You and AI - with Jim Al-Khalili at the Manchester Science Festival

[Music] hello good evening ladies and gentlemen I'm Jim al-khalili I'm a professor of physics at the University of Surrey and I'm also a fellow of the Royal Society so thank you all for joining us for this this is the penultimate in the Royal Society's year-long series you in AI so yes penultimate event in this year-long series of events we've been holding around the country Unai which has been exploring the world of artificial intelligence and opening up the conversation with leading experts on how a AI is going to affect all of our daily lives in the coming years now the Royal Society has played a part in some of the most fundamental historical life-changing discoveries in science for over 350 years and the Society's fellowship of esteemed researchers continue to make outstanding contributions to science in many importance and diverse research areas now in April of last year April 2017 the Royal Society launched its landmark report on machine learning based on extensive research into public views on AI and engagement with industry and with policymakers this called for action in a number of key areas over the over the next five years over the next 10 years to support safe and rapid developments of machine learning and artificial intelligent technologies it also called for an informed public debate essentially while we're all here this evening about the development of machine learning and how its benefits can be distributed across society so the Unai series aims to help develop this public conversation about what machine learning and AI are how these technologies work and the ways in which there they may and indeed will affect all of our lives from health care and transport to scientific research itself it's clear that there's a wealth of potential benefits to be gained from the effective use of AI but there are concerns as well and I'm sure many of you have them and I hope we will explore some of those this evening about how these technologies might develop how they're going to be used who's going to use them who's going to control them how my a I effect the job market how can we create trustworthy AI systems how when would we be happy or indeed unhappy about an AI system making a major decision by itself for example when a prisoner might be released on parole or what medical treatments are prescribed to a patient this is coming AI will be able to make those decisions without human intervention how do we feel about that if we want to bring the benefits bring about the benefits that a I can deliver both safely and rapidly then we need a broad and well informed public we need well informed public debate about how AI technologies might affect us and how we as a society want to use them or indeed not use them this we hope it would be up to us in a democracy part of this is having a forum to ask questions and debate the answers which is why we're all here this evening with me this evening as some of the UK's leading thinkers on AI and together we'll be discussing ai's capabilities and frontiers the ethics behind the technology and the potential impact on jobs as well as the implications for society as a whole we're also joined by these 30 small guests they're already here on stage with me I think they're already on for this various flashing lights as the odd one fidgeting here and there which is sort of a bit disconcerting and I also want to make sure I don't walk backwards and squash one of them because apparently they're quite expensive we're going to hear about them a little bit more later on and of course we've been asking you for your burning questions on artificial intelligence which I now have and I will pose to our panel of experts before I formally introduce the panel I would of course like to take this opportunity to say how delighted we are to be here this fantastic venue the Royal Exchange theatre also we'd like to to thank the Manchester Science Festival program this event is part of the Manchester Science Festival programme and in particular we'd like to thank deep mind the company who've kindly supported the whole you and AI series right then so without further ado let's meet our panel thank you right so some brief introductions first of all Dame Wendy Hall Regius professor of computer science at the University of Southampton executive director of the web Science Institute and a fellow of the Royal Society professor Neil Lawrence chair in your own computer science at the University of Sheffield machine learning researcher at Amazon and a member of the Royal Society's machine learning working group and dr. ever lugar Chancellor's fellow in digital arts and humanities studying AI epics at the University of Edinburgh another round of applause please so I want to start with it's sort of a generic question to sort of get the ball rolling with that I want to ask all of you and it's what does AI mean to you come to you ever faster okay um so AI I suppose if we think of all then ticket elegance as being human intelligence then artificial intelligence would be the intelligence exhibited by machines so machines that adapt learn reason that's what kind of thing Wendy hard work i co-wrote last year turned 17 I co-wrote a review on AI for the UK government to do with how we use AI to help grow the economy create jobs and since then the phone hasn't stopped ringing or maybe more the emails haven't stopped coming in and the invites because we're in this hype curve at the moment of AI and I'm going to tease that out during the panel but it's it's how we manage the people's expectations during this period of hype about AI you know so I like David's answer separating the difference between human intelligence and artificial intelligence I don't necessarily know what it means to me anymore because everyone tells me it means so many different things but maybe to define intelligence for me what intelligence means is that you use information to reduce your resource consumption so I can do something in a more efficient way if someone gives me information I know that's not human intelligence but in terms of what we're trying to do with AI I think that's very often what we're trying to do is you have a goal you're trying to do it more efficiently by using information I mean in a sense this is really I guess a continuation of the Industrial Revolution we've always used machines to do jobs more efficiently more quickly more intelligently than us but now I suppose it's it's not the mechanics that we're making use of it's the it's the software it's the information now for I guess for a lot of people when you you think about AI and when we're sort of all infected by Hollywood movies and so it's often thought of in in terms of futuristic sci-fi terms so what is AI doing for us right now so I think this is an amazing spectrum one in one part you see the reference when you were saying about the hype AI is just extreme statistics you know there's a lot of parallels between machine learning which I've worked in in statistics it's just the emphasis of what we're looking at was slightly different so um I mean it's it's doing things around but mobile phones intelligent agents in a recognition of people the first thing from the came out of my community was that little box around your head when you you take a photo of someone identified where the face was we were shocked wow we've done someone someone from it but today it's everywhere whether you know it or not my favorite application that I've worked on is was actually in Africa trying to better understand the distribution of disease it was very simple systems but it was allowing us to inform the Ministry of Health in Uganda about where to target drug transport so there you see use the information use your resource more efficiently by putting the drugs where needed in the right place and we're gonna see this happening more and more I mean when do you say this we seem to be talking a lot about AI at the moment and that's because the technology over the last even year or two is moving so rapidly well it isn't just the technology is the fact that there's a lot of data that's the big impetus behind this AI revolution there