 The next talk will be about something that really is disconcerting for me, but that's also very exciting. We've come to terms with the fact that machines have surpassed us on many fields. For example, in calculating and as scientists, we use computers for data structure and data analysis. But one region, one field that we can still retreat to is art, is creativity. But the question is, is that still state of the art? Is that still true? And for that we will have a wonderful talk on a person who is an expert for this, Simon Hegelich, who is a professor for political data science at the University, at the Technical University of Munich and he uses computers for an artificial art project. I'm impressed and I think it's great that so many people want to listen to this and I had the chance yesterday to talk to a lot of people here on the Congress who use the computer to produce art. So I will try not to use up all the time in my talk, but leave space and time for questions and answers, because there will be many controversial issues. Because the two terms that come together here don't belong together, art and artificial intelligence. And if you think so, you're right. But I want to show you the state of the art, the state of technology, what is done at the moment. And I also want to present two poems. One was written by an artificial intelligence and the other one was written by Jim Morrison. And after that I would like to ask you who thinks which poem is by whom. I'll start. My mother is good for her. First poem. Second poem. Second poem. Who thinks that the first poem was by Jim Morrison? And who thinks that the second poem was by Jim Morrison? I can't really see in the lighting of this room, but you see that both answers would have followers. Let me quickly start the video. We're already at a point where computers can generate art that we cannot easily distinguish from human-generated art. So first I would like to tell you what did we just experience. So for some years we've experienced a hype about deep learning. And deep learning is much more than just a hype. It's a procedure of machine learning. The basic principle of this is if you break it down very simply that we use unstructured data and we already know the answer and this is supervised learning. We construct a very complex mathematical formula to transform the input into something that we think should be the output. And this principle is basically something that I do not need for this principle. I do not need deep learning. The question here is what kind of mathematical formula do I need here? What do I construct here? And I can do this with texts. I can do this with words. I can transform words into numbers and then put them into a formula and then I get the output that I expected. What I used here is a sequence-to-sequence model, which is a deep learning model. First there is a class of algorithmic learning called a neural net. That's something very simple. I use words and every word gets assigned a number. So many people say that neural networks are like the human brain, but that's totally not true because it doesn't have to do anything with the human brain. It's just another way to represent mathematical formulas with a graph. And the mathematical function for this is fairly primitive. It's about multiplying input with parameters that have to be defined and all of this is added up. And the question is, does this activate another function? And then I can shift layer after layer for that. The advantage of neural networks is that I can put in a lot of parameters into my mathematical formulas and thus have a very efficient algorithm for those parameters for optimization. I present the model and the computer calculates the values for that. Deep learning goes beyond that. The difference here is that signals aren't just transferred from one to the other, but I'm starting to do something very complex with feedback. This is a sequence-to-sequence model that works with attention. This is actually fairly primitive. I have a sequence as input, a sequence of a sentence, and an encoder produces a signal and the decoder gets the output that is already known in different texts that can vary in length. It's not a word-by-word translation. And this transforms the input from the encoder into something else as a decoder. So, one other thing that we will see is that there is a lot of randomness in this. And it's very interesting to have several neural nets at once and train them at the same time and then see what is the output here. That's the principle of the attention network. So, that's what we just saw. That means we have an input. And now the question is what was the input? So, we train the network that I used on the canon of American Lyric, the complete canon. And the input is the first line of a poem and the output is the second line. Actually, this is an algorithm that is used to translation, automatic translation, for example from Spanish to English. But I could also use it for something else. For example, here, I feed the network the first line and then it should tell me the second line. Approximate the second line of the poem or predicted, sorry, predicted. So, now we have the model, which is in fact a mathematical formula. We have 160 million unknown parameters in this model that need to be optimized. That's the reason why we define deep learning models as black box models. Because once they're done processing, we don't really know anymore what influence which parameter actually had on the model. We can only with generative models, for example, so further proceedings understand which patterns actually have been learned by the network. And that's an interesting, this is an interesting procedure to understand, to generate new creative models. So, one thing we use is style transfer that you have seen in these videos, but