 Our next presentation comes from a man who is from Great Britain but now lives in Greece. Now I don't know if he just tried to escape the bad weather in Britain or the Brexit, but we're sure happy he made it all the way here to Hamburg today. Now the latest elections in the US and the referendum on the Brexit have shown us that with all the collected data and technology that we use to predict the outcome of such elections, we can do it. We cannot really know everything or predict everything. A new dark age is an exploration of what we cannot no longer know about this world and what we can do about this. Our speaker James here is an artist, a publisher and a writer and his writings have appeared in magazines and newspapers all around the world, such as Wired, The Observer and The Guardian. He's also a very experienced lecturer in conferences, in universities and at different events and we're very happy he's here today so please welcome him and give him a warm round of applause. Thank you very much for that introduction, thank you very much for having me, thank you all very much for coming here, this is huge and amazing and I'm very grateful. I was, yeah I just put hello there rather than new dark age because I didn't want it to sound quite so frightening at the outset and I'm not, it's a thing I'm working on at the moment and I was also asked today to talk a bit broadly about some of the work that I do so a lot of this is going to be like a bit me, me, me in that I'm going to show some things I've made as a way of explaining the way that I think through technology and society and various stuff around that and hopefully there'll be connections through that and I'll talk quite a lot about one particular project I did this year and then some kind of extrapolations from that to kind of where I think I and perhaps we need to be thinking next. Actually my background is as a computer scientist but I'm a seriously lapsed one, I managed to graduate with a master's degree without actually learning to program which tells you everything you need to know about CS education in the UK and I've since been trying to teach myself desperately ever since and really badly but I've also been a literary publisher and a journalist and a number of other things but it was my kind of background in publishing and my interest in the web that sort of got me slowly moving towards the place I'm in now which is increasingly in the art world but I'm always always crossing over and this is a good example of the kind of where I put things together which is a few years old now but this one of his work is a publisher but wanted to start to talk to people about how technology and the internet in particular changed the way in which we're putting knowledge together and so what I did was I took the complete history of a single Wikipedia article I took the entire changelog of it which was the article in the Iraq War which at the time I made this book was it was about seven years of data on a single Wikipedia article and because I was working a publisher at the time I learned how to make books properly so I type set it all it's all in like neat little three columns quite small type but when I had that all print and bound that single Wikipedia article changelog took a 12 volume size so that the scale of an a proper full old-style encyclopedia and what I was thinking through this work is the fact that the way in which we build stuff with technology doesn't really necessarily change the world so often but the way we understand how that technology functions changes the way in which we can interact with the world and so this is still an old-school encyclopedia Wikipedia is still an old-style encyclopedia but the technology we've built that underlies it allows us to put that together in a way that allows us to see all the changes we made to it we can see the fact that encyclopedias are not final statements of fact but processes of arguments over time with many many many voices contributing to them it's a machine for historiography rather than for history and that for me captures something quite fundamentally different about the way we're capable of building stuff today and so my projects go back and forth between like things that exist as big solid objects which are kind of quite easy to understand and more kind of immaterial software things this was a project that I ran for three and a half years on various social media tracking reports of drone strikes so I was working with journalists from place like the Bureau of Investive Journalism who were tracking drone strikes in undeclared theatres of war so these were the covert ones in Pakistan and the Yemen in Somalia these reports were being posted and those who are kind of interested in them could find out slightly more about them but it struck me as deeply strange that this entire war was going on without images which is deeply strange in our current times I mean and for a long time for more than a century we've they used to send illustrators to battlefields right with pencils and papers and an age of mass media where we're saturated with images there was this whole war going on without them and yet at the same time we spent the last 10 20 years photographing the entire earth from space you can pull out your phone and look into any spot on the earth surface and I wanted to pull those kind of infrastructures together to close those loops so every time there was one of these covert drone strikes reported I went to a digital mapping service use various ones to find as close as I could the actual landscape in which that occurred to kind of resupply some of these images back to this ongoing fight and I post those to Instagram to Twitter to Tumblr as these long sequences and trying to create a kind of visual record of what was occurring one of the good things happened is that in these images started to then be used in other places so media when discussing drone strikes instead of just running like a photo of a predator drone supplied to them by the US military would actually show images of the places where these things were occurring the actual landscapes that we're kind of talking about here and that attempt to make these kind of hard to see I don't want to say invisible because you can always see thanks and good and I don't want to say that but these hard to see things kind of more visible I take that kind of out into the real world as well so for several years I've been doing these drone shadow works which are one-to-one realizations and that came of me just wanting to know more about these big machines a thing which I'm very unlikely to ever see in the real world but I have you know the tools to find out just the size of it