have been for AI waves probably and there's a usually a winter in between each of them a hey I started 50 years ago or more in the 1950s people started talking about good machines think and since then we've had waves of development and the impact of research is often 20 years after the research is done the universities so what we're seeing today is the impact of research that was done maybe 15 years ago but we have very fast computers and we have a lot of data on which to train the algorithms and I think that's the major impetus behind what we're doing today but there's a lot of expert systems work in many industries and expert systems were developed in the 60s and 70s so you know we're seeing the impact of what's sometimes called good old-fashioned AI we're seeing breakthroughs with machine learning deep learning which are again based on research that was done 15 or more years ago but the difference is we've got a very fast computers and a lot of data and I think we're going to talk more about the use of data and part of that is just increased computing capacity in memory and the fact that we've got somewhere to store all this yes yes and of course with all this data ever comes the issue or the ethics you know who controls it whose data health- want one day presume you're gonna have our genomes or maps and personalized medicine so so where are we heading with a eye on maybe on the ethics side today well I mean ethics is something that's getting a huge amount of attention at the moment you know everybody's super interest in it most university departments will be now hiring people around ethics industrial labs are hiring ethicists so there's a lot of thought going into this but I suppose like my primary concern is that this tends to be at the moment a sort of a push towards the idea that there's one ethics for everybody so if we understand ethics as the rules that govern our moral conduct morality being what we consider to be right or wrong then we can sort of a sense immediately that that's there's not just one you know if I live in China my sort of ethical reasoning would be different from if I live in the UK so trying to find one set of rules that we all did here too is going to be incredibly problematic so one would hope that artificial intelligence could adjust its moral reasoning in accordance but we're actually you know we're nowhere near that so these are sort of the challenges that will come up and how about things like putting regulations in place you know about you know who controls the data you know even as almost as a separate issue from the ethics yeah the rules about what you can and can't do the rules about you know transparency and so on yeah absolutely and regulation is really important and we're seeing people being enormous ly concerned about this the EU general data protection regulation that came into effect in May this year has sort of started to talk about algorithms including I mean it's it's a bit contested in terms of what it means but the notion of a right to explanation that if you feel that an algorithm has made a judgement about you that you're unhappy about all that it causes harm you're able to push back against that so we are thinking about this I mean the importance of governance of ensuring that the data is managed appropriately that it's not shared inappropriately that we think about harm right there's sort of the starter system design it is incredibly important and I think regulation does play a role but we had you quite rightly pointed out we do need to have a sort of a separation between what we understand to be the law which should be the basement that's the least we should do to what epic so I should morally do and that should be what we're aspiring to so you know but that's the kind of distinction that I see though right so I would now start on the questions that have been sent in by the audience and so the first one comes from Jack who asks which sectors or industries do you see as benefiting most from AI and machine learning and if you were to draw a timeline how far along it we in terms of uptake yeah it's a great question and probably goes to the crux of the issue my feeling is it's a bit like asking what sectors are going to benefit from the telephone or the mobile phone the computer or the Internet it's a very pervasive technology and it's difficult to predict because humans are so imaginative in terms of what they can do with technology but I would broadly say that there's two things that we'll see happening because just I'm just copying what happened in the past there'll be new businesses we just can't even imagine or can't even think of like who would have thought of social networking in 1994 you know and an internet search for these are the largest companies in the world but I think things get really interesting when they change the existing business things that will always be here like transport and health logistics and that's much much harder because those systems have evolved into something that makes use of humans as an integral part of the lube it's not easy to just take them out and put in automated agents and there are many many issues and my feeling is you'll see two things you'll see things happen very quickly in terms of new innovation and all the imagination of all the wonderful entrepreneurs we have and then things will have much slower than people expect in those sort of more traditional sectors like transport and health and you know for me health is what I'm really interested in but everywhere but at different speeds and perhaps sometimes not as fast as people think I it's happened time and again with with technologies that replace humans and we went through sort of the you know with the silicon chip in the in the 70s and and certainly it did replace jobs you know on factory floors production lines for example automation we're now talking again about robotics automation with with AI coming in but I get the sense that what's different now is we simply don't know how many jobs are going to be affected or you know what new jobs might and we've never known what new jobs in advance might be created with the new technology that we can't even dream of at the moment but we still seem to know the extent of how far it will push humans offering the offbut of the job market I I think that um I mean we're sitting in the Corn Exchange in Manchester surrounded by buildings that were industrial looms and are now flats where I professor in Sheffield the effect on the steel industry you know the devastation actually the north suffered as a result of the changing nature of work I'm not actually sure I don't think anything like that scale or even the industrialization as people move from the land what I do think is really really difficult to replace is human contact and what I hope happens is you know I quite like hipster coffee bars and all this sort of thing where people are doing things you know computers can't do that you imagine walking into a hipster coffee bar and having the computer say this is single origin prepared by a computer in Guatemala it's not the same my hope is that that the world around us will allow us to do more of the things we feel as humans what we've seen in the past is we've had to adapt to the computer not the computer but the form of automation we have to get up when someone you have to come into work and put a punch our time clock we have to do everything to make the machines more efficient well wouldn't it be wonderful if this revolution was centered around us and centered around what's nice for us rather than what's nice for the machine as the previous revolution so that's my sort of optimistic hope the example I always give my father was an accountant well actually he was a bookkeeper and by the time he retired in the late 70s he was just moving on to using calculators that he'd done everything by hand up until that time huge columns of numbers long division of pounds shillings and pence all done by hand and that you know copying Ledger's all those jobs have gone everything's Joan's been completely automated and the banks have disappeared off the high street but the finance industry is bigger than ever before and this is because of