also deep dream, which has been developed to see which patterns these neural, deep neural networks actually learn to visualize what they learned. So, these are the new tools that we use to see what the networks are doing visually and what they learned. So, now we can understand, we're trying to understand what is artificial art. I don't believe we have artificial art, but I think we have pretty cool tools right now that we can use to understand, that do interesting things. The question is, if we manage to make a computer produce something that looks like art, is that enough to actually make it art? And I would say no, that's not enough. We have a lot of creative tools that we can use, but they don't really seem to be different than just novel brushes. We have new procedures, but it's us actually who are the creative entity. The computer seems to be creative, it seems to generate art, but I feel like most people will agree with me that this isn't really the thing that we mean when we say art. So, if we sit down today and everyone who has studied art could probably sit down and could probably draw some pictures that look like Picasso, because sometimes art isn't even that difficult, for example Picasso wouldn't probably be difficult to draw, and that's not even the challenge sometimes. So, we realize if we're talking about art, it's much more than just tricking the audience into who produced art. We have to go away from the old idea of the Turing test, because Turing based his idea on the question, if we cannot tell whether a communication is led with a computer or with a human being, then we have real artificial intelligence. But that's not true, just because a computer is able to trick us, that does not mean that the computer is really able to think or to produce art. So, just because the computer produces something that looks like art to the untrained eye doesn't mean it's art, because there is no creativity, there is nothing new, and that's the main reason why those systems are really, really conservative. So, because it only uses data that's already within the data that it is given, so it doesn't create anything new, so it's just pattern recognition, right? If there's just a pattern that's encoded in the data, so, you know, these systems are great at recognizing these patterns, so that's definitely something that they do that we can do maybe as humans, but after recognizing these patterns they're only able to reproduce these patterns, and I think this is like the opposite of what we talk about when we talk about art. And so, now you might say, you know, this is some sort of like circular logic here, sort of dialectical. You could argue, well, determinism and art, that doesn't go together well, but so, if something is random, it's not determined, right? Yeah, it's true, but the first point is that, speaking from the technological point of view, these models are huge. So we have 160,000 parameters that I've talked about earlier, and all of these are random, and so now you could ask, well, is there randomness within computers, and that's a question for computer science, but we're not going to talk about that today, and so, if we assume that this network is initialized at random, then it's very easy to initialize even more randomness, if you want to. And so I can create systems that randomly create new things, but again, randomness and art isn't quite the same, is it? And so, if we do that, you know, we have the question, is this art or is it not? Is it rubbish? And so that's what gets interesting, because art theory says, you know, even though art theory deals with these questions, it's not really suited to answer this question, because how do I define what is art in that case? And then very quickly, you know, the answer refers to the onlooker, the audience, but who is the audience, the people looking at this? Well, then we have to say, you know, the poems we just heard, you know, all of it was art, both of those was art, because there were people who said, well, I think this is the one written by a human. But if I go away from this abstract general audience, then it becomes quite a lead. And then we have the question, is art something that is defined by institutions through power or persons with authority? So then art would be what art critics define as art or institutions define as art or what sells on the art market. And again, we have, that's not a clear-cut answer, and many people would not agree with that. And so, yeah, if we talk about what's being seen as art, then we have to ask who sees it. And so if we disagree with these elite approaches, then... Well, we have this authority that decides what is art and what is not. Then maybe today we're at the point where neural networks can define these authorities. So we have GENs, Generative Authoritative Networks, and the principle is quite the same as we've seen with the poem example. And so while I'm training the network that's building these poems, we add a second network that gets real poems as input and these neural network generated poems. And so every time the second network makes the right call, it gets a treat, basically. And that way we have a feedback loop and a quite complicated mathematical formula to optimize these different parameters at the same time, but in theory it's possible. And so if only we had enough information about this authority that decides what is art and what is not, then we could actually have this decision-making on a technical level and maybe could even make more competent decisions than the real, quote-unquote, authorities currently do. And so yeah, we're at the point that it all fits together, right? The question, what is art? If we want to hand that question over to an external authority, then we could do that this way. Maybe we have other problems that way, like is it enough and