right just a simple thing like that and go and put that out in the street where people can actually see and kind of measure oneself physically against the thing it really changes the way people see and understand these things and I've painted these all over the world and is totally open sourced it so there's a guide online to join your own and people have then subsequently taken that and done it both as art installations and at protests in all these kind of I just get like people sending me pictures occasionally when these things have appeared around the world which is amazing is very simple to just set out to draw a thing in order to understand it and then realize a whole bunch of stuff by doing that physical action what the drone shadows really emphasize is actually the intentional invisibility of these things the fact that they're designed not to be seen not to be seen you know physically because they fly so high they're invisible but not to be seen politically as well because because they automate systems that would otherwise lead to body bags and all the reasons you know the military's like drones and that got me into the habit of drawing other things in the street as well and this is a rainbow plane that I joined Ukraine a couple of years ago if you've spent half as much time as I have on Google Maps you might have seen the rainbow planes they kind of I find them deeply beautiful I find them if you're scouting around airports or whatever you might spot these things in Google Maps and what they are the artifacts of satellite mapping so the satellite is passing very fast overhead taking images of the ground the planes are flying very fast below them and the thing is that satellites don't see like humans see and this is the point of which I become fascinated because I'm interested in seeing how technology sees the world in fundamentally different ways than we do and satellites see using a variety of sensors both visible spectrum red green blue and in a high definition black and white in this case and deep into the infrared and the ultraviolet so they see things that we're not capable of seeing and this these little glitches are very interesting because they allow us to glimpse inside that system and see the way in which the machine is seeing the world and attempt to kind of reflect that back by by making them visible and addressable and kind of pointing these things out in this way it makes us possible to to think about them for me in slightly different ways and a couple of years ago I did this project in south London and this was called the right to flight and this is on top of a multi-story car park in south London where we built these silos one of which is a hanger for this blimp and the blimps kind of amazing it's the thing called a helikite which is a hybrid of a balloon and a kite invented by a slightly mad British guy who I had to spend several days in the pub with and the helikite is amazing because it's better than both balloon and kite it has helium to lift and but also a kite to fight the wind so it doesn't get blown down like a normal balloon but has much more lift than a regular kite and I was flying it off this rooftop to kind of investigate what it was like to be able to have that view from above that kind of classic surveillance God's eye view of the world how would it feel for me to be able to make kind of aerial images like this to to become to to access this airspace above London that was is otherwise kind of heavily restricted and it was a beautiful project and it ran for several months but it also weirded me out quite a large extent because obviously the first thing I did was whack a camera on this thing because aerial photos are beautiful and I want to see all this stuff from above and very quickly I started to become really uncomfortable with that because I was doing surveillance essentially and I was while I tried to inform people what I was doing and I and I looked at various ways doing I kind of released the footage openly so having to see the product the products of what I was doing and this kind of stuff. No one actually cared. This is in part London is not really giving a shit about anything but also I thought whether it's just a lack of awareness of the thing or whether it's just the fact that London is so subject to surveillance all the time anyway. This is kind of off people's minds but I realize the responsibility was entirely on me to stop doing this essentially that I suddenly got to a point where I couldn't make any more images like this and I realized that in many ways it's quite hard to make artworks about surveillance that don't essentially do more surveillance and that's a really bad pattern to get into because I don't think it helps I think it just kind of reinforces the logics of surveillance and normalizes it for people. So I stopped myself. I took the cameras off. I ended up putting like a kind of pirate box on there so people could share that felt like a more useful kind of relationship to have with it. And so I started looking at more kind of pedagogical ways to talk about these things like this project which is called CitizenX which is a browser extension very simply anyone can download it tracks your browsing but entirely privately data doesn't leave it. I don't share it with anyone including myself but as you're browsing the web you can hit the button and it will pull down and it will show you very simply stuff that's very simple for most people in this room I'm sure where the internet thinks you are and where the internet thinks the website you're visiting is. So it shows you this relationship of the physical infrastructure of the internet. And like many artists I've been kind of investigating the physical infrastructure of the internet for a while and thinking it's important to sort of point at this stuff and go you know this magic cloud place you hear about isn't some far away strange place it's like it's a real thing it's here it's large buildings filled with computers in legal jurisdictions subject to certain laws. But there's weird stuff going on under that as well. So with this project I didn't just want to highlight that physical internet I wanted to highlight what that physical internet does to your rights which is in this case based on NSA documents showing how the NSA assigns citizenship. And what the NSA does is it when it's pulling in all that data it looks at your browsing habits and it says well based on where this person is going on the internet and the sort of behavior they have there's this percentage likelihood they're American because that's an important determinant for the NSA because they're supposed to not track Americans so they have to have some