the creativity of human beings because when the Jeanne's can take the repetitive jobs then we can use our creativity to think about new ideas and this comes back to your idea of someone said taste and knowledge it was something you said earlier and you know getting new knowledge as well I starting to do that yes and this is happening in health for example creating a new creation of new drugs potentially using AI using AI to diagnose cancer earlier but I do agree with Neal that human contact is important I think coffee bars as we know them will disappear unfortunately they may be but I I'd the one the thing I the Japanese have put a bet on robots helping the elderly right they have a much bigger problem us we have the same one but they have a much bigger problem of too many old folks that need look caring for and I worry that we I don't think we can use robots to care for old everything that people's needs are and I think human contact is very important we're miles away from general AI you know the robots do what they're programmed to do at the moment they they can't think outside the box they can't be they don't have understanding they don't have empathy they only know what we tell them there may you know we need another whole wave of research development to be able to develop general AI robots that care and empathize and by that point we need to have the ethics of the morality sorted out I do think though I will something that some I have a lot of I think we can use this revolution to think about the social contract I mean we take for granted the fact that on the whole we have a five-day week in a weekend well this revolution could lead us to a four-day week and I'm not saying this as a you know on a political stance I'm just saying it could we could we could and then and we could spend more time therefore with our families and we could use it in that way if we're clever but we've seen industrialization doing that for us anyways yeah try you know in 50s housewife and say what you can't have this washing machine anymore what you have to spend a day doing what I mean it's opened up and we take the weekend for granted our Victorian grandparents did I think kids worked for seven days a week in those days you know I mean the society has changed dramatically alongside the the Machine Revolution right I'm as I'm listening to you I've got one eye on these little monkeys and and some of them are sort of stretching out to the to the periphery of their little den so it's probably a good time maybe to give them center stage literally so here to demonstrate one of the technological applications making use of machine learning in AI and we can welcome him there on to the stage is Danielle Kerry or sapota from Professor Sabine how it's lab at the University of Bristol Danielle I would like to start with a question have you ever wondered how ants are so good at finding your picnic table they even create the shortest train between your foot under nested that's an example of swarm intelligence were thousands of individuals with really simple rules and reacting only to their local environment and neighbors can self-organize to perform really beautiful complex behaviors and they do that without any leader telling them what to do not even the Queen and my field swarm robotics takes inspiration from swarms in nature and here I've got some running robots and that I would like to show you these is a keno bot we work with these kids Obot on the bridgestone robotics lab and I brought 60 today but you know what we've got a thousand in the lab so they are quite simple they've got two vibrating motors so they can move as you see they've got an LED so they can signal in there in internal state they've got an ambient light sensor so they can sense where where they are in their local environment they've got the computing power of a scientific calculator so not too much and most importantly they can communicate to neighbors within 10 centimeters here I've got a demo that I will show you I'm going to upload a decision making program inspired on how bees do decision making I'm sure they want their really simple you're safe so did you know there's democracy in bees actually in house hunting bees I'm gonna show you an algorithm a program inspired by that decision-making in democracies so I use this controller to send commands to the robots but I'm not telling them what to do or what they're going to do I'm just sending their code inside so they can run it what you will see is that some robots will flash blue and others will flash red in this case blue represents a really good option whereas red isn't as good okay so if I can have the lights being beat down for you to see thank you okay I'm going to running okay you ready let's see what happens I don't know if you can see but they all became blue even though some of them didn't respond to that command so they all decided blue but how well they've got a really simple program with two rules one pick up the option from a neighbor at random completely and random and communicate that option for a length of time proportional to the quality of that option so robots are in blue will communicate blue for longer the rest they will be more likely to pick up blue from their neighbors and so that they will eventually become blue but the most beautiful thing is that this one doesn't know what that one over there is deciding but in the end they all share the same decision so in summary swarm intelligence is about getting an emergent global behavior out of local information and no leaders and that's my field thank you and so we return to your questions so and I think this next one follows on nicely for my robotics demo we have Becky who sounds quite alarmed how do we stop machines and slaving us I mean in fact there were a lot of questions on the threats that AI might pose so what are the greatest risks of AI Neal can I come to you first so I was thinking during the demo um this separation between you actually have a swarm intelligence inside you and you're not very aware of it until you have a runny nose and that's your immune system it's constantly trying to detect threats attacks why doesn't it use your brain well there is a there is probably some connection we don't understand but just imagine that I've found a virus in 1662 see should I destroy it and in oh I'm talking to Jim I can't answer the sense we all have of AI is of this centralized thing that's going to dominate but I think that in that sense what we're actually getting will be more like a distributed intelligence I don't the idiot machines have enslaved us we are enslaved to machines already I was up here coming on the train up here and everyone was enslaved by their machine where they were entertaining them all sorts of things I think that the the point is the most beautiful thing about AI and this goes to sort of Ava's comment earlier that we all have different values and this sense of unified value it's just absurd you know I hope that we learn more and more about ourselves as we try and create solutions to help us and when we get a deeper understanding of who we are will be able to adjust the solutions we use to remove the enslavement that we have today of those machines but you know we're not I just think there's no the idea of a large AI that's going to create itself and do things is I mean I think that we you know the problems we face are the practical problems of the computer makes me do this computer says no I thought that you know and we want to stop that you know that's such a funny thing because it's what happens the computer says no so you can't do it right and we need to get around that and get the computer to be more tuned to us than we are to the computer because computers super stupid they have no idea of context or anything I mean they're shockingly stupid they don't understand us and so they constantly do stupid things now one hope is that maybe that we could get computers to better understand us and what we want the threats not not being enslaved by terminator or Skynet or whatever maybe yeah I think that's what the science fiction narratives are really exciting at me you know we will know them and you know robots are going to