Artificial Intelligence Meeting Aarhus Introduction

Welcome to the official Artificial Intelligence Meeting, ALMOS, as part of the Internet Week Denmark 2016. My name is Erik. How many in the room here know me already? Hands up. Five, ten people? Okay, cool. I thought there was going to be all my friends here, but even better. Nice! Thank you for showing up. This is the only event in Internet Week Denmark about Artificial Intelligence. The reason I organized it was because I got very, very sucked into this topic eight months ago. If you had asked me nine months ago, Erik, what do you know about Artificial Intelligence, I would have said none. So, this is not my life passion thing, but I have definitely spent a lot of time looking into it the last eight months, because I just can't let this go. And this is why I organized the thing today. I'm very grateful for you to be here, to hear about it. Why is Artificial Intelligence important? I call this talk a much-needed reality check. Why is Artificial Intelligence AI so important? I asked myself this question two days ago, because I really needed a motivation for this speak. It is... As far as I can see, relevant to all of the steps of this human hierarchy of needs, developed by the guy called Maslow. This is the extended version of it that also features the transcendent step, where you, after you get your safety, health, love, social esteem, knowledge, aesthetics, and personal growth in order, you are able to look at the world in a more altruistic way, and start to help others. You may be at different stages in your life, where you can relate to either of these stages. I, for one, think that they're all important, and this is a very good model. Let's not go more into that. Point is, Artificial Intelligence, I have found out, can be helpful to all of these things. We can get better food, more food, better sleep, more clean air, better financial security, better friendships, more knowledge, better realization of our own potentials, everything. I have so far not found anything where I say... where I was able to say Artificial Intelligence is irrelevant to this. As far as I can see, it's relevant to all of these topics. That we can discuss afterwards. But that is my motivation for standing here right now. What is Artificial Intelligence? It's just the name for intelligent machines, or intelligent programs, as a broad definition. But that's really what we need to know to get into this talk. And right now a lot of people are talking about it as machine learning, machines that learn. But I want you to look at it as something very simple, not a big black box of the unknown. It's just the name for some kind of machine or program that has been made that can think, or have thoughts, or do intelligent work itself. Examples that are already existing now that we can call Artificial Intelligence in some way can be Siri, that is voice recognition. That is the software that we all, or many of us know from the iPhone. Recommendations on your Netflix that can also be called Artificial Intelligence that helps make just the right recommendations for what you want to see when you open your own Netflix account. Now we have in San Francisco there is Google and Apple and other big firms developing Tesla self-driving cars. How can they do that? With the help of Artificial Intelligence. And that all comes down to, they have a camera on top of the car that can recognize if there is a person in the way. So the more they train that system, the better these cars can drive by themselves. And one very interesting new addition to the list that I've seen recently is cancer diagnostics. You now have software programs that can diagnose cancer. And they can diagnose cancerous cells better than the doctors can do just because they are better at pattern recognition. This is what it's all about. Can you recognize a pattern and then make predictions out from that? This is my short introduction to what Artificial Intelligence is. I'm 100% sure other people in the room have different precise descriptions, but this is just what we need to know to get this talk going. And one thing that's not on this slide but is actually a very good quote is a lot of the times what happens, something in the 80s that would have been called Artificial Intelligence, when it got so normal that everybody was used to it, even born and raised on it, you don't call it Artificial Intelligence anymore, you just call it software because it works. So for many people, Artificial Intelligence is just something that's new to us. But as soon as it's automated, like recommendations for what you want to buy on eBay just pops up, you're just getting used to it. You don't call it Artificial Intelligence necessarily anymore. Also a long discussion, but it's just a quote that I like. One thing that is relevant for all of this is that so far we only have what we could call narrow Artificial Intelligence. Why do we call it narrow Artificial Intelligence? Because there's been a new classification of what Artificial Intelligences are, namely the narrow ones that are superior to humans in some narrow field. For instance, chess. We already have that. There's already been built a machine that plays chess better than any human in the world. You have now cancer diagnostics that can diagnose cancer better. It's better than any doctor in the world. Those are all narrow Artificial Intelligences because they're only good at one field, one thing. They can run the fastest within one track or they can think the fastest within one subfield. But what really is interesting is the next step that has been called the general Artificial