mechanism for making this distinction. So moment by moment online your citizenship which used to be this really solid thing that used to be a thing determined by your passport by your place of birth that you could prove those of us who are lucky to hold stable citizenship. You could prove with a document and that's no longer the case. Your right to privacy is being determined algorithmically all the time by tracking of your behavior. And so what this next does is it looks at your very simple behaviors online and it builds you one of these algorithmic system chips to say this is what actually your web browsing behavior would look like if someone more nefarious was looking at it and how that might kind of come to affect your rights. And I was looking at the sky a while back because I was looking for these planes which are covert metropolitan police surveillance planes and hanging out in fields quite a lot looking for weird stuff like this. And what I found instead was was these flights. This is an example of why go out into fields and stare at the sky and find myself stood next to a bunch of kind of weird interesting people because you end up in the fields around airports with the planespotters, the chemtrailers and activists working on various subjects. I ended up working with a bunch of activists on tracking deportations from the UK. This is the system the British government uses to deport people which is basically where they hire private planes because it got embarrassing to do it on general general usual aircraft. And in the middle of the night they put people onto these planes and they fly them out of the country. And this is a very hard system to see because it happens in the middle of the night at private terminals, airports. It happens within closed courtrooms and within privatized detention centers. And so working with some friends of mine who are architects and architectural modelers we used planning permissions went to the council, we got the files on these kind of places, we did these kind of classic investigative journalism practices of finding eyewitness accounts, finding council documents, visiting and sketching places where we could, where we weren't allowed to photograph in order to visualize them. So this is a courtroom in the centre of London. In the UK it's illegal to photograph courtrooms. So there's no photographs exist in this space but there are very, very shiny architectural visualizations that we made of it. This is an interesting courtroom because it's a courtroom in which special laws apply. The UK passed a special law for immigrant appeals commissions, where there's suspicion of terrorist activity or some kind of security risk, where they can present secret evidence, which is evidence that the defendant and their legal team are not allowed to know about. It makes a complete mockery of the idea of a fair trial. And what's interesting is the way in which that appears as architecture in spaces like this, because you can actually point to it. The divider on the left here is what separates the public viewing gallery from the spooks viewing gallery. The closed witness box over there is so that members of the security personnel, security service, can appear without giving away their identities. And this culminated in a eight-minute film you can watch online called Seamless Transitions, which documents these various spaces. The courtroom which you just saw, a detention centre near Heathrow Airport and the final airport through which people are deported, which is Stansted if anyone goes to the UK on holiday. And what again I was trying to do in this film was not necessarily paint like just what you would traditionally do with photojournalism. So this was commissioned by the photographers gallery in London. And it was, when it first went on show there, it was alongside a huge amazing deep dive through the archive of something called the Black Star Archive, which is an archive of human rights photography from the 20th century. It's extraordinary but it's very much what you think of when you think of 20th century photojournalism, which is the kind of sharp black and white photo of face in a crowd, this frozen moment, this moment from which we can supposedly draw like this particular story, this one signal thing which will explain everything else. And it's a fantastic way of working but in this I kind of realised I was doing the opposite, which is I was using the architecture of these spaces and also the way in which work like this is put together, which is hugely based on research, assemblage, cooperating with other artists, visualisers and also cooperating with the software. Architectural software that makes this is the same software that generates a huge amount of the built environment around us and yet we still kind of assume there's one master architect at the heart of one of these things rather than several thousand people working on code or to CAD or however it is who now have a huge stake in our built environment. So all of these different pressures come into play on trying to depict not a particular place but the entire legal and infrastructural system that goes into creating the situation that results which is horrific deportation system. So that's a very brief and swift run through of some of the sort of things I do. I want to talk a little bit more about about this project, the cloud index, which requires a little bit of history, which I hope will be interesting. This is one of my favourite people, Lewis Fye Richardson. In 1912-13, Richardson was a meteorologist who was working the Esquimedia Observatory in Scotland doing a very, very early kind of weather calculation, mostly just recording the weather and trying to like think about what it might mean to start to have this kind of huge body of weather data. And then in 1914, he was called up for the First World War and he was a Quaker, so he was a pacifist, so he joined the ambulance division and spent the war running around with stretchers. The other thing he spent the war doing was that he'd taken a bunch of data, a bunch of readings from the observatory and from other observatories around Europe that had been gathering for the war. And over the course of two to three years during the war, while in trenches, while under fire, he performed the very first mathematical calculation of a 24-hour weather forecast, just over the area of Northern Europe, just over the area that you see here. But he figured out what the maths was to take a bunch of historical weather data and to advance that 24 hours to push that prediction forward into the future, which is an incredible achievement and I'd say it took him several years to do this with pencils and paper. And