the current state of research, but I think monthly we see progress in this field. There's so much going on. Or we could say this doesn't suffice to really treat this question about art. And that's my personal opinion and I'm going to go out on a limb here and just say that art is intentional. And maybe you might disagree with me, but I think it's about intention. It's about using your own skills through painting, through lyrics, to express something, feelings or abstract thoughts and ideas. And so it's not just about what the artist wants to say with that. So if the artist intended to write a poem about winter and everyone else sees it as a poem about death. So the intention of the artist isn't that important. I think the intentionality reveals itself in the piece of art afterwards. And if we ask the artist about is this what you meant when you created this, they mostly lie. And so it's not really about having an overlay there. And so I think this is a negation of randomness. We had randomness as negating determinism, but now I think we have to have the negation of the negation. And I think that would be intention or will. And now you might ask, do we have that right now? And the answer is no. Will we get this into computers in the future? That's difficult. There's not any approaches that look fruitful right now. And you know, with deep learning, you have this approach to build deeper and deeper neural networks that somehow interfere. But I don't think that's a viable option because you can't escape the sort of determinism. But in the field of art, you know, what is intentional about it, it's not that complex, right? It can be, but it doesn't have to be. And so if you would get a computer to, and this really is a project I'm currently working on with my company, working on strong AI to sort of take the next step in general artificial intelligence. And so the problem is to really get the computer to disagree, disagreement as the basic form of negation of intent. And so you would need to get the computer to not just disagree, but also to have these disagreements synthesize on a higher level. And so now we're at helium logic, which I also really love, but I'm not going to talk about it so much. And you know, this is a logic that's not binary, but that really enables disagreements to be able to be fought together. And of course there's a Latin saying omnis determinatio negatio. So everything that's determined is sort of a negation. And so this is sort of the basic saying every time I talk about something, I really talk about what it's not. And the second thesis would be identity and difference. And so if you think about that coming from a computer standpoint, you have this recursive loop. And then that's the third point that says, you know, this movement from nothing to nothing back onto itself. And I think that's the way to really soon have computers, not just in like 50 years or so, to have computers create art on their own. Because they can distance themselves from what they've learned by negating it. So they learn through deep learning. That's really important because they need the numbers. So I need deep learning to have structures, to have patterns. And that's just the same as for humans. But then you need to negate what you've learned. And this negation has to be transformed into something new, into the negation of negation. And how this will be possible exactly is a very difficult concept. So this was roughly what I wanted to tell you. Have I forgotten anything? Actually, we can just start with the discussion and see if this has inspired us. Okay, then we have a lot of time to discuss. Hi. Thanks for the inspiring talk. Thank you. You really led really well through the discussion of artificial intelligence. I really like that you didn't take a stance right away. But I still think that it's not the right way to see in the intention or in the entity that produces the creativity to look there for the definition of the art. If I glue banana to a wall, then that's not art. But if I do it at Art Basel, then that context allows it to become art. And so we need to distinguish between the system and the environment to say art is a social environment. And what is accepted as art in this environment? And so that's fluid. We discussed, we consent on it every time again for years. We redefine it, what becomes art. And that carries us closer to the understanding of what is art if we want to actually answer it. Well, thank you very much for this statement. First, I think at the very beginning I tried to allude to something. I do not want a definition of art that is without contradictions. It's a basic thesis of Hegel that our false softness towards things that we do not let contradictions touch us, even though contradictions are those very things. So, for example, you can tape banana to the wall. What was the intention behind that? Especially in modern art, we have an intention behind that. I want to show with that that art has reached a limit. It's contradictory because it only works if I do that during an exhibition and not at home. With this example, you can also see that the critics are not sure how you can see this. It's about this abstract idea, this intention. Is it really the point that if you're in the correct room and the correct space and then suddenly it's art? Or can I just say, okay, I've seen this a thousand times already. But if it's so difficult evaluating things that we as puny humans think what is art, then the threshold to decide what is art for a computer has to be very difficult. And one more important point that I forgot before, my entire summary, what I wanted to say, please follow me here for a moment. Imagine if we had computers that suddenly could produce art, we agreed on a definition of art, and we agreed that computers could produce art, and now