Global governing body feed chose a marina of brazil and one and hell not putative paraguayan were charged with racketeering conspiracy liberations about the guilt of a third will continue next week theres i had lines more news here on aljazeera after talks Al Jazeera Stephen roberts say with us if you can. Session staged a serious debate up front at this time on aljazeera. You can watch all. You see. We live in an age of rapid technological advances where also efficient intelligence is a reality not a Science Fiction fantasy every day we rely on algorithms to communicate to all banking book a holiday even introduce us to potential partners Driverless Cars and robots may be the headline makers but they always being used for everything from diagnosing illnesses to helping police predict crime hotspots algorithms also make important decisions affecting whole lives all be eligible for a credit card which applicant should get the job. Critics say the more our personal data is gathered the grace of the threats to our privacy as machines become more advanced how does society keep pace when deciding yet excess and regulations governing technology we find out is Steven RobertsMachine Learning professor at the university of oxford talks to aljazeera. Professor Stephen Roberts thank you for talking to aljazeera about a fascinating but often confusing subject so perhaps you could start by explaining exactly what these algorithms Artificial Intelligence on the scene learning actually are and how theyre involved already in our everyday lives i think most people have heard of algorithms and many will know something about them these are much more traditional ways of solving a problem with a computer an algorithm is basically a sequence of mathematical steps which are performed inside a piece of software in principle they could be dark with a pencil and paper by you are but of course with big problems its much easier to let a computer do them we critically we design the steps in a traditional solution in an algorithm so we tell the algorithm what the first step in the training is and then it moves on and performs the next operation of the next and so on and at the end the answer is attained and these are found in everything from our mobile phones to engine management units in our cars to intelligent toasters and microwave ovens and so on but we wouldnt really regard those years as a i in a more modern sense the very big difference that i am Machine Learning bring is we dont tell the algorithm the sequence of steps that it should take ahead of time in order to solve a problem we merely give it a lot of data to try things out on and we tell the algorithm very clearly what success or failure looks like and what it would do is it will try various strategies to succeed. Those strategies that it finds lead more likely than not to success it will reinforce inside its memory and those strategies which lead to failure it will begin to inhibit and after enough iterations all runs through this procedure the algorithm itself the ai system has learned the likely ways in which a construct a solution which gives rise to a high chance of a successful outcome and that really is a learning process the only thing we need to do is tell it the rules of the game what success looks like and provide it with an environment in which it can. Iterate and work out these strategies typically by providing it with a lot of data ok lets deal with the dystopian fear that some people have thats Artificial Intelligence is going to lead to these robots these machines that can take over the world they can replace human beings is that conceivably remotely possible. It sounds much more like Science Fiction and we certainly nowhere near that particular point where there are going to be swarms of sort of armies of robots that are taking over the world i think we have to remember that automation and autonomy are something that is very deeply embedded within our world already whether its from algorithms that are trading on Global Financial markets to smart algorithms that are scanning a miles checking for viruses and working out what a response might be that we would make predictive text on our mobile phones is another such intelligent algorithm so i think these kind of things with very familiar with and were not really afraid of the i would say the fear on a much more rational level is that like all automation it would displace us as people from the jobs that we have so you can imagine a what we really are in the midst of a second Industrial Revolution a technological revolution. And whereby traditional jobs whereby people use their intelligence in order to sort things work through things whether its accounting working in factories making things robots or ready taking those jobs because they dont tire theyre cheaper long term to work with and so on the advances though theyve been quite rapid i mean i would say what in the last ten years theres been a tremendous push towards Artificial Intelligence towards Machine Learning what have been the bigger the advances you think that have been happening whats changed dramatically in the last ten years the last five years especially is two things firstly we just have computers that are capable of so much more we have machines that we can go out and by ourselves which have supercomputer like performance from twenty years ago and thats Pretty Amazing in itself and also the amount of data that is available out there publicly or commercially available on the World Wide Web is absolutely phenomenal the scale of data is almost beyond belief and you couple those two things together you have the methods that can make smart algorithms that can look for the world around them to solve a task you have data to feed their investigations as to how to solve that task and you have computers to hope you do that in a reasonable time so you dont have to wait a year for an answer you