Imported from Digital Frontrunners 2019
David Rowan on AI - Digital Frontrunners 2019
1,828 views
View transcript
[Applause] welcome inside please welcome to the stage to be David Rowan to pronounce it nice to be here thank you humans for joining us because we're gonna take a little journey into where the nonhuman is taking us the smart machine that is advancing at an exponential rate and I'm gonna share some of the reasons I'm pretty excited and some of my anxieties that mean collectively we have to work out the value code that we want to impose on this world we are creating so I am a journalist I travel a lot I get to know a lot of startups and I'm constantly shocked at how quickly smart machines are advancing it's everything from computer vision to large-scale data analytics to sensors understanding sentiment by monitoring all sorts of data outputs and it's got to the stage now where I've kind of stopped being surprised because I'm expecting everything to be a bit strange a few weeks ago a bunch of Russian scientists showed what would happen if you took a two-dimensional image and trained it with a neural network so you can see in three dimensions what the Mona Lisa would have looked like if she was on Instagram doing video selfies we've now trained the smart machine by feeding in the data of typical United Nations speeches to generate a fake speech that is as authentic as anything that you hear in the General Assembly the Security Council very quickly we have startups that are quite well-funded that a taking talent from some of the best university labs and creating iterations on what we already knew in very creative ways we for instance are able to simulate how people are speaking by manipulating video and translate in real-time there's a startup called sinneth Synthesia that has taken a footballer who is not famous for being super smart but he's very good at kicking a ball and they've made him sound beautifully polyglot malaria isn't just any disease it's the deadliest disease that's ever been sir isa al-ma tomas telling me that the policy only exists in video you don't and would it be well my zealot O'Laughlin could literally retain will approve of the metaphor pretty quickly and obviously you know there's a dark side the fake videos did you see this what about my point in time even if they would never say those things so for instance they could have me say things like no kill monger was right Ben Carson is in the sunken place or how about this simply President Trump is a total and complete [ __ ] he may have said that but he didn't say that on this video public address so we thought creativity was the last bastion of the humans but even creativity is now being understood by the smart machine so we've had music getting better and better we're now starting to get the plots of Hollywood movies based on sucking in all the data based on what the audience responds to we now have companies like Nvidia showing you how to be a great artist you just do a sketch on the left and it takes the database of image and makes it work on the right and this is of course hitting real Industries retail Amazon about three years ago showed its concept store which uses computer vision proximity sensors you check in with your phone by scanning it but there's no cash desks everything you put in your basket the store understands so you can go out without paying and this store a lot of other retailers mocked saying it's just a gimmick by a West Coast tech company and then last year they announced they were opening three thousand of these stores across America a few months ago they opened one in New York and suddenly it changes the rules of retail now if you are giving people the friction of forcing them to pay for something in the store then you're falling behind so everywhere I'm looking there are interesting developments so when you take deep learning reinforcement learning the ability of the smart machine to create a real-time simulation and you get interesting projects this is a project from the University of Berkeley in the West Coast it's called blue they worked with open AI to simulate what would happen if a robot moved its arms in various directions in order to see how the role would be helpful doing your laundry at home and we're just at the very early stages of functional machine learning with low-cost robots I don't think we can keep still you can't really afford to rest because it's moving you've seen the videos on YouTube of Boston Dynamics the dogs that go to the battlefield every time I look now I don't know if they're making any money but they're making some great videos this is the parkour video and it's simply learning based on simulating and based on doing together with the sensors inside the robot itself so it can get a sense of where it is in three-dimensional space so I guess probably the area that I'm most excited about and that's going to affect us in our lifetimes more than lots of other industries is healthcare because we've relied so much on the human expert the revered doctor and of course humans can only take in so much data and there are research studies published the whole time this is one recent research study published in Nature Medicine and for it's complicated journal title what it's saying is the smart machine can learn from your ECG to spot arrhythmia better than the human experts I mean this is one of the graphs in the article the little red dots are cardiologists and how sensitive they are at spotting things and the blue curve is the algorithmic model so it's actually better in many cases of spotting heart defects than the human and every week there are new journal articles published that show how health care is being rethought by the smart machine I think we need to celebrate this it's going to prolong our lives it's going to help us live a higher quality of life especially as we get older the Machine is better than the human radiologist at spotting tumors it changes the nature of the human job you still need the human but maybe that role is less about doing the physical legwork