself I'm on a different note what do you think are some of the biggest problems that artificial intelligence can uniquely solve well the if something is complex enough like the you know you take the microbiome it you know that's billions of data points even the subspecies matter a great deal we prove them recently not just macro statistics like diversity our lactobacillus or something but you really have to get down and look at those gene profiles we have this incredible result that if you give kids in some countries once a year an antibiotic that cost two cents called the serum iesson you save a hundred thousand lives and in a sense it makes no sense because you can't the the that antibiotic is disappearing from their system within a few days so there's something about their microbiome the intestinal gut Junction health that has this profound effect and I don't believe that without machine learning techniques we will ever be able to take the dimensionality of this problem and and be able to to find the solution about what is going on there and once we understand it of course we'd like to magnify that effect and avoid using a broad-spectrum antibiotic which has resistance type effects at all so many complex problems and many very complex datasets only with these techniques that are in a sense pattern recognition techniques the upper bound before the breakthroughs in machine learning was such that many deep societal problems were not tractable now if we get the datasets make sure they're used appropriately because I think we can deal with privacy concerns and yet still have the type of deep longitudinal information that would would reveal these patterns and so it's a chance whether its governance education you know how to accelerate the advances in all the sciences yeah you mentioned the potential that AI has you benefit society in many ways could you talk about an application that has already been positively transformative to society well I'm one see there you know that many you know certainly the the search engine technology that Google or Bing are using which has been greatly beneficial the amount of AI that's being applied there is super impressive in that you know led to the sort of foundational in terms of the cloud platform and how that was created in a in a very generalized way we in terms of actual medicines that would not have been discovered the next 10 years or where you're going to see that in this dramatic form in particular that work on on prematurity you know to give an example we took the 23andme data working with them and saw by using AI learning that there was this deep association with malfunctioning selenium processing genes and risk of prematurity and so we literally have now 20,000 women who live in areas of Africa that their natural diet has no selenium in it that we are intervening by giving them a small amount so we'll know 18 months from now and based on preliminary data we expect to see about a 15% reduction in in prematurity which for Africa as a whole would project out to be about 80,000 lives saved per year which always you know when you say show one picture in one life it's more dramatic than than tens of thousands so I think we're you know it's the current set of things the the deep machine learning didn't really get into the drug discovery process or what had been called systems biology until quite quite recently and so and in the case of education it is not yet you would have not even begun to do that work in terms of understanding motivation and engagement and teaching styles and teaching assistants that would really improve the output of the system ie better learning less dropouts you know key key things that the current status is deeply on unsatisfactory I think for the remainder of the time we would want to open the floor for some questions from the audience when it is your turn please state your name and affiliation and in the interest of time please keep your questions brief so we can get to as many as possible hi I'm commander Chatterjee and the founder of unmanned life-form you trying to use human centered AI to do autonomous systems and so my question back to you dr. gate and of course to you was where do you see the boundary between ethics and machine learning particularly means applied to autonomously moving to an autonomous society where most things will be done by machines in one way or another so where is the boundary there between the ethics and the machine learning in the data sets that we're getting particularly because the autonomy can be very well in some sense very anodyne so how do we ensure that ethically it does the right thing well it's a very broad question there's a there's different domains and there's different degree of autonomy you know the book army of none talks about the current weapon systems that we have like the agus missile firing system that by most definitions is an autonomous system that it is authorized to fire based on incoming targets and there were a few cases where it accidentally shot down a commercial airliner so even in that case where people thought it was for very well founded it it turned out to be very complex now then again you don't want to be too risk-averse on these things because the idea of solving very tough problems you always have to compare to what the current solution is so you know if you're not going to have as many car wrecks you're you know you might not want to set the criteria at 0 then again enforcing good behavior or understanding what the liability regime will be that's probably why autonomous cars the US and certain respects would be one of the last places in the world that you'd you'll see very widespread views because our sense of liability and our desire to preserve the status quo if there's any chance that something might be even in in a framework considered a step backwards that's that's very tricky the place that I think this is is most concerning is in is in weapon systems in the medical field you know we just don't have doctors most people are born and died in Africa without coming near to a doctor and so there are definitely things like we're doing a lot of work with analyzing ultrasound and we can do things like sex blind the output because we're not having anybody actually see the image we can tell you what's going on without revealing the gender which is of course when you do that it drives gender sight and yet we're doing the analysis the medical understanding in a in a much deeper way and that's an example where it's all done with a lot of machine learning that was meeting with the guys at Google are helping us with this this morning and there's some incredible promise in that field we're in the primary healthcare system the amount of sophistication to do diagnosis to understand for example is this