when he analysed the results, because obviously the day he was doing was by that stage almost a decade into the past, he turned out to be broadly correct. Some things were a bit more exaggerated, whatever, but his thesis worked. And because of course this was a time before what we now think of as computers, he thought this was a kind of achievement, but thought realising its full potential was a very long way off. He, in this book, the one I showed before, he goes on this kind of beautiful little fantasy where he describes what it would take for his weather forecasting to actually achieve some kind of industrial scale and be economically useful. This is a painting of that fantasy by Stephen Conlin from 1986 where he, Richardson described this vast globe on the inside of which will be painted a huge grid covering the entire earth which weather forecasters could kind of point to read data off, calculate it, have it replaced. You can see there be teams of mathematicians, then still called computers, working in the middle. And then this kind of amazing infrastructure of information, telegraphs and all the kind of communications infrastructure of the age, pouring information into this place and pouring it back out again. And so this was, he knew it was possible, but he thought, you know, mankind would never actually get to the point that it would be possible. Mankind did get to that point in 1950 when the very first mechanical calculation of a 24-hour weather forecast was performed, which was done on the ENIAC, which is a computer that some of you have probably heard of, which is my personally favorite computer. The first mechanical stored program computer built in the 1940s, the University of Pennsylvania. And this is the process for doing it. And they brought together a team of meteorologists from 1948 onwards. And in 1950 they ran this very, very first calculation, all the very first successful calculation of a single day's weather forecasting over the northern United States. And when they did it, it took about three weeks to run, which doesn't sound so great, but you have to remember that the ENIAC broke down a lot of the time. And what they realized was that actually if they subtracted all the time when the machine hadn't been running from the process, they got it down to 23 hours and 30 minutes. So they had, for the very first successfully time, calculated the weather faster than the weather itself was happening. This is an incredibly significant moment. And John von Neumann, who worked on this project, wrote in the project log that we have achieved the dreams of Lewis Frye Richardson. Computation now advances faster than the weather itself. And I like the fact that it happened on the ENIAC, which as I say, is my favorite computer. For those who aren't familiar with it, here it is. It was an early mainframe, so it occupied two huge rooms, first at the University of Pennsylvania and later at the Plymouth Proving Ground, Aberdeen Proving Grounds, where it was moved after the war. It required a lot of fiddling and plugging in. If you've read Chewings Cathedral, which is a great book about the kind of history of the early computational programs, there's a beautiful quote in that from a guy called Harry Reid, who was an engineer who worked on this computer, in which he says that I always thought of the ENIAC as a very personal computer. Now we think of a personal computer as something that we kind of carry around with us. But the ENIAC was a computer that you lived inside. Right? It completely surrounded you. And actually, for me, like that division of personal computer is obviously and I think increasingly not true because we all live inside this kind of shell of computation now. We know that these are mostly terminals to other connections, whoever owns them, whether we build these things outwards or not, whether we're talking to satellites or not. The ENIAC hasn't contracted into these things, it's actually expanded out into a kind of vast shell around us. And the other thing that I really like about the ENIAC is, you know, as one of these computers, these generation of computers were the last ones that were kind of truly physically legible. So computation now is legible to those of us, probably not me, who are really smart enough to kind of follow logically step by step what kind of process is happening within machines. With the ENIAC, you could do that physically with your eyes by looking at the blinking lights so that you could watch a computation kind of unfold on you across the walls. You could actually see that as it went from different parts of the machines through the architecture, you could kind of follow that. And that kind of legibility also seems to me something quite crucial to kind of go back to and reflect on. As I mentioned, one of the people who worked on that was John von Neumann, who I'm sure you're most familiar with as the kind of inventor of the von Neumann architecture who worked on a lot of these early projects and obviously was working on both the Weather Project but also on the Manhattan Project because this is here with Oppenheimer, who ran the Manhattan Project. And this same computer, the ENIAC, was used for both the weather forecasting programmes and for development of nuclear bombs. These are the kind of two grand computational arms races that really kind of got computing going in the 1940s and 50s. And von Neumann had this line which he wrote first in the commentary on the weather project but later recycled to talk about the Manhattan Project, which was kind of his view of what computation could do and particularly computational prediction. And he said that all stable processes we shall predict and all unstable processes we shall control. And that was his view of what you could do with computation. You could model and control the world's sufficient degree that you could predict the outcomes. So I wanted to take some of those ideas of the weather and our computational predictive abilities and bring them somewhat up to date. And I got particularly fascinated with what's currently happening in neural networks. Because I studied CSU Arbaic, I studied AI in the early noughties but basically the AI of the 90s that was dying. So we were taught neural networks as a kind of historical anomaly, like a nice idea, some interesting theory but basically a dead end. And the field kind of went quiet for 10 years and it's just starting to go kind of massive again, right? The last five years, neural networks have become starting to kind of become inside everything, essentially, in rather interesting ways. And so I thought it was important to kind of re-evaluate them and explore them again. And this is from a