of spotting and more about advising the patient based on the knowledge as I think as we're gonna hear a lot more startups even in Copenhagen doing some pretty effective things in healthcare so if you add the ability of the smart system which is able to process more and more data and we're just about to open the quantum computing era which is gonna create new drugs through simulation in silico in the machine before it is actually tested with molecules in the lab it's gonna affect all sorts of sectors it's gonna affect transportation so we're already having startups like this is lilium in Germany the vertical take-off electric jet able to do things that until recently couldn't be done and this is just one of a bunch of startups doing autonomous vertical travel there's another one there's a zipline active at the moment in places like Rwanda using these drones that were able to self pilot delivering medical supplies among other things in places hard to reach by Road and what we're entering is an era where everything about the physical world is just another data point and the friction that we had the inability to get to a place to deliver to a place to know what's happening on the ground is diminishing there's a startup called planet and the west coast that's putting thousands of these nano satellites this big in constant orbit around the world creating real-time availability of imagery to show what's happening on the Earth's surface so for instance when Apple was building its Cupertino headquarters and Apple didn't want cameras crying it couldn't stop the subscribers of Planets feed getting access to how the building was developing and this is all data that we've never had before but suddenly you've got satellites we've got drones we've got all sorts of other data outputs that are creating new ways to understand the world which means pretty much every business has to start thinking of itself as an artificial intelligence or data analytics business even McDonald's now is seeing itself as an AI business it spent three hundred million dollars acquiring an Israeli company so it could understand better in real time what its customers wanted as they approached the drive-through so I think we have to stop assuming that the way we're doing things today is going to be the way we're doing things tomorrow if the Machine increasingly understands not just what we're telling it but our sentiment our emotion the response of our voice our facial expression and there's an upside to this which means the machine is more empathetic it's more intuitive about what we want there's an example an austere newzealand startup called Sol machines that's developing AI customer service BOTS but they're not BOTS as in the instant messaging BOTS these are BOTS that are computer-generated people that are connected to a system that has a microphone and a camera and responds to your expressions so these are not videos of people these are the company sol machines CGI computer faces and they respond different languages Wayne don't war Ville common in Deutschland and the founder used to work in Hollywood I'm doing CGI for movies like conch he won a couple of Oscars and but he's now developing these partly for healthcare advice partly for governments to interact with citizens partly for customer service and it may be a bit spooky but I suspect this is one of the futures so there's a lot of ground to be optimistic there's an awful lot happening talked about healthcare talked about monitoring the natural world so we can predict earthquakes so we can understand whether of course a huge amount of investment money's going in to autonomous transport which will save lives fewer people dead because of human error but it's hitting every sector and I feel we're just kind of the foothills of what's going to happen so often um games projects early to see what's possible and there's a bunch of open AI projects involving games training the network to learn to play better games you've seen deep mines alpha go beating the worlds best go champion but in the last few months there have been a bunch of other projects that have outdone the humans this is a game Montezuma's Revenge and an algorithm called go explore kept getting better and better [Music] it is easy to dismiss games you can start to see how the machine is starting to write in a compelling way so you can log on now to a website called transformer I was playing it at 6 o'clock this morning starting to suggest the beginning of a sentence and it will complete the sentence for you in real time and it's not great news for people who write for a living because the machine is gonna be pretty good at writing but it's kind of interesting so it's not quite there but the more data it can suck up and learn the better it gets the novel that's compelling isn't gonna be that far away so where does this lead there was a cover story in American Wired in February written by Kevin Kelly and his proposition was we now have so many sources of data in real time we have so many cameras billions of cameras out there storage is pretty much cost free so we're going at some stage soon to have what he calls the mirror world which is a simulation of the physical world because it's just so easy to capture what's happening in the physical reality and to make it searchable so if you want to teach kids what it's like in the rainforests of Brazil where the fires are starting you can take them there augmented reality virtual reality is going to help because you've got all these data points and we've kind of seen some early proof points you know the Pokemon go speed at which it became a mainstream game capturing a physical reality and embedding a virtual reality but we have to think about where we want this to go because it's not all happy shiny aren't we a great startup news it's also gonna be troublesome just as emails incredibly useful it also gave us the problem of spam what's gonna happen when we have this virtual world embedded over the physical world a filmmaker Katie Matsuda gave her an example he made a short film about