imagine we have this computer. And then suddenly we would enter a new era because suddenly something that had been exclusive to us would no longer be exclusive. Just think of chimpanzees producing art. So we will arrive at a moment at which computers will have cognitive abilities that used to be a human prerogative. So how will we protect computers that are creative as artists from humans? Hi. One sentence. I'm from media art. And I think it's a steep thesis to say that humans who studied art can draw something. That's a steep thesis. I studied art, I never held a brush in my life. But besides that, there's many people that do art today that shouldn't be art after various definition. But it sells well still, but according to all definitions, that's not even art. So art is a social practice. And we sell it and we buy it. But also we hang around vernisages and we drink beers. But how should I make art without the context, basically, without hanging out at vernisages and drink beer with it? I think the hanging out is something that we can really, really easily do. But surely, art is a social phenomenon. It's produced within a social context. So I can't take the position that there is no art without social context. So I could imagine an exhibition where computers admire computer art and just consume electricity. But in general, I just like to say, yes, you're right. There are contradictions. We always think that there is a fixed definition for art and apply that to everything. What you're saying is true. A lot of the works of art that are sold at Sotheby's are something that is produced in the context of Sotheby's and of commercials. Thanks. Thank you for your presentation. I also studied art and I have to say I was desperate throughout your whole presentation because, first of all, a small critic from me. There isn't art anymore. We only have artists. We have a widely spread definition now. We can't work with a unified definition of art. I'm just wondering if you are doing research in this field and if you want to pull it all together. You also talked about Picasso and Hegel. So my God, Hegel is over 200 years old. How can you build a scientific base based on Hegel in this day and age? Take something from the 21st century, but not take a different basis for your theories. We only took Hegel because it was exclusive back in the day, but today we should have better fundamentals. If you do research in this field, do you also take into account more novel artistic phenomena or only the older ones? So what do you use as inputs to generate this artificial intelligence output? Thank you very much, but I have to contradict here. There is no single definition of art. If you say that there are arts as a plural, you do not escape from the central question that there is no definition. You just expand that one definition, but I think that is not true and I said that before. And there are inherent contradictions in term art and that's what I wanted to show. I want to show that if we talk about artificial intelligence art, we suddenly apply categories that we would probably not apply to humans. And concerning Hegel, I'm a bit of an outsider here. There is a philosophical break here starting with Kant. And that questioned our worldview at that time. And we can integrate that into artificial intelligence and I'm also happy if other people do something different. Questions from the stream? Stream has a short question, simply. Can you please say something about artificial intelligence and how it generates music and summarizes state of the art? That's a quick answer. It's basically the same algorithm. What is done is that we take media data and we then can treat them like texts with the same algorithms. I tried it, it sounds a bit random, but probably I shouldn't have used all yes records. Thank you very much for the interesting introduction. I have a comment. I think art is to free ourselves. That's why I think external authorities that you want to feed through the artificial intelligence or you want to play with that, they don't belong there. Any type of authority has nothing to do with art, in my opinion. Because everything that has happened and everything that hasn't happened is relevant for art and has been like that. And I also think it's difficult to only look if something can be generated because that's not even the point. The point is there must be an attention, but also an intensity. And that's actually the relevant thing, I would say. And I would totally agree with you. I think it would be great if we could arrive at a point where computers can produce something that moves us. But that in turn produces more questions. Will that move the computer emotionally? There's a meta level to this, but basically I think you're right. Hi, I would like to say two things. One thing. So is photography also art? I would like to ask, because what we saw on the screen, we could say that's just information processing. So that's like what a camera also does. And also the development of it. When did photography start? When was it recognized as art? And I forgot my second point. So that shows that the definition of art changes, the idea of what is art changes historically. There is Walter Benjamin's art in the age of mechanical reproduction, and I think we have moved beyond that. I think we should think about the work of art in an age of mechanical production. Something is changing. And in this talk, I hope I could stimulate you to discussion, because I don't have the single answers. Humans can stand up and signal can't. In the question what is art and what is it about, I ask myself if artificial intelligence or the computer can actually give the answer. Yes, that's the point. So if we have a very strict idea of art, on the one hand, as a pure simulation, then we will find that it's not satisfactory, it's