Intelligence, where you have something that is as smart as humans in any field. Some predictions from a guy called Ray Kurzweil, that we're going to hear more about later, has been that we could reach this level within the next 15 years. This is stuff like that that got me very, very awake while I read this, because what the hell. And then there's talk about super intelligent Artificial Intelligences that are smarter than humans in all contexts. Pretty hard concept to wrap your head around. But there's people right now discussing this. And some predictions say that we could reach this level of machines or intelligences by the year 2045. That's pretty soon. Two ways of looking at this could be either the optimistic or the pessimistic way. One of the most optimistic persons in the world looking at this calls for focus on hybrid thinking, mainly the fact that this is just man and machine merging. We're just getting smarter. Right now if you have a phone in your pocket connected to the internet, you have access to Google in your pocket, and that makes you smarter. It's a tool that can make you smarter. So it's just thinking enhanced. Another way to think of it is a more pessimistic way, that if you have a super intelligence that is not part of humanity, something else, then what do you do when it's smarter than everybody on earth? That is Nick Bostrom, a guy who is founder of an institute three years old in Oxford called the Future of Humanity Institute, that is one of the main proponents of this angle. He says we should really start looking at this now. This is one of the graphs that also got me to wake up the most. Many will recognize this as Moore's Law, that says that you can fit the data of how many calculations per second a computer can do as a function of how far we are in time, and you can see that it can be fitted to an exponential curve. And if you buy these assumptions that this graph is made of, you see that we have a prediction that around here, 2029, we could reach a computer that has the calculation capacity that could be compared to a human brain. We don't know that we're going to get there, but this is something we can predict and make a good estimate of, and I think that's pretty good motivation to start thinking about that now. If we have a computer that's as smart as any human brain in 13 years, what are we doing not preparing for that now? So let's get into some real-life talk. What is happening right now? Image recognition. This is a good example, also a very, very clear example, of machines that get intelligent right now. It's okay if you don't want to call this intelligence, but we definitely have to recognize that they're doing something that we usually only thought people could do. Recognize images. You can train a so-called neural network by first recognizing small bits like these colors and corners, and then after you get it to understand the concept of a corner and the concept of green, then you can take it to another level and start looking at building blocks that are bigger, and Google has done this, and other firms as well, training it to, after this, being able to recognize more complex concepts, going up to things that we thought really only the human eye could recognize, potato fries, a guy smiling, a car tire, until you have the most sophisticated image processing recognition here, and you, I don't know how many of you have tried this, but you can already now upload a picture of your dog, and the software can tell you, this is a dog, just by recognizing the patterns. This exists already right now. I had some fun with this, took a picture of myself, and then combined that through an app that is nine months old, called Deep Dreaming, with a Picasso photo, and made it generate a new picture. This took five seconds. And if you had shown me this ten years ago, I would have said, that cannot be done in five seconds, but okay. There's another artistic new version of my photo, also generated in seconds. So, another thing that's going on right now, is companies building AIs, artificial intelligences, softwares, call it what you want, that can play games. In 1997, IBM built a machine called Deep Blue, that beat the world chess champion, Kasparov, in chess. How many have heard of this event, number one? Okay, that's good. I forgot that, but okay. It just seems like it's so far away. Up until 1997, the main consensus was like, chess is the epitome of human intellect, no computer can ever do that. And then IBM showed that this could be done. This is 19 years ago. Five years ago, David Watson made a computer, an intelligence that beat the world champion in Jeopardy. How many have heard of that? Okay. And three months ago, Google and the company called DeepMind, created an intelligence called AlphaGo, that beat one of the earlier world champions in a board game called Go, that is the biggest board game in Asia, and is considered to be maybe the hardest board game in the world, that is played by a lot of people. They managed to build an intelligence that beat the world champion in this, and immediately after that, South Korea, the country where this took place, they announced that they invested $800 million in development of artificial intelligence, because they don't want to be left behind in this game. How many heard about this event? Okay, sounds good. I've been three, last three months I've been in California, so I'm not so updated on the talk here in August, but my impressions from Facebook was that nobody was talking about this, that's why I'm holding this event today. DeepMind is the company that did this together with Google, now they're looking at beating