take over the earth but I mean as we've just heard it's unlikely really we're not we're not at that stage and also for an AI to be able to engage in the kind of activity that result in enslaving it would have to be smarter than I mean when we think about intelligence we think about it largely in two different ways so there's sentence which is like how sharp and smart is it you know and computers are really good at that and say piensan say penis is more like our ability to experience what we call phenomenological things so things like empathy sadness and we don't have computers that can do that but if you think about any you know relationship that would result in one person being observing it to another it would require both types of intelligence so there's that and the other thing is that rather than enslaving I think we should be more worried about manipulation so a lot of the systems that we currently engage with whether they're sort of social media systems or whether it is you know applications on your phone that unrelated social media they're designed to be sticky they're designed to make us return to make us exhibit types of behaviors we saw from the Facebook emotion like contagion experiment that actually which was where they ran a sort of an experiment whereby some posts were positive and some posts were negative and they wanted to see how far that that traveled where the people you know whether it was contagious and we know that that this is true that they're basically you know we can manipulate each other through these systems into doing things otherwise we might not have done so I think speaking to the earlier point we are being more maybe not enslaved perhaps but but certainly manipulated and managed which is why governance and ethics are so important because there's a lot of money to be made in merging us and our emotions and behaviors but I mean we get the sense I would say enslaved by our smartphones for example we sort of have a sense of you know well I may be addicted to it in the sense that I find it fun but I'm controlling it I'm switching it on you know and it may be feeding me you know it's not feeding me what it wants me to its leading me what I want to hear cuz it's too dumb and and a lot of AI experts get frustrated by this constant yes but you know when machines take over the world in principle presumably principle artificial general intelligence when a machine can ultimately become sentient there's no sort of technological thing that is stopping that from happening ever even if it's not in the 21st century our experts still arguing about whether that would ever be possible well I think that have to be human in you know but it exhibiting the higher level states that like human emotions a lot of the people that work in the AI sector that that I I work with the goal is you know AGI artificial general intelligence so but the way that they were getting there is through tiny incremental projects that have really clear parameters around them so you know the examples are an AI that beat a human it go or no at chess those those are really clear parameters you know if you broaden out those parameters and then it would fail so I think we're so far away from that point this it's sort of it's difficult to imagine and also it comes back to the notion of you know if we're designing these systems now we have a certain level of control as to what those systems will or will not be allowed to do and the fact that we're having these global discussions about what what ethics ought to guide the assistance that we I think is an important one I mean I know in Germany whether you know focusing on driverless cars one of the things is you know should protect preservation of human life is a rule that they think should be hard why didn't he much all of those systems but then when you start thinking about that in more detail what does that actually mean there's a project MIT the moral machine if anybody's been on line with that basically it's like a quiz and you can they give you a series of sort of thought experiments where you're driving this car do you kill this person or this person or this person and the parameters change and then you can see how people across the world voted so we get a sense of how diverse our morality is but in reality it also makes you realize how hard that would be to program into something like who do you kill over whom and even that question is ridiculous really but it's an interesting ride it was discussed in depth in the good place on Netflix okay Jeannette would like to know what do you think are the gravest political challenges the development of artificial intelligence and machine learning will pose to human society I guess this is for und well the immediate answer is probably job loss versus job creation and what that imposes to society but I think that there's also the issue of well there's the ethics and morality issues which we have to worry about I want to just make the point that Stephen Hawking said not long before he died that if we could have machines that are more intelligent than us then that's the end of the human race because if machines can evolve themselves to do tasks which some computer science including deep mind the company that's funding this event that's their stated goal on their website if they can then machines will out evolve us because we're biological and that takes actually longer to evolve their machines and so it's I think it's a philosophical question about how far we can get it but I I do think we have to because we are pretty safe at the moment we are in control mostly at the moment although the companies that use AI with the social networks do play the psychological game to get us to use stuff more we have to be thinking about the future that's what we need our politicians to help us do and that has to be done at a global level and even made the point about the culture that you come from very much determines your ethical or moral views and I spend a lot of time in China now China has billions of people on the internet behind a you know and and they have a very different view to this than we do and the government has access to all the data and they have face recognition on their CCTV and it's we're approaching the time when in January they estimate 50% of the world will be on the Internet well that's awesome in two ways one in 30 years we've got 50% of the world in the internet but there's 50% still to go and most of those are in rural China rural India and rural Africa and you think about well who's gonna be managing the internet in the future and that's where all the data is and that's where all this manipulation is going to happen so our politicians have really got to think on a geopolitical level about this as well as well as on the social level about managing through the peaks and troughs of the job losses and job creation and you know and so there's all the issues that we brought up in the review we did about how we get existing industry to adopt AI how we help the startups how we reduce the skills gap so we get more and more people able to work in this industry but I really think they they have to think on the global and scale as well which is really difficult at the moment I mean the debates are happening you know people like you know you hear the government was a big house of laws reports as well you know III and it's very good that so you sort of get the feeling that the policy makers and politicians are talking about it and are starting to be aware how quickly the advancing yes I think I have a theory about that which is that as well as you know the the AI this AI revolution is driven by computer processing power and and the data that's available which is largely been driven by the availability of the internet but also all the politicians go every year to Davos and in January 2017 shuara who clash barber who runs Davos released a book called the fourth the Duster revolution and that was all about you know how are you gonna deal with AI and they all came away from Davos thinking we've got to do something about this so there are every country and yes even China yeah all the developed world is having an AI strategies how we're gonna deal with