paper from the beginning of this year, things move fast in this world, of a Facebook research paper about DC GANs, Deep Convolutional Generational Adversarial Networks, which is something that they've been pushing and using. And this is from a neural network generation from that paper where they took thousands and thousands and thousands of bedrooms, pictures of bedrooms, and they asked this network to simulate new bedrooms. So each of the images you see here is not a bedroom that exists in the world. It's a bedroom that's been dreamed up by a neural network that's seen many, many pictures of bedrooms. And it's fascinating to me what we decided to train these things on. And the more public image of neural networks, which kind of everyone, a lot of people have seen, is the deep dream stuff from Google. And so what deep dream is, is an attempt to see back through these networks and try and understand how they think about and make images. So it's kind of instead of asking the network to look at an image and tell you what it sees, it's asking, it basically gives it blank or fixed images and I'll sit to kind of reverse them back through that network to kind of imagine and visualize what it can see in the world. And it turns out those things are terrifying. And I think this should be acknowledged more. When the guys on the machine learning team at Google first came up with Deep Dream, apparently they showed it to Google PR and were like, look what we made. And PR was like, never release that. And in fact, they weren't allowed to. The Deep Dream leaked, first of all. And I don't know how, but it's interesting. And then they had to come out and say, okay, this is what this thing is. And everyone went, whoa, man, computers are tripping balls. And didn't, and didn't, you know, there hasn't been a huge amount. And I don't think like, you know, the people behind it that are quite annoyed that there hasn't actually been a bit more thinking about what this kind of stuff entails because it's what neural networks are doing in the world is kind of interesting. You can do weird stuff with neural networks and visual neural networks. And DC GANs in particular, this one's the same paper. So you can do additional subtraction. You can take a bunch of images associated with certain concepts. You can then produce new images that combine elements of those things. So this is a network that's never seen a picture of a smiling man, but has seen these other distinct stages. And you can do, you can just matic on that to produce new kinds of image outcomes. And so I thought, well, if a computer can dream of bedrooms and it can dream of smiling men, it can dream of all things and it can possibly also dream of the weather. So I took, you might have noticed from the other works I'm quite obsessed with satellite imagery. So I took eight years of satellite data of weather over the UK. This is from the EU space programs, Meteosat, which sits up there well above Null Island watching the earth every day. I took several images a day of weather systems over the Western Hemisphere. Chopped that down into images over the UK, which came to about 16,000 images, which I stress is really not enough for doing this kind of work. But I only had eight years of data on the other input, which was polling data from the UK on their voting intentions in the EU referendum. I'm sorry about Brexit. I'm not gonna speak of that again. Except I am, because this is what this project was about. This is a short snapshot. I actually had eight years of polling data. It's a bit sparse at the end, but it was enough to extrapolate eight years of information. And so I built one of these networks with the assistance of Gene Cogan. He might be around here somewhere. And he's brilliant. And Fedet, eight years of weather images from space, eight years of images of cloud formations, and eight years of this polling data about the EU referendum. And my neural network dreams new weather formations. I dreamed new kind of cloud patterns. And once it had started to make these kind of associations, it was then possible to say to it, well, you've seen what the weather looks like on these certain occasions, when the voting is trending in these particular directions. So just as we did the addition with the smiling man, et cetera, let's do some addition on the voting and say, well, what should the weather look like if the entire country was to vote in one particular direction or in the other direction? To generate whether potential weathers that we could simulate if we wish the democratic outcomes to be different. And the ones at the extremes are what they are. I actually kind of enjoyed more the outcomes that played around with undecided voters. This seemed to me actually something that really required more attention in the aftermath of very divisive elections of the kind of various kinds that we've seen this year to try and put a little bit more emphasis on not on hard Brexit or Trumpism or whatever it is, but on the huge space within that of indecision and unknowingness that actually guides and hugely influences a lot of our elections. And the reason that I thought that this would be, in part, a worth all things to do is because we can change the weather now, right? If we want to. This is Vincent Schaeffer in his laboratory at General Electric in the mid-1950s. Vincent Schaeffer's the man who invented cloud seeding, which is the realization that you can place into early forming clouds crystals of silver iodide that will cause those clouds to precipitate. You can raise smog levels, you can burn off fog. You can enact quite large scale changes to the atmosphere. When Schaeffer's work started to appear in the 1950s, there was a huge public debate in the US and elsewhere about what the possibilities of weather modification might be. There was a huge belief, there was a huge military belief that the weather would be the weapon of the future. The US used weather modification extensively in Vietnam. It had a two-year program of seeding with silver iodide, clouds over Vietnam, in order to basically obliterate the Ho Chi Minh Trail. They reckoned they extended them on soon period by an average of six weeks each year, making it rain more and more and more. When that leaked, again leaking, they stopped it immediately and there's been little willingness in the West to continue cloud seeding experiments ever since then. But it doesn't stop other people. These are Chinese weather-making rockets. The Chinese government has a weather department of some 50,000 employees whose job it is to make the weather. They are active very much in agricultural locations, either making it rain on the crops or preventing it hailing on