StarCraft, and after that, Doom, GTA, whatever. And that makes me wonder, what's the limit? If you can make an intelligence that beats all these games, what can't be done then? Google DeepMind's motivational mission for existence is to solve intelligence and then use that to solve everything else. That sounds pretty ambitious to me. But it's straight talk, that's what it says on their webpage. And they have beat every single, or at least a lot of well-known Atari games, by just having a computer learning to play them. Not giving it the code, but giving it the screen output. So just like a human being, looking at the screen, trying and learning until it got better than any human ever was at some of these games. This is... Let's see how fast this goes... Giefer just wants to show you of DeepMind's artificial intelligence playing Pong, anybody remember this game? Faster than any human has ever done. And it... Initially it lost every time it played this game, but after enough training, it obtained a level that was better than any human had ever been at this game. So you can start to think, okay, what about other things? If you start training the computer or the software intelligence to do this stuff, what is the limit? Okay. And it is of course also a big business that's coming now. I really don't have business as my motivation for getting up in the morning, but I have to recognize this and tell you about it. It's a field that attracts some of the finest minds on earth, and some of the biggest business too. DeepMind, the company that just helped Google win over this world champion at Go, was sold for 400 million dollars two years ago to Google. That's an unfathomable big amount. It's like 3 point something milliard of kroner, 50 guys. That's insane if you ask me. McKinsey has estimated that this will be a business of 50 trillion dollars towards 2025. That's now. And for comparison, the whole GDP of the United States is 18 trillion. This is going to mean everything. Flemming Beesbacher, the chairman of Carlsberg, has called this the fourth industrial revolution, a couple of months ago after the World Economic Forum. And also there's predictions about 50 % of all jobs that will be disappearing as a result of all these things that you can start to automate. That's two billion jobs. That means 50% of anything we're in this room imagining we can work with in the future is just not going to be there because it's automated within the next 15 years. Are we ready for this? I don't feel like it. That's why I'm standing here. More about business. Have a look at this slide if you're impressed by numbers. Toyota invests one billion. IBM, one billion. It's a huge business opportunity. But it's also a lot of other things that's important to say about it. Startups, 300 startups have recently been estimated at a collected value of 1.5 billion that are working with things that can be called artificial intelligence. In general, as you can see, these are different things. Computer vision, smart robots, speech-to-speech transition, stuff like that. But the main takeaway here is that it's the next big thing. Here's a slide I took just to show you how you can think of this as the fourth industrial revolution. After steam, water, and mechanical production revolutionized everything in the 1700s. And electricity came in 100 years later. 100 years later we had electronics and IT. And I would still think of artificial intelligence as just a branch of electronics and IT. But the thing we need to realize is that now when computers and everything get smaller and smaller, it starts to be able to interact with our physical systems as well. That means our body. Stuff like Google Glass. 10 years from now it could be just a lens or something even smaller. That makes me wonder. Okay, cool, everything is going at top speed with tech development. But will this make us happy? This Happy Planet report says yes. I wanted to show you this. We have Denmark in the highest right corner there with the most happiness and the most techiness. So you can be positive looking from this. Another thing that makes me really positive and that I think is way more interesting than talking about business and money is that this actually has perspective to get rid of money. Thank you. 60s of information technologies is the way the Singularity University calls this evolution that they have so far shown that everything, every information technology that becomes digitized, think of music that came from LPs to CDs to mini-discs to MP3s, has gone through this evolution. And it goes first through a phase where it's deceptive. People they say, meh, it's not happening. Until one day it's disrupting and it changes everything. Then suddenly it's dematerialized. Nobody needs a physical store to buy music anymore because it's just not material anymore. This is my favorite part because then it gets free for everybody and that means it's democratizing. So that is very very good news if you ask me. Because if you can democratize everything, you don't have to stress over money. I like that perspective. Peter Diamandis is one of the guys who co-founded the Singularity University, which is a university in Silicon Valley that specializes in looking at future predictions of what we can do with these technologies. He said in a radio interview, he called it, we're demonetizing living. I say, yes, thank you. He says, this is like the most important and exciting time ever to be alive. Science fiction is becoming science reality every single day. If any of you are hooked up at a newsletter from some tech news site, you will agree that nowadays it's every day you get