it how it and there's there is effectively an arms you know there's an AI race and the super powers are the US and China and a country like the UK has to work out where it fits in in we cannot be a superpower in AI but we can be a balance of power and III and we can do good things and we can look after our economy and and Europe you picked up GDP our Europe has gone very much for data protection and that's sends good signals out so it's you know when when Chinese companies truck like Alibaba trade here then if we buy and sell anything on Alibaba then the Chinese government have access to all that information it's we can't afford to so ease off or step back from from debate technology we are one of the countries that lead the world we don't do it we we leave the world in terms of what we don't lead I would say the US and rapidly China are coming up in terms of the number of AI researchers but we have some amazing AI research in this country because it goes back the 50 or do you I mean you you can argue we invented it Alan Turing and the I whole concept of can machines think and so and we have a fantastic legacy and wonderful universities that teach and do research in AI you've seen examples here from Bristol and you know we will represent universities a very good and so we can we can still we are very inventive in this space the card thing is that the even deep mind sorry deep mind you know the business model of the AI startups at the moment is to be bought by Google or Facebook right that's that's your business model I mean demos has a massive deep mind says they had to sell to Google to get the data to develop those algorithms as well as the money and that's a big problem for Europe big problem and that comes back to how do we how do we tax these companies I mean there are huge issues here and it moves faster than any chance of the Exchequer can though these companies have bigger turnover than the GDP in most countries big issues okay next up we have Josh who asks do you worry that AI will reflect the implicit biases of the designers how can we address this yes we we should worry absolutely and people are already thinking about this so there's there's a number of trends at the moment within I guess industrial academic thinking one of them is the fairness awareness and transparency movement which is it bridged as fat and also the responsible research and innovation movement which is thinking about sort of ethics and responsibility right from the get-go but in terms of buyers specifically there are sort of two ways that we might think about that one is that algorithms are designed by humans who are inherently biased and therefore potentially some buyers could bleed into the way that we designed those systems but the more pressing concern in the one that I guess we can illustrate to ourselves if we have a Google anything is that there is unconscious or implicit bias in the data that these models that sort of run the systems are trained on so for example if you google doctor and I'm not picking on Google particularly it's just the most obvious search engine Bing it yeah so like nobody does but you know if you google doctor predominantly the images that you'll see you'll be men and most of the faces will be white so there's some unconscious and implicit bias that is in that data set that's that's generating that output and also this is particular concern for things like facial recognition particularly if we're going to be using facial recognition in things like driverless cars you know if you have to recognize passengers out a lot passengers sorry pedestrians out on the street because predominantly most of the models are are sort of trained on white faces so you have a massive potential issue there because obviously people are just enfranchised through that model so one way that we can deal with that potentially is by extending the the data upon which the systems are trained so diversifying that and there are projects that are looking to do that and the others to train people that sort of design as these systems so AI specialists interaction designers interface designers about a conscious bias and what it means and how to detect it I mean you know fundamentally the systems that we're designing are just reflecting reflections of us as human beings you know we're all bad and therefore the systems are bad that's the only way of thinking about it really but the problem with artificial intelligence is that it has the pop but you know it encodes it it has potential to encode these biases and also the more that these biases are embedded within systems that make judgments about us that might result in harm the more problematic it is so if you have a bias model running in a criminal justice system potentially or to decide whether you get insurance or decide whether you get education then that bias becomes problematic so in answer to your question yes it's a big problem yes we should be doing something yes we are but no we haven't solved it Wendy I mean usually we hear that benefits of of AI is that they can circumvent or somehow be more objective than humans who are so riddled with societal biases and and and and some subjectivity in some instances but if you're training on data that's been produced by humans then they'll pick up the bias that is there and then you have to decide do you program that bias out and who has the right to program that bias out and in what way and and so yes sometimes they can be more objective and I often say when you talk about automated cars nobody asked me about my moral attitudes when I took my driving test it's just to make up your own mind who you know and we all know about the instinct we have to you know avoid a dog and killer pedestrian because we've seen that animal and you know the highway code says kill the animal but you know the instinct and so cars may well make the right decision in those circumstances as scary as it seems but I think I want to just move the conversation a little bit to the issue of diversity because it's one we tackled and talked about a lot when we were doing the review is that we have you know we have to address the skills gap we need more people working in this industry we need to open it up to as many people as possible you know if you're thinking of something to do you can write your own paycheck in this world so you know it really is a fantastic future for P but you don't all have to be not everybody works in AI has to be an extreme machine learning programmer and it's very important that we have diverse teams that come from different disciplinary backgrounds we need the ethicist we need humanities we need the you cannot economists we need the lawyers we need the social scientists we need this philosophers and psychologists involved in this world and we need to create I think the an essence of interdisciplinary teams working on in our in companies that are developing AI algorithms products services to sort out these problems so it's we don't have the answers but I my pleas for to think about diversity everywhere so next question this question comes from Gideon who wants to know with algorithms becoming commonplace in various industries should we be putting greater emphasis on teaching mathematics probability stats calculus in school to better equip future generations for challenges that lie ahead well I think generally yes anyway you know because we're moving into an environment within which these are going to be cool skills but I think sort of a long side that is maybe this the software idea of teaching computational thinking and teaching people enough about a system to be able to understand what it is they're interacting with so and I'm thinking as well not just children because actually when you think I often think you know yes we should be teaching children to do these things I think that now but those children become adults in 20 years and how useful was that information so to some extent I kind of think that we need to be focusing on adults as well so the post-16 education sector people who are interacting with these systems on a daily basis are just ignored and marginalized in this context and actually you know we very many people don't understand