a high-risk pregnancy yes let's escalate that person to go to the hospital level even though you couldn't afford to do that on a wide widespread basis so this stuff is going to be very domain-specific in some domains like education I'm more worried that the privacy concerns which are appropriate they're good privacy concerns but if you don't put a lot of creativity and how you have longitudinal data access while not violating privacy you're going to default to the data sets not being there and in the u.s. education day that is the default that there isn't much information that would allow you to find positive exemplars either at the teacher or school or district level and therefore really examine what what inputs are allowing for that unusually positive performance so if you're thinking deeply about technology and ethics are there any things that you think in retrospect you might have wanted to do differently in your time leading Microsoft or any lessons you learned thinking about that post well certainly the really profound societal changes from personal computing are really just beginning and so we didn't disrupt the way that people get news or or communicate it led you know that PC led to the Internet led to the cellphone led to social media today and so the the awareness that once you had made that access to information including information that stimulates you or that you agree with and that you could cluster in that way there wasn't a recognition way in advance that that kind of freedom would have these pretty dramatic effects that we're just beginning to debate today you know the a lot of the personal computing early period we were worried about the so-called digital divide that is that the computers would be available to the kids who are better off and accentuate rather than reduce now at a classroom level the actual data about the value of computers in the classroom is essentially nil so that's good we didn't create this gigantic dividual divide that is the schools with the computers are just as bad as the schools without the computers which in an absolute sense they're quite bad and you know so sometimes you get false positives when you worry about because you think your solution is so incredibly magical there are things in terms of internet access you know getting that out to rural areas getting it into parts of Africa we still that's a unfinished agenda that through a variety of you know cheaper cell and antanas so-called whitespace type access I do think that that general connectivity issue that we've been working on for over 20 years largely will will become a solved problem and I hope that computers prove to be very valuable in classrooms so that then we do have the need to get them out on a very widespread basis but only at the individual levels do you see in in terms of the highly motivated learner do you see that it really has changed the learning outcomes and that's only in say the top 15 percent of the highly motivated learners hi my name is Ron Lee I'm a physician and focusing on integration of AI and clinical processes for the healthcare system at Stanford we often think about in medicine and other fields of relying on AI to reduce error and even in medicine we have seen algorithms with error rates that are lower than that of the human but at the same time when AI an AI system makes an error that you effects on society but also just how society perceives that error is very different than when a human makes an error so doctors make mistakes all the time but then when you have some AI system making that same mistake the reaction is very different so I wonder how do you think about this dichotomy and it effects on how AI would progress and be accepted by society yeah a good example of this is another group that the foundation founds has done work where you just use a cell phone camera to take a picture of a woman's cervix to predict whether she has cervical cancer and that you should intervene and the results the National Cancer Institute is very engaged in this because the results are dramatic compared to the very best humans and a core the typical humans particularly should get out into developing world settings are either not available at all or their performances well well below the gold standard which we were able to exceed here and so certainly on those image recognition things you know that's getting to a point of maturity and hopefully that it will become accepted one thing that we're going to have to build in is a feedback mechanism that is when the algorithm makes a mistake the ability to take that training set and constantly improve it because something new may come along that the original training basis it wasn't good enough and completing that circle even in the u.s. that's a very difficult thing to do when you're out in rural Africa and you don't have these electronic health records to say okay you tested this person you told her she didn't have cervical cancers and an interesting that three years later she died of cervical cancer you know let's go back and look at those images so you want to you want to complete that loop and as usual if you have you know negative consequences for mistakes it kind of discourages that completing the loop type system there are a few cases like in in civil aviation where the willingness to look at mistakes and apply massive resources like we're seeing today to say okay what went wrong here it really is pretty mind-blowing and of course you have software based elements including the the 737 max case and so we we through software driven surgical tools software driven flight tools software driven weapon systems we are accumulating a sense of understanding it is fairly troubling that today's deep learning systems are mostly opaque and so one hopes that sometime in the next decade somebody comes up with AI systems that are both as good or better than what we have today and yet have a degree of explained ability including the sort of strange false positives that make absolutely no sense to human cognition that still for most you know take the visual side of these things do trigger in a way that is is somewhat it would would not have been predicted it is impressive today the FDA is taking in diagnostic tasks there's three that were given the early approval where there is this notion of dynamic improvement that will go on and that will actually have more impact in the developing world for things like tuberculosis malaria you know we just have way more lives to save than the u.s. does I mean if everybody in the u.s. you know live to a hundred it when would not match what we can do in the developing world in terms of the net changed human benefit so it's nice that it gets piloted here but a lot of the impact is where you don't have the human comparator is is