not enough. We want to have something in there that we attribute to humans, creativity or intention. I don't know how we can define that. But on the other hand, if we can imagine computers that can actually do this, then we can think about this. But maybe it makes us humans to recognize that a banana glued to the wall, that that is art, and not because we can do it, but because we can recognize that is art. That's what I meant. That's really what I meant. It's a philosophical experiment, because there are no computers that can do this at the moment. But if we can imagine such computers, then we can imagine computers who can also appreciate it. And that raises the question, what makes us as living beings special or different? Those are basic ideas of human existence that we have suddenly transferred to a computer. Could the computer not come up with the definition of art that is actually hidden to us humans? I think computers could produce art that we cannot understand as humans. I have a question. According to your opinion, is there the possibility that you will say, okay, this art is by the artificial intelligence, because everything plays together. Who is the interpreter? We can relate this to Hegel that every single detail will have an effect on the result. And also all the programmers who worked on the model, who is the artist? The question of authorship is something that is really discussed very widely. If there is an author, then that author should be a collective, because the algorithm has been developed by several people. A lot of it is open source, a lot of it is developer communities, the large internet companies, Google. They really further this development. Next question, who provides the data? So if the artificial intelligence learns with my paintings, am I the artist? So right now the question is, who has the rights to those algorithms? Who is the inventor of them? This has to be discussed and cleared legally. It's a fascinating philosophical question. What do we got from the social networks? Question on Twitter. Do you have tips for newbies who want to get into the topic of artificial intelligence and art? Well, yes. The fascinating thing right now is deep learning. And the community here is very, very open. But it's also influenced by the large companies for obvious reasons. For example, TensorFlow by Google. And there you can find very, very nice tutorials how to build models. But it also requires a lot of knowledge about coding. You assumed at one point that artificial intelligence can only recognize pattern while humans can generate something novel. Which now I have the question, isn't the brain doing the exact same thing? What do we do differently than just recognize patterns in the environment and putting them together? Because in principle, that's nothing novel. That's a huge question when it comes to cognitive sciences. And here you're really touching on something difficult in my argumentation. A lot of researchers and scientists would say there is no free will, there is no creativity. It's just a lot of neurons firing. But if I say that is true, then I can just use deep learning and deeper and deeper neural nets. And at one point you will have a creative spark that I do not understand any longer. But I think we're doing something different. We're not just recognizing patterns. I think we can generate contradictions. We can speculate. And that goes beyond the patterns that we just have. For example, just recently I met a young child on a street. And his father came up and I was wearing a helmet with a visor. And the child said, look, there's a motorcycle rider. And the child used its own ideas of a pattern recognition with my helmet, with my headgear, to say that this belongs to a motorcycle rider. So this is what humans do. An expression that I wanted to add is glitch art. And then we can get back to Hegel, which you mentioned. Because according to that, we could say this isn't just interesting looking processing errors. We can sometimes actually see negation in what artificial intelligence produces. So that's kind of an interesting field of tension. So the inspiration, I'll look into it. Twitter is the negation. So the controversy in the computer isn't projected from the outside and nothing that the computer did internally. Well, of course it is. But every form has a constant. And every constant has a form. And this goes round and round. But at one point I said that the question is whether something that a computer produces can be random, since it's a deterministic model. We would need something from quantum computers here, because there is a contradiction. Can something grow here beyond systems? I wanted to talk about the intention. I'm a painter. I studied art. That's kind of boring. My intention when I start painting a picture, I think it's super boring actually when it comes out exactly like I intended to. I think it's boring when I paint exactly what I intended to paint. So when I paint, I react to the picture. And with the picture I have to listen to it. I have to take it away. Then there's also an aspect of reality that influences me. So there's multiple elements that come together and I just react to them. And the more I listen actually, the more most interesting the output will be. And sometimes I change all my ways. And still not everything I produce is art. So I'm still selecting from what I produce, which is art and which is not. And so if I had artificial intelligence that would analyze that for me. That would analyze which would be art and which wouldn't be selected from my output. If you would take many artists and you'd do the selection process with many artists and artificial intelligence would learn this from many artists. Maybe we would get places doing that. And for me art, the best definition is science without aim. Which