some new breakthrough to listen to. So with these ways of looking at it, it's very very positive and there's lots of optimism to have. That was my first part of it, the positive note. Now it's time for some blood circulation, so everybody let's just rise so we can wake up for the second part. How many recognize this picture? The seventh floor of physical... The physics department at Aarhus University. My favorite place in the world, I think. Cathedral of Aarhus. Okay, move your limbs and get down again. Just a standard recommendation. Yesterday was the Internet Week Denmark opening reception and they made us do this as well. Okay, so a lot of positive perspectives. So let's just sit on this. That's good news. One of the reasons I'm standing here now, devoting my time to this, is that there are also a lot of things to worry about. And this is the first thing we should let Raymond Kurzweil say. He... I'm just going to read this. He says, we need to pursue AI. We're diagnosing disease, curing disease, cleaning up the environment, overcoming poverty. We need artificial intelligence. A lot of thumbs up there. But every technology back to fire has its problems and peril. Fire kept us warm, cooked our food, but also burned down our houses. We have much more powerful technologies today, so the problems and perils are much more powerful. Recently there was a lot of concerns expressed by Stephen Hawking and Elon Musk. How many have heard about those concerns? Okay. Um... How many are still thinking about those concerns? Cool. Because that's what happened with me. Eight months ago, I got introduced to some of these concerns for the first time, and I have not been able to let them go. Stephen Hawking says, I think the development of full artificial intelligence could spell the end of the human race. Let's just leave that there. Elon Musk, the founder of Tesla and SpaceX, says he calls artificial intelligence potentially more dangerous than AI. And Bill Gates says, I don't understand why some people are not concerned. So, if you're not concerned when you leave this room in one hour, please come and tell me why. Okay. Elon Musk had a visit to DeepMind, the company that's developing these arcade game winning intelligences, two years ago. And he came out with this pretty exhausted statement. The pace of progress is incredibly fast unless you have direct exposure to these groups. You have no idea how fast. It's growing at a pace close to exponential. And the guy who founded this company, Google DeepMind, also says, human level artificial intelligence is several decades away, but we should start the debate now. I have really not heard anybody say that. I have not heard anybody in Denmark take this debate. So this is me saying, let's start the debate now. What do we do if we have a computer that's smarter than any person in the world and it doesn't do what we want it to do? I feel like this is a good picture of the situation that could be the reality. This is my favorite cartoon drawing from a guy called Tim Urban, who made a 30 pages long blog post about this that I really recommend you read. This could be us standing there saying, haha, that's adorable, the funny robot can do monkey tricks, while we look at the supposedly linear development of artificial intelligence. But then, if it turns out, which many things indicate, that it's actually an exponential evolution, we could, bam. Stand in a situation like that, but fuck! And you have a supercomputer smarter than any human being in the world. In a matter of years. This is another brilliant re-illustration of this. People standing in Humanville station. Hey look, AI is driving. It's coming fast. I have been looking the last eight months for anybody who could tell me that this is not the case. I have still not found anything else than people researching this. So that's why, again, I'm standing right here. Then there's the field of Value Alignment. This is a concept developed for people who say, okay, if we cannot... What do we call that? Unko. Unko. Avoid. Avoid. If we cannot avoid development of artificial intelligences surpassing the human brain capacity, we need to at least work on now, that whatever we build has the same values that we have, so it doesn't kill us or do something weird with us that we can't even think about. I will let Stuart Russell, artificial intelligence researcher, explain that. This is a 60-second clip taken from the World Economic Forum last year. ...products of our intelligence. So if we can amplify that, then there is no limit to where the human race can go. But I actually want to point to a problem. And that comes in the utility part of the equation. So imagine, for example, that you ask your robot to maybe make yourself some paper clips. That it might need. And your robot's very, very clever. It takes you very literally. And pretty soon, the entire world is six feet deep in paper clips. So this is the Sorcerer's Apprentice and King Midas all rolled into one. Now technically what happens is that if you ask a machine to optimize and leave out part of your preferences, the machine will set those elements to an extreme value. For example, if you say, Google Car, quick, take me to Zurich Airport, it will max out the speedometer. And they say, oh, I didn't mean break the speedometer. Well, it will still put its foot on the gas, and then when it gets to the airport, slam on the brakes. So this is the problem of value alignment. And if you combine misalignment of values with a super intelligent machine that's very capable, then you have a really serious problem for the human race. So the point is that machines can and will