crops, but they also do a lot of work clearing the air for official parades for the opening of the Olympics for these kinds of events. So the weather is a tool that we could be deploying should we wish to and one that we might increasingly want to think about deploying in the not too distant future. When we have to ask ourselves quite serious questions about what we want to do with the environment, how we're willing to use the things that we know, how we wish to use those things, whether we have the strength of our beliefs in our ability to act in the world. And why do I do this? It's because I'm no longer convinced, actually, that it's possible to understand the world just through processing kind of vast amount of information that I don't think this is any longer a proactive, useful way of dealing with the world. But I realize that I kind of wanted to retain the cloud, not as a kind of bad metaphor of kind of lossy engineering or corporate control and not as that kind of materialist, look, I found a data cable, kind of like materialization of the world, but as a way of thinking the world, as recognizing that the cloud is cloudy, that what it reveals to us is often actually a kind of difference and differentiation around the world and that it is not a purely kind of fixed channel for information that's there to tell us how the world is, but asks us to question it and think about it more. This is increasingly necessary because this informational view of the world, this view that by gathering information, we kind of make more knowledge and more sense is increasingly at risk. I'm sure, again, Moore's Law is familiar to most of people in this room, the rule that the number of transistors doubles every couple of years in certain space, so you get more and more power, more and more processing power. This is its inverse. This is something that's been named a room's law by people who work in the pharmacological sciences. This is the graph of money spent on medical research against drug discovery and it's going the other way. The more money that's put into this, the less and less we're learning. And there's various proposals that have been put forward to various analyses of this, which is a long-running and long-standing problem in pharmacology. And one of the most prevalent theories is essentially too much information. If you know how drug discovery is done now, it's not basically people in white coats in a lab, it basically looks like a server farm. It looks like large machines, it looks like robots, and it looks like very, very large data sets of information being automatically run through these high throughput screening machines in order to test new drug reactions against one another. And this problem seems to be continuing. And one of the ways in which science labs are dealing with it is increasingly putting people back into the process. Not as some kind of like airy-fairy, computers are bad, people are magical kind of way, but simply that there seems to be some other way of thinking the world that doesn't rely on vast, vast data sets, but on plotting slightly distinct and non-machineic ways through them. The example I like of that is the story of advanced chess. So this is the famous game in 1997 between Kasparov and Deep Blue, when IBM spent a decade building a computer to beat this one poor guy at the only game he'd ever loved. And it was this insane moment, because we basically, our idea of intelligence is always like the one thing we're still holding out, and you'll notice in any discussion of AI that what constitutes intelligence has kind of receded for decades as the machines have got better and we've got, oh no, not like this. You kind of continually move the barriers. But at this point chess was the thing, right? And it was a really shocking moment when possibly the greatest chess player of all time was beaten by this machine. And Deep Blue was a brute force machine. That's what it did. It basically was capable of doing such vast search through the kind of development process space of the game as it evolved that it could outthink any human by a pure kind of brute force approach of thinking ahead. And that ultimately seemed to be right. We finally built enough processes to beat a human. And what was interesting is that while lots of people were kind of like, oh God, that's it, the machines take you over. Kasparov came back just one year later with a thing he called advanced chess, which was a new kind of chess, which was not humans versus machines, but humans and machines versus humans and machines. Because it turns out that today, and this has become very successful, not even a supercomputer today, a advanced computer will wipe the floor with any human alive, but a human using quite a relatively weak computer assisted by it will wipe the floor with even the biggest supercomputer. There's some kind of interesting complementariness to the way in which this thought happens. There seems to be one model of how we can think about our relationship technology, not as this kind of completely oppositional thing, but as a form of cooperation and kind of thinking through problems together. But at the same time, there's something also that's coming out of that at the same time, because this is where we next put intelligence, right? This is the AlphaGo game from earlier this year when Google's DeepMind, again, just a massive sportsport machine, played Lisa Doll, one of the greatest Go players in the world, and despite a heroic fourth game, third game, fourth game, DeepMind won. And so Go is now, again, one of these things that we held up as being such an important part of intelligence and is now kind of, you know, has again been taken by the machines. And while I really look forward to advanced Go and all the kind of interesting things that will come out of new ways to play it, there's something fundamentally different between DeepMind and DeepBlue. And that is, we don't understand how DeepBlue, how AlphaGo, sorry, does what it does. We don't understand how the neural network actually beat this role. There's theories and there's kind of ways of watching the game and picking them apart, but essentially this was a machine that was trained first by a team of humans and then trained against instances of itself and its learning ability while it played against other instances itself was kind of exponential. And it's thought processes because it's in a neural network in the way it's entrained that neural network is unintelligible to us. At some kind of deep fundamental level, we can maybe produce the kind of weird, deep dream, mind state images for AlphaGo, but they won't tell us what the process is that allows this to happen. Everyone's