what it's going to be like going to the supermarket where you've got all these messages customized to you so I am ran wide in the UK for eight years until quite recently and I'm constantly kind of going to visit the university research labs and the startups and trying to understand where this is going and pretty soon we're not going to have conferences that are devoted to artificial intelligence just as we don't have conferences devoted to electricity because it's just going to be part of the infrastructure is going to be embedded and if you see what's happening sector by sector AI is being normalized in finance there are so many startups putting machine learning into things that weren't there the day before yesterday this is CB insights with just some of these startups using AI for different reasons for personal financial assistance for predictive analytics to help with detecting fraud and regulatory compliance to help with credit scoring to help in asset management everywhere you go the ability to process real-time information at scale and decide what to do with it is getting very interesting I mean certainly by a hundred years out but I would guess by the lower numbers as well I think we have general-purpose artificial intelligence that is more capable than humans at all are almost all economically valuable work this is Sam Altman who runs something called open AI which has taken quite a lot of money from the industry to kind of push forward the knowledge on a and make it widely accessible so his opinion is that we don't know yet when but sometime in the next century we're going to develop a general artificial intelligence that we better than the human at doing most tasks um which has its own risks partly we haven't really worked out how to constrain whether we can constrain this general AI and it may not do things that are in our interest and one of the problems at the moment is the people coding this magical future have their own inherent biases there's some obvious biases gender biases so these are some of the big research papers over the last couple of years and almost all of them are published by men who have certain assumptions we have cultural biases if we're hiring people we're gonna choose people who think who speak more like us we've got some startups already using artificial intelligence to try and counter these sorts of biases companies like biometrics that try and use neuroscience online games that stop the human making that decision they just monitor in real time but I think we have to accept that the Machine the smart machine is culturally determined and it's not inclusive there was one research paper quite widely reported recently about the ability of the network to identify people's gender from thousands of photographs it's pretty good for light-skinned men it's not so good for darker skinned women so we're getting some nice cultural commentary based on some of these prejudices there's growing computer vision recognition tracking us tracking our faces tracking our devices there's now a fashion company adversarial fashion that is designing and selling clothes designed to confuse the cameras if it sees these patterns on your clothes it doesn't know how to respond there's girls versions first but I think if we really want to know where some of the risks lie we have to go to China which has declared its intent to be the world's leading AI power within the decade and a combination of the state plus the universities plus private money are focusing their efforts into understanding pretty much everything and processing it for tracking and I guess this is my version of the Chinese flag and some of it's quite fun and a bit harmless there's the AI newsreader hello everyone I'm an English artificial intelligence anchor this is my very first day in Xing wonders agency my voice and appearance are modeled on Jack Jiao a real anchor with Tsinghua the development of the media industry calls for continuous innovation and deep integration with the International advanced technologies and then there's Education if you can track individual pupils performance that becomes quite palatable is artificial intelligence based adaptive learning system issue Education Inc realized the five times of learning efficiency for middle school students compared with traditional approach and students need to take an intelligent knowledge diagnosis assessment at the beginning of studying and this ally adaptive test is to quickly and accurately detect students knowledge witnesses but this is where I get a bit more concerned this is the face recognition in real-time this is a company called cents time there's a bunch of these companies that have become supremely highly valued and the story they tell is will make your life simpler IVs are you able to get into your office building just by looking at the door and yeah sure it gets rid of friction but it's the little ways in which these sorts of technologies inside China are being used to exert control that are pretty worrying sometimes you know it's a bit quirky the machines that make sure you are not taking too much toilet paper the face recognition toilet paper dispensers that are turning up through bathrooms throughout China but this is where it gets a bit more worrying Chinese city of Shenzhen but unlike most other mega cities across the globe you won't find many jaywalkers here that's because this city like others in China uses facial recognition software to dissuade pedestrians from crossing when they're not supposed to using strategically placed cameras at Troublesome intersections the city is able to detect jaywalkers and if you get caught then your image gets displayed on the nearby screen and I'm fine it's sent right to your cell phone it's one thing getting fine for crossing the road in the wrong place what about if you're an ethnic minority in a place in the northwest of China in Xianyang where the government is effectively jailing more than a million of you because it's scared that if you're Muslim maybe you'll threaten the unity of the state in Xinjiang