make better decisions than humans, but only if their values are aligned with those of the human race. Yes. Is everybody getting the concept? The concept of value alignment? It's okay to say no? Okay. Otherwise, come talk to me later. This... Let's go back to... Value alignment has now been a big topic, after people realized that this could be the only real thing we can prepare if we suddenly have a computer that's faster or more powerful than the entire human race. This is the guy, or one of the guys, that thought about this earliest. I think 10 years ago or more, he started posting stuff about this, that he was worried about what are we going to do if we end up with a machine intelligence that is smarter than the whole human race. As he says so chillingly, the AI does not hate you, nor does it love you, but you are made of atoms which it can use for something else. Wow. Elias Yudkowsky has founded the world's first research center that looks into this full time, every day, with some of the smartest, brightest mathematicians in the world, Machine Intelligence Research Institute, or MIRI. It's based in San Francisco, and I was there five weeks ago. He was not there, he was working from home, but a lot of his colleagues were there, and I met them, and they are working full time on looking at what can we do in advance if we assume that this is going to happen. This is an illustration that I like to use to explain this. MIRI's motivation is on top there. We want to help humanity prepare for the moment when machine intelligence exceeds human intelligence. If we imagine all of our thoughts, 7 million people on earth, right now, through all time up until now, is this pink circle, human minds, and we have then some look at the future, that you could call transhuman, you could just call it future human, you can imagine that when we get better machines, and more knowledge exchange, and new inventions, we will have a bigger mind space. We can think brighter, beautifuler, and better, and a lot of new thoughts. And maybe in the distant future, you have something you could call post-human mind space, 10,000 years from now, that's going to be even bigger. Then the value alignment problem lies in creating an artificial intelligence that is aligned with where we want to go. And if you just create artificial intelligence without a concrete end goal, you don't know if you end up with a gloopy AI, or a freepy AI, or a beeping AI. And the main concern is that the technological development that's going on right now is not controlled, so we could end up anywhere. We don't know what's going to happen. That's the concern. This guy is the one I mentioned in the beginning, he's one of the main people concerned with this. He wrote this book called Super Intelligence, Past Dangerous Strategies, that came out two years ago. It was a New York Times bestseller. I called Gyllenhaal, they never heard about it. I was like, what? I called a lot of people, I called the Danish, nobody heard about this book. But it is one of the biggest books that gathers all, all that is known about the subject, and also looks into the, the future. What should we do if we take it serious, that this could happen? So, there is a copy of that book in the back of the room afterwards, if you want to have a look at it. I bought this book, I got around 100 pages into it, it's very, very, very hard to read. How many knows this book? That's good. This is the first time I experience this. Okay. One of my favorite subjects, or parts of this book, is that the author calls what needs to be done philosophy with a deadline. For many years, humans have done philosophy, but maybe not thinking about that you need to get to some sort of solution. But if you have a concrete date in the future, very near future, where this could be of high relevance, we have to start thinking, but also have a timeline, on the thinking. And the most beautiful part of the book is the last section of the book that he calls Will the best in human nature please stand up? Which is basically a call to the whole human race to take this issue seriously. The guy is, as I said, called Nick Bostrom, and he is the director of an institute called the Future of Humanity Institute. I announced that we would have a guy called Kajsotala from Finland coming to talk at the end of this event. Unfortunately, he turned in sick yesterday, but he has hooked me up with a Skype connection with another guy from the Future of Humanity Institute called Stuart Armstrong, which is going to join us after Henrik Fruint makes his talk after me. So we will hear from the Future of Humanity Institute in around 45 minutes. He has made, Nick Bostrom, which is the leader of this institute, has made this overview over what are the various threats to mankind. As you can see here, climate change is down. Up here, and then you have synthetic biology, then you have nuclear weapons, then you have nanotechnology weapons, and up here you have machine superintelligence, which is, at least from that Future of Humanity Institute's perspective, the highest risk threat to mankind. This is also only two or three years old that has been taken serious to this extent. And let me just give you another reality that I also checked that really chilled down my spine a couple of days ago. This is a movie from 1970 called Colossus, the Forbidden Project. Has anybody seen this movie? One person? Nice. We build a supercomputer with a mind of its own, and now we must fight it for the world. This is IMDB's description of the storyline. As soon as the link is established between two supercomputers, they become a new supercomputer and