can now get the answer within an hour or even minutes in some cases so who do you think is actually behind this push to make such big technological advances obviously slices of people like you with want to innovate but is it also down to business is it down to investors money politicians is almost like a space race it feels to me very much so i mean youre absolutely right previously there have been big methodological advances in Machine Learning and i. I mean that tends to been done within the academic sector people working on the theory developing efficient algorithms to implement that theory and showcasing how they might work in principle at industrial and commercial scale with the rise of data and the rise of computation it means that industry can take these ideas build upon them and use them to commercial advantage and of course that becomes then there is enormous Business Case money gets reinvested back in to make these methods even better make them work on bigger bigger datasets extract more knowledge make the business even more profitable and i think you can see the enormous rise of a i in some of the event Big Tech Companies to date who are using Machine Learning as really one of the central pillars of the business all we as a society they equipped to be able to keep up with the pace of these developments i think as consumers we enjoy the benefits of those advances but as a society we are definitely lacking behind simple situations where you can think about crowded streets in a city and sat in traffic jams everybody driving there the kind of dirty car wed love the idea of self driving cars and electric vehicles that smartly charge themselves up autonomously and so on but in order to do that we as a society need to deal with ethical legal frameworks i was talking to a lawyer quite recently who was worrying about whether you could actually take legal action against an algorithm. It becomes almost bizarre philosophical commentry but as a society we need to address these kind of questions head on if a piece of machinery goes wrong is there a chain of litigation through which we can. Bring the legal system to bear should we worry about that even if the systems are performing better than human beings what happens if a robot surgeon gets something wrong who is to blame the hospital the designers of the robot the people who created the algorithm remember the algorithm has developed itself learning from data and that data might be a very public open repository with thousands or millions of stakeholders so its not clear where the blame or the audit trail of accountability lies i think as a society this takes us into a very New Territory were used to people being the people who are responsible and when algorithms are now responsible its very difficult to to talk and negotiate with an algorithm for a particular task i mean just on an everyday basis for example when you go to get a mortgage or a new bank account we all know that if the computer says no because you cant take the right boxes youre not going to get a label that that credit card is that what youre talking about the ability to be able to absolute yes a lot of machines theres a very very much so and in fact there is an enormous push now amongst the Academic Community as well as the very big commercial players to try and have some kind of accountability an audit trail of reasoning built into the way these out these machines are operating this is absolutely necessary because we have an algorithm that drives a car or makes a decision about whether you get a credit card or a mortgage and you are denied or something goes wrong you need to know on human understandable terms what that reason was big problem is if you force the computer to only use a trial of reasoning which is understandable to us as human beings guess what the performance of that algorithm drops and often drops below Human Performance so if you want to algorithms that do better or as good. As human beings we need to have algorithms which are not understandable necessarily by us is that why so many people are worried about what they might term as biased algorithms the way that an algorithm is used by people what is in proceeds by a person might be biased what is it put it by another person might be biased and im trying to come to something that is completely neutral i mean thats very difficult to do isnt it i very much agree and i think much of the bias goes back to the data which the algorithms often the fed if we take a there are often classic examples if we take say the word scientist then we go on to the web and we search for pictures of scientists because of cultural bias depending on what culture we come from most of the pictures we see would have be of date white men. The world is much bigger than white men and and yet an algorithm doesnt necessarily have the sensitivity to understand that it is looking at a very biased collection of pictures and so will believe what its given it doesnt have a kind of high ethical. Process and so that bias will tend to be amplified in the results it performs so when you next asked the small town of them find me a picture of a scientist it would tend to favor in favor of the statistics of the data its been given if those statistics are skewed or biased it will amplify that and we obviously going to have work very hard in order to produce unbiased algorithms which take these sensitivities into account fortunately many of the big commercial players and a lot of Academic Work is going into trying to do this so teams that places like facebook and google and so on are working very hard to try and remove these sort of almost cultural biases from the data the algorithms are using in the first place so how does all that feed into the criticism that some people have of the way out. Rhythms an Artificial Intelligence all use for example within the police the police allocate certain resources to certain