probably heard, this is the example of the strangeness of neural networks. I've checked this story and it doesn't appear to be true, but it's the classical story of how strange neural networks are or why they do things that we can't quite understand, which is that the U.S. Army commissioned a neural network to recognize tanks. They gave a team of researchers a tank, a bunch of soldiers, and they said go and hide in the woods, take loads of photos, feed them into a neural network, and we wanted a neural network that can see tanks in the trees, that can see camouflaged artillery and so on and so forth. So they went out, they took loads of these photos, they went back to the lab, they built a neural network, they ran all the images through it, and it worked perfectly. On this kind of X thousand images that they had, the neural network every time classified, yes, there's a tank in this image, no, there's not, even when the tank was hidden deep in the trees and invisible. As they went back to the army, they gave the neural network, and we're like here you go, we've done it all, and the army put its new pictures in and it completely failed, didn't work at all. And when they went back, because they were working with a relatively small dataset, they looked at the images again and they realized that the tank had only been there in the morning. And so all the images of the tank had been shot in the morning and all the images without tanks had been shot in the afternoon. And the neural network was really, really good at working out whether it was morning or afternoon. But it completely and utterly failed to see tanks. And these are the kind of strangenesses that occur when using technologies where we don't quite see the way the machine sees the world. And we're building neural networks into all of our things now. This is NVIDIA's programme to train cars to drive where they basically just put the neural network in the car, connect it to all the computers and driven it around for a while so that it learns to recognise the world around it without any external cues. It doesn't know the rules of the world, rules of the road, it doesn't know what anything else looks like, it's just watched humans drive for so long, it figures it out for itself. These things are becoming inside the things around us all the time. And we're actively building the tools of our own mystification, this I find particularly strange. This is from a Google Brain experiment just a couple of months old where they trained two neural networks to hide their communications from a third party. So they, Alice and Bob in this arrangement, evolved their own cryptographic system to protect their communications from Eve. And I myself think it's a really great idea that we should train the machines to be able to talk in secret behind our backs. I think this is a, I'm struggling to see why we would want to do this. This idea though of course of being able to gather all the more data about the world, making it to collect everything in order to make better decisions is of course one that we see across domains. This is the approach of surveillance agencies. This belief that by gathering all of the information, they will be able to make better decisions. They will be able to pull out and they will build a model of the world that's so complete that they'll have a full understanding of it on which they will be able to act. And it strikes me that this is not actually all too different from the way in which we also oppose this belief that actually the belief in information itself being somehow sufficient to change opinions, to sort of enact some greater good naturally on its own behalf is also the approach of transparency, of kind of full information release. That the NSA and WikiLeaks essentially believe the same thing, which there is some smoking gun information at the heart of the world that if we only bring it to light, everything will kind of magically be made better. And that seems to me to be an increasingly insufficient way of viewing the world. This is a small example, Facebook's analysis of hot topics for November 2016 with WikiLeaks right up there in the old men quadrant of most discussed topics from the time of the election. I'm a big fan of Julian Assange's original network analysis of secrecy that posed the kind of leaking as this kind of like grit in authoritarian regimes that will kind of make them grind to halt. But it does appear to me at the moment that right now it's kind of merely oiling a machine that none of us want to keep running. That we're locked in a kind of deadlock of an information war that we're both playing against each other in ways that are not gonna break us out of this opacity transparency dialectic anytime soon. And all around us these, it seems to me that it's becoming harder for these predictive mathematical information-based systems to work. And we need to ask very specific questions about what that's doing to us and to our societies. There's a huge fuss around polling and the poor performance of polls both in the referendum and Trump and everything else. But most of that seemed to focus on the fact that like the mathematical models were wrong and they'd have to be revised and we kind of build better ones rather than seeing these things as things that actually poison the discourse as part of the process. That when you are running polls, running metrics, running 40 models like this all of the time alongside everything, you're actually shaping the outcomes themselves. So when polls are done during a campaign, they don't just say what people are gonna do at the end, they change people's behavior during the campaign and they also change what is politically possible because people run polls and they say, oh, people don't like that, we're not even gonna put this out into the public discourse. Modeling and predicting in this way restricts the kind of actual possibility of discourse around us. The debate around fake news is a kind of startlingly interesting one, but one that is not also historically new or novel or should not be and also needs to be, I think, very specifically historically situated as something that as an idea of certain, it's suddenly that we can't know this stuff or that it doesn't emerge from quite specific ways of seeing the world. I was particularly struck by some of the reporting in recent weeks that went on and on about a lot of this fake news stuff emerging from the former Yugoslav Republic of Macedonia that's cited it particularly in this place without any particular context around the political situation in that place as possibly being related. So the story, if you don't know it, is that lots of kids in Macedonia and particularly