threaten the world with immediate launch of nuclear weapons if they are detached. And then this jumps up. IBM brain-inspired supercomputer helps to watch over U.S. nuclear arsenal. Who's got a solution for this? I don't know. So in conclusion, we might be able to cure all diseases, which means immortality in the end, if we have artificially intelligent and guided medical equipment. We might be able to print everything with 3D printers, if it's helped out by artificial intelligence, and smaller and smaller atomic precision manipulation. Hello, abundance. We might be able to overcome poverty on a global scale. Goodbye to the importance of money. We might solve global warming with artificial intelligence. This is on the agenda of the D.I.F.E. mind firm right now. We might have increased artistic expression for everybody. If everybody can paint Van Gogh pictures or do Mozart symphonies, maybe we can become happier people, because there's not going to be that artistic move. You have to be a genius or a talent. Everybody can do it. That's good news. Did I mention cure death? I think that's the most crazy thing about it. But the dark side is, we might all go extinct if this fucks up. So this is why I organized this event. I have been zigzagging between those two states of mind the last eight months. And I still cannot conclude that either one of them is the truth. And after traveling the world and contacting a lot of people, all I can see is that nobody has a certain answer about this. You can have beliefs, but I would like to hear all of your beliefs on which side we're going to end up in. But don't panic. This is Stuart Russell that you saw talk about value alignment before. There's actually now a research field looking at this. This is very new. Fifteen months old. January last year, we had the world's first safety conference on artificial intelligence in Puerto Rico. Elon Musk from Tesla was there. Demis Hassabis, the founder of DeepMind, was there. A lot of the people you heard about was there. All the people you heard about was there, except Raymond Kurzweil. He was not there. But after that, they got together 8,000 of the world's artificially intelligent researchers and wrote an open letter calling for robust research. And we should focus on actually knowing what we're doing when we're developing these things. And in January, also, Elon Musk of Tesla donated $10 million to this field. 35 different people received these grants. One of them was who's supposed to be here now. Another one of these guys was Stuart Armstrong who's going to be with us soon. How many heard about this $10 million donation? Okay, good news. In October, there was the world's first UN meeting in New York about artificial intelligence as an actual existential threat. And in December, Elon Musk, together with other people, founded a company called OpenAI that focuses on that we should open source the development of AI so as to not centralize that in the hands of some evil people with $1 billion. And a couple of days ago, the White House said, the White House announced that the US, that they're now also putting down a strategic committee to look at the risks and benefits potentially of artificial intelligence. So, from what I have gathered the last eight months, we have four of the main institutes in the world looking at this MIRI, Machine Intelligence Research Institute, in Berkeley that I visited, and Future of Humanity Institute in Oxford that we're going to get Stuart Armstrong talking via Skype in just a moment. And then there's one in Boston called the Future of Life Institute that also looks at this. And the Center of Studies for Existential Risk in Cambridge. Some of these are founded by Jan Tallinn, the Skype owner and other big business people. So I see no other logical conclusion than we need to join the game and create an artificial intelligence research in the new halls inspired by the same for nanotechnology, 10 years ago. I studied at the Einem Center. So this is actually also why I gathered you here. Everybody who's interested in helping me with creating a research center about this topic will be welcome to talk to me afterwards. Time for action. There's other ways that you can get into this. These are just my ideas. Contact people of the local parliament and have them actually look into this. Contact your local school, university. Tell them to have a theme day or week about this. Have a look at the blog post called Wait But Why that I featured these comical strips from. It's very easy to read even though it's long. Or dive into Nick Bostrom's Superintelligence book, which is a very hard book. Contact the most influential people you know. Tell them to get on this board. Or contact the new news channel that I've created for this called Reality Check Denmark. That is the official organizer of this event today. And I'm also looking for help to organize, found and fund a research center for this. I'm meeting with City Hall about this in two days. But do get at me if you want to join this. And then finally thank you to today's sponsor, which is Zern from M2 Rock Development Center sitting in the back that have given me some money so that I can invite you all to dinner. Now. At 5 o'clock when this is over everybody who wants to participate in a workshop about this, we have a meeting room on the other end next to the cafe where dinner will be free. So please feel free to hang around and join the rest of this discussion because as you hopefully feel right now this is a pretty big topic that needs a lot of inputs. This is my found work. Let's save the world together. And here's my contact info. That was it.