areas because theyre told by a computer that area is deprived there is more likelihood of crime taking place there the criticism is that that reinforces stereotypes in that area. Absolutely in principle of course what people are trying to do is is use very effectively a finite resource if we turned that problem slightly differently and said we had a finite amount of money and were allowed to spend it or buildings to make those buildings safe for earthquakes wed preferentially spend that money in regions where statistically earthquakes are much more likely to occur and i dont think anybody would question that that line of reasoning because the moment you start bringing that to putting more police in areas with higher crime rates you actually beginning to have a much more complex story and thats because it involves human beings where algorithms are good at is dealing with data but that data doesnt have. An emotional context behind it or sensitivities and the depths of understanding that we need to go to in order for algorithms to understand the emotional impact and the social impact on us as human beings its just not there yet so i think ai Machine Learning is extraordinarily good at analyzing data working out very smart things about how you would best use a finite budget in some our own work with for example try to work out how given a certain budget you can have maximum impact to prevent malaria again these kind of more Health Implications i dont think anybody would be too concerned but the moment you begin to start looking at things to do with policing and crime and so on there is always a danger what you are again doing is potentially reinforcing social stereotypes the problem isnt necessarily with the Machine Learning its being done rather clumsily and perhaps the worst you can say is naively the problems are actually deeper and they are societal problems which im afraid even i icon solve. On an individual basis of course many people concerned about privacy and the security of their data always getting the balance rice all mass do you think people are quite right to be worried about bruce in security there is a lot of data out there in every interaction that we have in a world where there are now more internet connected things than the arse human beings most of the time we leave a trail of data throughout every day of our lives and that data can and often is collected often it will help prevent us against Identity Theft and fraud but sometimes it can be used against us either maliciously or surreptitiously for marketing or advertising purposes and so on. I think in some areas we need to have ideas a little bit like opt in or informed consent so in many countries there will be Organ Donation so upon your death you give few organs to medical science trying to help others to live i think there are many cause for our data a medical data to be under a similar kind of framework so we can offer the data we generate as human beings with all the the bumps and bashes and illnesses that we have possibly even sections of our genetic code to try and help medical science to improve intervention tests and medicines which are of Global Benefits to all is ten years since the beginning of the first rule smartphone when we think of these the the i phone when steve jobs went on stage in two thousand and seven introduced the world to the i phone of course the smart phone i dont think any of us would want to be without but it is a device that collects data most of the Big International players are actually extremely good when it. As to the use of that data they but they will use it for their own commercial gains between advertising recommending friends that we might get in touch with or some or some such a game we wouldnt think of that is terribly impactful but we lose track i work in this field and i i couldnt tell you where who and how my data is being used and thats thats slightly frightening ive got to say the words to of course fake news to what extent is algorithm Machine Learning how are they involved in the spread of fake news which is you know come to dominate the headlines over the past couple of years and theres been a lot of people saying well this is all down to computers managing to spread the misinformation who is to blame for that. Ultimately people want to spread fake news for particular purposes whether thats too for advertising or whether its to try and destabilize nations. Is almost the same computers and i are used in two areas the first area comes back to the question you posed earlier which is if we only had a certain amount of money to spend where and how would we spend it most wisely to get biggest effect we can use algorithms to say if im going to spread fake news what outfits be they media or social media or wherever should i put that fake news so that i believe it will spread like wildfire and we can use algorithms to tell us how were going to do that you can imagine i could create fake news but if i if i put it up even publicly in the wrong place hardly anybody ever gets to see it its a bit like a contagious disease that need to Critical Mass of interaction and we can use algorithms i say we course not me but people can use algorithms to seed fake news in the right places at the right time for maximum effect the second thing is the fact that algorithms are used to spread information around the world and they do that without correcting it they dont know the difference that precedence between true news and fake news they simply propagated on and that means that if if somebody wants to spread fake news part of the question they ask themselves is going to be where can i explore it the loop holes in the existing algorithms are the social Media Companies then. Doing enough to combat that because theyve had theyve come under a lot of criticism i see theyve got people working on that rather rapidly right now they certainly do the very big social