how the models operate that run the systems and why should they but actually we put a little bit of focus on ensuring that all of society has equal access to the information and understanding so they can actually interact with AI in a way that benefits them and doesn't harm them I think that's super important and the other sort of part to that question is that we should be teaching computer skills but actually I've been doing some research with the Institute of Chartered Accountants for a little while spitting up when your accountants exam plus like yes that's what they said when you were talking but what we're discovering there is that yes you know the sort of the skills that could be automated are being automated this is sort of a triangle where you've got three partners and senior managers at the top the people at the bottom in terms of the job roles are sort of gradually being eradicated and actually what's left in the middle is you know the you've got to work out what skills these people need just to be able to enter the middle of an employment triangle and one of the things is soft skills so the fact that in accounting particularly you still need to interact with clients you still need to convince people that your system is a trustworthy so actually there's as well as those kinds of skills we should be thinking more about the sort of the human skills that allow these technologies to embedded in social life and professional life I first of all the government recently ended when Michael Gove when he was Education Minister changed the curriculum we dropped IT and everyone's doing computer science and we all said fantastic idea but I knew what would happen and what's happened is we've got less people now reading computer science a that was than we had before and partly it's a problem unless girls particularly because we would change the curriculum but we haven't trained the teachers we don't have and last in the last budget last year when our AI review was put into the budget we've got the money for the AI review and the Center for data ethics and innovation that the government also set up they put in money a million pounds I think it was to train computer science teachers so that we can get the right education in the school so that we can encourage people to more people to learn these skills but I also I think we it's about time we stopped accepting it when people say that oh I can't do mathematics and accepting that that's okay and I think AI can help us here I would give the example where I am was very I was naturally good at mathematics at school and by the time I was well in primary school I was being asked to teach the other kids math because it just came so naturally to me and I I couldn't understand why other people couldn't do it and then much later in life I started to try and learn to ski and I was hopeless absolutely hopeless and I used to fall over on the nursery I couldn't get up and the instructors eventually get bored with picking you up and everyone else moved on to the higher slopes and I went to the bar what I needed was a personal coach I needed someone to help me get the confidence and and that's what we and I think we can use AI to develop personal tutors and coaches for kids I think we need we need we do need more people with maths and stats skills and but we need the all-round skills as well it's not that everyone's got to become a mathematician and a computer programmer and do you think Evers point about teaching adults constantly learning are we going to move into a world where you don't qualify whether you do apprenticeship or go to university or whatever and you become you know somewhat expert in a particular job or career and you see that so it's the end of your life you will have to be constantly because if AI is constantly catching up with us we're gonna have to well I think you're very trained and ready seeing that today's the Millennials whatever today's kids will not have the same expectations of jobs that for life and and so we're ready into that we're academics only obviously their job for life but the I think that I think we can use AI to help with education I firmly believe that we can to help people get develop skills throughout their life Neil Andy has asked how can we ensure that the benefits of AI are shared outside of London it's a really interesting question I mean I've spent 15 years living in Sheffield but five years commuting to Manchester working here we have a lot of problems with the London centric myths of the country today it certainly wasn't historically the case there's this weird thing that if you look at where all the innovation came from it came from the north and when everyone got rich enough they all moved south there's this sort of common pattern I come in for the name of the family and gossip that's founded the mills there but the hill woods you know and they ended up running a stall how about we I think we started I there's no simple answer this question and actually Wendy's sort of saying I don't fully agree that China in the u.s. so world leading ahead of us I mean and I think that that victim culture is really something we have to watch like if Manchester's gonna look and say oh those things being done to us all the time and and and and it's out of our control well that won't work we have to stand up and say no we can affect things we can change things and Manchester's in a great position on that in terms of the devolution deal which my understanding of which is devolved healthcare as alongside social care there's devolved transport here it is very hard to do changes good changes to on a national scale you need regions that can try these things Edinburgh has a lot of things going on in health because Scotland which is only the same population as Yorkshire can trial a lot of these things the Manchester City region can try all these things that can't just be someone up in London the Royal Society should come in two events here all of these things help in fact um you know I do really worry that you know we should not be reactive we should be proactive and I'm and that's maybe easy to say but it's not going to happen if we're just saying oh people are doing things to us and Manchester's leading University one of my great friends and colleagues here Magnus rattori runs the data Science Institute here he's a top academic individual and it's an amazing Institute a data science is just AI without the hype you know so really the seeds are there one problem is I don't think we you know there's been this tradition that we have to cut costs everywhere and that's closed down the room for innovation so how do we recover that innovative culture that says that the for everyone whatever their job whether it's doctor social care work or whatever that says there is room for me to innovate in my job and make things better because I think that's one major problem we have when you speak to people who are doing these important jobs on the ground where we could get a lot of benefit by introducing these technologies they just don't have the time they're so busy with what they're doing and that as well as the training it's great they can go and get training but in the innovative ones have it knocked out of them and that really worries me I don't know how we change that I mean I guess big cities like Manchester are a well placed to be able to make sure they're not left behind but let's say outside of London and Manchester and the big regional you know the big cities there must be whole areas of society who are gonna feel disenfranchised or who gonna feel more threatened by automation absolutely just look at the places that feel most upset with the current status quo so my son used to play football in Sheffield and we were going out east all the time so old mining villages where the football clubs are still associated with the miners welfare clubs right and but the mine doesn't exist you know they took their kids football a lot more seriously than I took my kids football because that's a major thing going on in the community and what we've seen is that there wasn't a massive conversation about how we shut the mines down and everything else ironically like we're now having