province the Chinese government is wary of the separatist threat posed by the Muslim weaker population according to local NGOs an estimated 1 million weekers are being detained indefinitely in secretive internment camps why some are being subject to abuse it's been called the largest mass incarceration of a minority population in the world today the authorities are using facial recognition cameras to scan people's faces before they enter markets total decision world median jeongja to our ship tankers a sashimi and told his old shoe EMU to the fire you greeting me with other each year chisel I think this is the moment to ask whether we want these technologies to be used in similar ways here whether they threaten our values because if we don't have this conversation there are some powerful people and some big tech companies that are going to have it for us and we're starting to see the risks of not enough public debate so Google owns deep mind an AI team and it's gone into a bit of trouble lately for working on a health project with hospitals in North London and not getting active consent from the patients for their data being tracked Amazon too is getting very active in health care but there's a lot of fuzziness about what actually they're declaring some of these startups are starting to get big rebellions among their own team so there's a company based in New York will clarify that's developed a very effective face recognition tool and then they started saying we're going to work for a US government project project maven because we think we can be useful the staff rebelled some of the staff the boss starts to post matthew sailor why we're doing this the pressure grew he gave an interview in which he said ok ok we're going to have a full-time role thinking about the ethics thinking about minimizing bias in our work it seems that role never materialized people inside the big tech companies because of this pressure and now starting to respond in Amazon they're talking about now a role focusing on algorithmic fairness we're gonna hear a lot of these phrases I think it's Salesforce they've talked about having a chief ethical and human use officer Google now is still facing the wrath of quite a lot of its employees they're increasingly public in the American edition of Wired this month there's a big feature looking at how this debate is starting to damage the core of the company and there are new phrases being used by Google staffers one of them which we may hear more of you've heard of greenwashing will efex washing is when you imply that you have an ethical process in place sundar the boss at Google posted the AI principles they're operating by things that they will do they won't do we're not developing AI for use in weapons but we're going to still work for the government and military and it's not been resolved there are people who are leaving these big companies and saying well they're pretending to be guarding ethics morality but they're the ones who are unaccountable so thankfully the debate is coming into the open there are academic and semi academic institutions that are starting to think about the values the European Union a few months ago issued its own recommendations for AI guidelines and a few of the things it's asking for a human oversight a safety protocol transparency so you know exactly how code is being developed what the code is accountability and of course this but it's hard to know how easy it is unless we get more involved now we demand this accountability this fairness it's hard to know whether we're going to achieve what we want and avoid the Chinese vision of total surveillance and everywhere you look things are moving I just wrote a book which is looking for innovation around the world proper innovation inside big organizations not just the big tech companies and I'm seeing the application of AI in the most unusual places as a fertilizer company in Norway that had problems delivering its fertilizer from the factory to the ports for exports so they decide to develop the FIR of the world's first autonomous electric cargo ship this is a fertilizer company going into the autonomous transport business this is a model the real one is coming into use in a year or so I was in California where in this barn a bunch of michelin-starred chefs are making recipes that are being filmed for an app and they're employed by one of the world's biggest manufacturers of saucepans based in Hong Kong this is a company called mayor mayor realized because everything is now connected the offline saucepan is soon going to be redundant so they're developing an online saucepan connected via bluetooth to the app with the recipes in that these chefs have designed that connects to a conductive heater below the saucepan which has a temperature sensor embedded in the middle of the metal so it controls how quickly and at what temperature it's heating up so you can cook alongside the michelin-starred chefs the smart kitchen is about to replace the world of offline sauce pans so I'm gonna review I guess with optimism tempered by realism so technology sensors the smart machine robotics it's going to improve our lives you know the baby who is born deaf and can have a cochlear implant implanted and can hear for the first time is a wonderful thing to wit but I don't think we should just leave it to the people making the tech to work out what the use cases should be likewise I didn't think we should resist these technologies because they're coming whether we want it or not and we're biased as human beings we will dismiss things because they're uncomfortable the first time you are in a completely autonomous car it's gonna be a bit a bit weird bill Rimmer put his mum in his Tesla in a film does he set it to autonomous mode but you know like the next time is just gonna be another way to go and play cards with her friends so I'm gonna leave you with a certainty that as flawed human beings you're just one part of this equation but if you don't demand the same now you may get left behind thank you for listening [Applause]