apparently in this one town of Veles set up loads of these websites which were getting more hits on Facebook than New York Times or any kind of traditional media by completely making up stories and so getting tons and tons of ad revenue. And it's nice to me that this is a place in which they have a government in Macedonia which for the last 10 years has pursued a policy of fake news as government policy. This is part of Skopje 2014, part of the Macedonian government's project to build an entirely fake history for Macedonia by erecting statues of Alexander the Great and Philip of Macedon and classicizing buildings with kind of Hellenic columns and this kind of stuff in a completely kind of a historical move that's been opposed by lots of people in the country but as a result of a kind of huge resurgent nationalism a nationalism that actually depends on feeding a kind of fake stories. And we seem to be optimizing a lot of other stuff for this as well. We seem to kind of desire this inability to fully to see and understand the world. These are images that AI researcher Robert Elliot Smith discovered in his own photo stream when he posted originally the two pictures on the left and Google's, I think it's called like auto magical algorithm or something produced the image on the right which is an image that combines them both to have both people smiling but it's an image of a moment that didn't happen, right? It's a memory of something that never actually occurred. Likewise Adobe is pushing a thing that you might have seen called Voco which is a way of editing sound as you would edit text. So basically a tool for a kind of public tool a kind of Photoshop for audio essentially that allows anyone to kind of create these kind of things. Our entire, a lot of our technological advances seem to be in order to kind of create and change and mold the fabric of reality in ways that are not always visible and transparent to the visitor. And the outcome of that is a world in which people don't believe anything when complete authority of any kind of trust in institutions. No, I'm no one for blind trust in institutions either but the fact that when in the UK you had a referendum and something like more than 25% of views of participants believe that was going to be rigged by the security services is a very worrying situation to be in. And that is a result of a kind of a crisis of literacy and inability to deal with these increasingly complex narratives that are supplied by our technologies. More information as we saw in the pharmacological research we've seen in the neural networks doesn't necessarily lead to better decisions. In fact often kind of hardens people against them. The sociologist Frederick Jameson in his work Cognitive Mapping described conspiracy theory in particular as this attempt cognitively to map landscapes of information which are simply too vast for people to grasp. That no one person can hold this entire system of information inside their head. It's the same problem as big data. It's also the same problem we encounter every day on the internet because we are continually faced with stuff that doesn't map into our model of the world. And that is existential terrifying because we're quite weak lizard-brained creatures. And that we have not equipped ourselves with literacy for dealing with complicated and often paradoxical information. And as a child of the kind of hippie internet, right? This is deeply worrying because for me the internet has always been a place of incredible emancipatory possibility. A place in which you have access to all of this information and it makes the world more comprehensible to you. It gives you greater agency and allows you to make more social, better-ordered, whatever your decision may be, naive view it turns out, right? But a really important one to get across because we're living in a world where information has been increasingly available for a long time now and is increasingly driven by fundamentalisms and deeply entrenched differing points of view. So we need to think very carefully about what our responses are, what are the new literacies that we need to build not to understand raw information but to understand these differences between information. There was a, I took the title of this talk from a piece in the New York Times from a few weeks ago by the director of the Global Weather Foundation, a big weather forecasting agency, huge computational weather forecasting dated the full deal. And what he was saying was that alongside these growth of information available to us is also a huge and cataclysmic change. Climate change is now progressing to such an extent, the atmosphere is warming to such an extent that the models that we've built based on the hundreds plus years of weather data we've been gathering no longer apply. The weather models are actually getting worse because the underlying models that generate them no longer work so well in a rapidly changing climate. And we're actually moving towards a place where we're going to know less about the world than we did previously. We're going to be able to predict less and that's going to affect our ability to prepare for it. This, no not that, this. I'm running out of time and I want to take a couple of questions so anyone's got them. Here's what I think. I don't think the internet and so much of the stuff that we've built on top of it is a tool for understanding the world as some kind of abstract entity that we can regard as static and stable. Rather it's a tool for understanding our own ability to understand and interact with the world. And it's the most extraordinary, complex and advanced such tool we've built as a species. For me it's a cultural tool. It's something that emerges like language and writing emerged. And what's extraordinary about it is that no person or no single set of intentions created this thing or imagined what we might do with it for so its uses. It's an unconsciously generated tool for unconscious generation. And it's this model of complexity. It's kind of cloudy cloud. And if we are to take our kind of technological understanding of the world seriously, I feel that we must recognize that it's teaching us to expect and understand differences in the world, to understand cloudiness and complexity and not conformity and similarity. Thank you very much. Thank you so much, James. We now have a couple more minutes to take questions. Please use the microphones or show up at the microphone so I can see you. Or beer after this. Anyone? It's alright if not. Or beer later is also good. Okay. Are you sure you don't have any questions? This is your chance. I think we're good. Alright then. Thank you very much. Thank you very much.