now I mean we talk about accountants but there weren't thousands of accounts on the street when stock broking disappeared as a job that the worm thousands of those are history but we did just put these people out of work and massively disenfranchised them and that is not clear to me at all how we put that right because it's not like you know there's nice things that I was at a meeting where Jeffrey Sachs said it's gonna be great we can just discuss philosophy in coffee shops and I said well that is great but around those communities there were a lot of things you know there was a social structure associated with hard work which was a very important part of the community and that kind of went when the work changed it's it's to me that this goes beyond AI and and I don't think we're talking about it enough I think you know you sat in policy meetings where you realize we're talking about all what's going to happen when when sort of you know in the future doctors are chair their work change or lawyers or accountants well in the past they were fine because they're all got an old boy network and they moved on to something else and they were educated but the people who are at the most marginalized section of society like driving taxis now I met a guy once who went into catering became a prison chef in Lancaster they closed Lancaster prison his only option was to drive taxis he was very happy driving taxi so what if if that goes you know this is this is very very challenging and I don't think AI is bringing the answers it may bring more problems I think the one of the things for me is how much local government can use AI to make services more efficient and I think we talked about we talked very glibly about smart cities but I think it's actually two best easier and more efficient initially to create a smart village because you can you can actually tackle that issue and I and when you think of what the local governments being asked to do today in terms of the cuts and the huge I had an email yesterday from Southampton City Council I don't live I live in the new forest but I saw the email how they've got to make these huge cuts and they've got to decide what to cut and we're asking the people what would you do now and it's that well really the benefits for AI in those situations I think that's where we should be putting the effort and that's how you get it outside London because actually the big metropolis is like London Manchester Edinburgh are much harder to make smart because they're just that much bigger and they're much less homogeneous so I think there's a huge thing about local community the other thing I would why I think the US and China have the apart from the scale at the number of people working the area I have huge amounts of data that's where China gets its powerful but there are things like the UK so it's all very well saying oh there's these big companies what is the budget of the NHS 400 billion that is larger than the turnover of the largest company in the world right 400 B and rather than saying oh look at that they've got let no we have a national health system with we say that we need to do more about it and think one problem with it is as we operate that we are constantly so obsessed with efficiency around it that we don't look for the opportunities to innovate and use and save save money there but it comes back the problem within the national health is a jewel in our crown in terms of an asset we can use to mine the data to give them better personal care medicine but you've got the huge ethical issues of how you who deals with the data and how you deal with it and I think you'll do it at a local level that's the point I was trying to make there and it's about scale I want to stick with you Wendy for this next question zigmund asks do we need to be educating politicians or on just what is possible with enough data to ensure that the privacy debate is well-founded that picks up very well I don't know if you planned this I think the interesting thing for politicians is they're already well aware of the issues because most politicians the most important thing to them is the next election and one of the things that's happening with a eye is threatening democracy in terms of you know the whole all the issues around the social networks and what the companies like Cambridge analytic air who have now disappeared completely but there are other people with those skills can do to try and manipulate the way that we vote what the Russians are doing and other countries Iran and other countries like that are doing to try and get you know not destroy democracy but certainly to tamper with it and I think this is a major threat for us and I will bring China back in again because it's not there's not that's not a democratic it never has been country they have a completely different view on all this and I think how politicians are quite aware of that because they are worried about the next election that because that's what they're all so I think they are I and in terms of educating them though there have been these great reports that have come out and I think that the problem we have is that motor I mentioned Matt Hancock who's just gone to health and he actually is someone who understands programming he can he can program right it's very few politicians who are to science or engineering literate and I think that's a worry and I think that's a worry not that we were going to make all our politicians scientists or engineers but that scientists and engineers are not going into politics I think that's the biggest problem we've got Abby asks if algorithms are dependent on using our data yeah well I mean so some people would say that we are being paid for it via the services I wouldn't necessarily agree with that but you know so for example you're using Facebook for free and how is it free it's free because it has a monetization model that relies on the social graph cells that that data on to advertisers who then push products that you okay so that's one model potentially so arguably we're being paid through service but you know the other idea that somehow you know we can we can harness or harvest all of our data and then sell it off through some kind of sort of you know horizontal market is is something that people I guess especially in the academic sphere people have been talking about for a little while so there's a few projects that are sort of dealing with this one of them is a hub of all things and that was initially intended to be I think it's more of a privacy focused thing now but it wasn't initially intended to be a platform that kept your data safe and then you could decide which of that data to sell on to which companies in return for some kind of service or money or whatever now in theory that's a really good idea because you think well it makes total sense right my data I get money for it you know great but there's a couple of problems with that one is that you know the implications for exposing your data about any activities in your life is that there's a concurrent reduction in privacy now you might be cool with that but actually who are the people that are going to be selling their data it's not somebody who earns one hundred fifty thousand pounds a year it's somebody potentially who's either on benefits or on low income and I think 50 pounds actually is worth it for that bit of information but we don't know what the long-term effects are of exposing our data in that way so in reality while it seems very simple as a sort of a proposition it's massively problematic in terms of what that would mean for people being sort of marginalized and you know ultimately potentially exploited so I mean I think there is a question that needs to be addressed you know are we being exploited is our data being used in ways that it shouldn't be but I'm not sure that's you know getting money for it solves that problem it almost causes more problems because it's like it finishes the conversation it's like you give me your date I paid for it right so you're not really engaging in a meaningful debate about it then so so you know I think all the face bits it's a good idea but actually in reality more complicated we've come to the