not at all what what we take for granted my name is Elia I'm a Stanford PhD student and I'm curious about like so in the past few weeks we just had a lot of rains in Northern California and it's hard to imagine like here we have been suffering from droughts in the past few years now we are suffer from the impacts of floods so I'm cursed like in terms of AI how do you think we can make use of AI to help those who are affected by floods or other kinds of natural disasters and allow the whole society work take together not just within Stanford bubble but also make the huge impact and work with the citizens together thank you yeah the last time I was in this room we were talking about climate change and of course you know climate models are extremely imprecise unfortunately AI alone will not make those models precise the amount of data that you would need to really understand you know over a period of months or years weather conditions requires a really pretty unbelievable amount of data you can sort of prove it because the the the huge nonlinearities in in the system they're these systems can't improve to some degree and so if you can predict floods if you can predict out that's good we do know that we need to make our systems more resilient because climate change we do know brings higher variance and so if you're a farmer in Africa today a subsistence farmer which is seventy percent of the people who live in abject poverty today you get about one out of ten years where your crop completely fails and you need a buffer stock or government programs it appears it'll get by the end of the century to about one year out of four now in the Western world you have savings you have government's with tons of money as long as your gross productivity isn't going down substantially then you just you know are able to cover that year if you're a subsistence farm or what it means is that your kid is getting so little nutrition that if they survive they are they are permanently damaged and so it'd be great you know to have the very best AI work in the very best weather model America and the data collection the that is actually a huge limiting factor is how you program up the resolution of the initial conditions including in the in the ocean the hardest piece being the biosphere because you have very nonlinear reactions to weather and heating within the biological systems including the biological systems that are in the ocean so I wouldn't sit here and make some fantastic prediction that we will be able to model out those negative things you want to have a lot of extra resources you want to be agile about bringing those extra resources to bear primarily in equatorial regions where you have subsistence farmers and and the world is not very good at that today we have the World Food Program that does some of those things but if you if your your figure of Merit is avoiding malnutrition when you have negative weather variants we do a very very bad job with it hi my name is Layla I'm a student at Stanford does it concern you that AI talent and innovation is concentrated in a few big tech firms and universities and if so how can we encourage more competition yet the in a sense when you have competitive something that's competitive where somebody's ahead of other people and is at the state of the art it's not normal that you'd have lots of people who are at an identical position you know so you take like designing nuclear weapons we didn't have like lots and lots of places in the world that were the equivalent of Los Alamos you know there was what we did create the competition with Lawrence Livermore labs but just to have a tiny bit of diversity there and so there's I think you have to draw out did it yes we should we should draw more universities in and universities in general are motivated to think more about societal benefit than the private sector and so it would be unfortunate if the universities fall behind and so it's great that Stanford is putting together these initiatives and there are even questions about access to cloud computing power that matches what the private sector has and how we're going to make sure that you know Stanford and hundreds of other universities actually can run datasets I mean for example if you want to look at bias in word embeddings you better be able to create the state-of-the-art word embedding system and have access to play around with that system and unless we're careful the the private sector will kind of run away not just with smart people but also with the ability to to do super super complex models so yes it'd be good most advanced technologies in the u.s. post-world War two were created as part of the military industrial complex and therefore the u.s. in terms of its application to weapons and sort of on the government itself being involved at an early stage to think through what these things mean it was natural that the government was seen it partly through wearing that defense related thinking now that these AI technologies are completely done by universities and private companies with the private companies being somewhat I had the government just doesn't see it in the same way that they did with with previous technologies and you know hopefully things like humanists your your Institute will bring in you know legislators and executive branch people maybe even a few judges to get up to speed on these things because the peace and the global nature of it and the fact that it's it's really outside of government hands does make it it particularly challenging the US was in this totally unique position for most of these breakthrough Technol oh geez and now yes the u.s. is still very much the leader but not in a the same dominant dominant way that you can be sure say hey ten years from now will the best AI that does reading and scans the scientific literature to look for biological advances will that will that be best here in the United States or might it be in other locations very hard to say and even the definition when somebody says to me is China ahead on AI that's an ill-defined question because there isn't a boundary where you know there's oh that's Chinese AI oh that's that's us AI the you know for Microsoft we have a lab in Beijing Google has a lab in Beijing some of the best AI work in the world is being done across the street from Shanghai University now what kind of AI is that its global AI and if people started thinking of this in terms of nation-state terms and try to draw boundaries that's going to be potentially difficult and potentially quite problematic I'm afraid that's all the time we have for today please join me and thank you mr. gates and Stephanie her [Applause]

Channel: Stanford

Views: 52,377

Keywords:

Id: Bdaq-KlyfLQ

Length: 57min 48sec (3468 seconds)

Published: Wed Mar 20 2019