would contradict a little bit what you just said. And there's another question if nature actually produces art that would kind of unify science and art. And maybe we just didn't get it and we just never have gotten it. It's actually just one thing that everything does all at once is art. It's nothing that only humans do, maybe. Well, about the first part what you would like to have for your practical work, that's very advanced. So something that professional artists would like to have. There's a certain problem with that. If I take single artists, then the computer could for example decide what fits that artist. Then I would not have enough data. But if I took all paintings in the entire world then the result would just be reduced to the mean, the taste of the masses and that's probably not what you want to have. A lot is happening in that field and what is working right now is that the style of your images can be transferred to another painting. And then I could start with something and try to imitate that and still something different would be produced as a result. Art is a process and you write about that. Every time I try that it's something different and natural art. And the question whether we have an intention when we produce art is something that really produces difficulties. You cannot just think of that as human stone and fixed forever. I'd like to say that again here. I wanted to say that what will happen to us as humans if that is technologically possible what I presented here. Of course it's okay if we use different ideas of art. But then I will never have to ask myself the question if computer produced art could be art. One aspect that I'd like to pose a question about is the idea of art and how it goes together with intoxication. Basically do they go together? Being high and making art and artificial religions do drugs and get high to make art? It's a question I have not yet asked myself. But I'll try to give an answer. It has to do with intention. Art can be abstract expression but it can also produce deeper emotions, deeper feelings. So I can imagine that drugs enable you to access those feelings. Maybe there is a different connection here that for example certain people are more liable to take drugs but that does not have to do with their artistic output. I don't know and this would have to be researched. The other thing is can computers get high? That's a tough question. Could you for example put in additional background noise into those neural nets to have different results not just from single random things or algorithms but that you introduce artificial noise? I'm asking myself whether the debate around intentionality at the beginning of the last century does it relate to the debate in the beginning of the 21st century? Can you separate the art from the artist and their intention? I'm asking myself why it shouldn't be as possible and can we maybe distance ourselves from intentionality to say artists have their own intentionality for themselves? If you start an opera or a painting then you can't arbitrarily go on creating it because in the process you arrive at obstacles and you deal with them and it is a bit of arbitrariness but it is what regulates art and what takes influence on the process and that artificial intelligence could also do that maybe. Yes, an artificial intelligence could find those rules but then again we would have the problem that art includes the breaking of rules and that happens again and again in the history of art. So I don't think that I have recurred to that debate but I wanted to show that the intentional process influences the production of the work of art and not necessarily the idea of the artist. Hello, I'm sorry, I slept through the talk but I still want to pose a question. So I will assume something. I will assume that the data that the artificial intelligence worked with to make art I will assume that they were ready-made. They were not empirically experienced data so you'd have to empower artificial intelligence to make their own experiences that they can record via sensors and then create something from their own experience and maybe that could be closer to something art like. I mean it's here that I'm something major in artificial intelligence research. I can't really have a spirit in a bowl but if it's about production of art, about thinking, everything cognitive you learn through living. So in effect I'd have to have a robot that moves through the world, that lives in this world but I don't agree with that because a lot of experiences are here, are present in the form of texts or music and why shouldn't a computer be able to learn from that? Why do we have to experience everything? I don't think so. A computer could experience that. But of course as with everything there are a lot of serious people who disagree with me here. There is no interactions with AIs and humans. I'm sorry I didn't really hear the question. So the only interaction here is the interaction of the artificial intelligence with the data and the data is provided by me and again taken from a large cultural pool. So what would happen if an artificial intelligence could start an artistic process and exchange into an exchange with other artificial intelligence forms? It sounds like science fiction but I think we should think about this situation in the present because we as humans have already given up on our monopoly position in a lot of fields and this raises a lot of questions for examples in terms of society, politics, art, finances and those are some major questions for the next years. Yes, thank you very much. We are already two minutes over. So let's ask the question back. I would like to thank you for discussing it so lively. It was extremely exciting and of course we want to thank Simon for the beautiful presentation. Thank you very much.