last question and this is a question from fuzzy great names by the way thank you all for the ask at what point will we be able to say that machine learning has become machine understanding that's a great questions here with me so I'm we talked a little bit about empathy earlier and machines having empathy and what's interesting about this is if we were machines if we were in the future when this debate is being had by machines the debate wouldn't actually happen because we would just communicate over Wi-Fi to each other everything we thought agree on an algorithm in your land because we are the most the reason why AGI won't happen and I totally disagree with what Wendy said is because we are a function of our limitations not the technology we're a function of our limited ability to communicate so I can only communicate on 100 bits per second I don't know what these little guys are doing here but the capability of a computer is to do billions of bits per second as a result what we spend most of our time doing is thinking about other people worrying about other people what did that look mean oh look at that grumpy person there or this person is in a bad mood this morning but our blahblah because we can't actually directly communicate what's going on and that's why I mean empathy and and that's why we're so cool I mean we've got massive computational ability but limited ability to express ourselves and that's why all we have all these drives and senses but the computers we build will never have that because they can just communicate in 10 immediately across themselves but the problem is at the interface right so you can build a system of efficient computers that could run our economy and everything else way better than we could do it and if you leave humans out of the loop that we find but what's the point what are they actually doing it for at the end of the day this is about humans I despise the idea of transhumanism because the real idea is about us you know and what we want out of things there isn't anything else and that's where the challenge is that we have this limited bandwidth of communication we all hopefully understand each other most of the problems are when we don't across cultures and whatever else and unless we get the machines to sort of have some sense of that and what we really mean we will never get machine understanding there's a possibility we can emulate it you know after with a lot of data well I think there's a strong possibility I think one day we'll be able to emulate it and I think that as we get closer to it that's when we'll feel more comfortable with our computers do the things that when you talked about of the personal coach you know and reflect the various different ethics in our computational systems but you know it's a it's a long way off but I think that's it's a great question and that should be the route now so I'm going to go a different route which is about you know flipping it around a little bit to say that it's also about humans understanding machines so you know we might reach machine understanding where a machine understands everything that's going on with a human being but we also have to have some kind of I guess like symbiotic relationship almost where the humans understand machines as well and that involves a greater focus on the things that we're already starting to look at so transparency how does how is that Racine reached its judgment why is it done that why that and not something else all of those things that we just can't answer and actually with the more complicated algorithms so neural nets and such like we we even a trained expert is going to struggle to tell you how it reached this judgment so for us to be able to understand system there's a lot of work that's got to go into that and also the interface design as well so these things at the moment a lot of the IOT you know the Internet of Things technology is designed so that it's invisible in use it sweats it frees us up sorry from sweating the small stuff and we can go off and do crazy things with our lives but actually while we carry on doing that we become more and more dislocated from what is actually happening and how we're being if you like enslaved or controlled or manipulated so I think it's bi-directional you know the machine must understand but also we must understand the machine Wendy hmm there's so many things here I think the next wave of innovation is going to come around in tension and I think you said that they know much and they say we have in the labs already the brain machine interfaces I think this is where the next wave of innovation is going to come from the fact that machines will be able to read our thoughts or at least read our brain waves and interpret that I was inspired a bit by what you said on this and it comes back again to your point about is the augmentation it humans and machines that's going to be so interesting and this again I'm getting bitten quite interested in the whole Smart City concept here because you know we're very glibly talking about smart cities but if we wanted to have you know an AI controlling the temperature in this room how would it decide what to do with all of us sitting here right it's hard enough with one person to decide what temperature it's going to be when you walk in with what's our intention is half you know we'd all want different different environments different temperatures and I think so that I think this is where where it's good I think the other thing I go back in time and think about Alan Turing and his Turing test to decide whether machines could think and he was you know for a long time that was the Holy Grail we could we get machines that solve the Turing test and over time so basically if you don't know the Turing test I'm sure most people do it's like you have a machine and a person behind the screen and you ask them the same questions and from their answers decide which one's the machine and which one's the human if you can't tell the difference then that you've got an intelligent machine well you might have a dumb human and this came across when Google recently did it was Google they had something electoral Siri one of them ring a restaurant to book a table and some hairdressers appointment and I thought was very sorry for the human being on the other end who was being asked to do something they didn't really understand by something actually was a bit not sharper than them in many ways and so I think this and and the other thing I want to just add to this is that we aren't always as intelligent or as able to communicate as you know the people sitting around this room we start off completely dumb as babies and we learn and there's a lot of research in is it a life and things like that we you know you think about how do you grow a machine and there's been a little science fiction about this that knows nothing and then grows and I think I like this idea and it came across in the Philip Pullman books about your daemon that when you're born you have a demon grows up with you and and it's your coach and you know your interface to life and so I think there's there's lots of new research areas that will play into this as we go forward very exciting very exciting indeed and unfortunately that's all we have time for fantastic debate on behalf of the Royal Society a huge thanks to our host Manchester Science Festival and the Royal Exchange theatre thank you to deepmind for their kind support of the Unai series and i guess perhaps well obviously thanks to mike my three panelists but thanks to all of you our guests for for coming here tonight and especially to those who are such brilliant questions so lots and lots of food for thought the debate carries on thank you all for joining us this evening and we hope to see you sometime soon thank you all very much you

Channel: The Royal Society

Views: 48,712

Rating: 4.5586205 out of 5

Keywords: royal society, science, scientists, scientific policy, scientific research, science uk, science research, international, international science, science education, science policy, AI, Jim Al-Khalili, Wendy Hall, Neil Lawrence

Id: E3qIExS4r-I

Length: 74min 32sec (4472 seconds)

Published: Thu Dec 20 2018