PassW0rd – 8th July 2020 (AI and Creativity)

Speaker A: This program is brought to you by Resonance FM.

Speaker B: If you like what you hear, please support our work by making a donation at resonancefm.com/donate.

Speaker C: Hello and welcome to Password on Resonance FM. Our monthly exploration of technology in society. I am not a technologist. I am a poet. Yet if I were a computer programmer, I could also be a poet too, and write sonnets like this one, generated by artificial intelligence. It in a circle pallid as it flow by this bright Sun that with his light display rolled from the sands and half the buds of snow, and calmly on him shall enfold away. I’m not a composer either, but if I had enough computational power, I could create concertos. I could even write Beethoven’s 10th Symphony. We all know that the Ninth was the last one he completed, but at the Herbert von Karajan Institute in Salzburg, Austria, Dr. Matthias Reuter and his team are piecing together his notes for the Tenth.

Speaker A: So in our case, the AI has, like a human student, has learned a lot of music and it has seen, you know, thousands of examples of music and it understands how to combine musical notes so that they make sense and that they, you know, create emotions in the listener. And when we give the AI the sketches of Ludwig van Beethoven, then it will take these sketches and continue them in a style that it has learned from these thousands of examples.

Speaker C: Okay. So you are going to use this to finish off Beethoven’s 10th Symphony, or have you done it already?

Speaker A: Well, we’re in the process. We went through many iterations. We already have a couple of versions finished. We will keep working on it until, you know, the last minute before the first performance and hope to present to the world then an exciting piece of music that Sounds like Beethoven but isn’t.

Speaker C: That’s the point, isn’t it? Sounds like but isn’t. Because it’s not Beethoven’s finished symphony, is it? It’s Beethoven’s symphony finished by a computer that, reasoning for want of a better phrase, implies that this is how he would have done it.

Speaker A: Well, yes, that’s right. Although I would say it’s impossible to say how Beethoven would’ve done it. What you can say, however, is this is one plausible way in which he could have done it. And that’s of course a big difference. So no one will be able to say, this is what he would’ve done, but because the AI has seen all of his symphonies, all of his piano sonatas, in fact, a lot of his music, string quartets, and so on and so on, It understands how he did compose, in which style he composed, and this is what the AI can do. So if we give it a melody, in this case, we give it a sketch by Beethoven himself, it will read the sketch and then it will continue the sketch in the style of Beethoven.

Speaker C: It’s such a difficult thing, isn’t it? Because I’m bizarrely, my computer at random played me 5 hours of Beethoven last night and it was his cello. Works. And I was thinking that, well, his Fifth Symphony is not like his Pastoral Symphony. And there’s another piece of work which I’m very fond of, which I think is in his Seventh Symphony. And they all seem to be so different. So in terms of an actual computerized progression of where he would be going, how do you go about sorting that out?

Speaker A: Yeah, that’s a very good question because it implies that there is some progress from work to work in the style itself. And that is definitely true with Beethoven. So with each finished composition, he has enlarged the space of possibilities in his music. And so in addition to that, there’s also the possibility that Beethoven, you know, with a new work would go back to something that is in the past, that is something that he’s referring to in the beginning of his career or music by other composers that came before him. So with Beethoven, there is a lot of possibilities that his music could go to. Nevertheless, there’s always something that is unique to Beethoven., and the machine can learn this from looking at all of his works. So in a way, this is the problem also that any composer is facing when they sit down to write a new piece of music. Do I write something that sounds like the stuff I did before, or do I want to have something completely new? And if so, how do I make it sound like it is my own? So you get to a point where you’re talking very quickly about, personal style, musical identity, and of course innovation. What is possible in music? And Beethoven has done that over and over again, that he pushed the boundary of what is actually even considered to be music.

Speaker C: So it’s a very, very interesting case study for an artificial creative intelligence because yes, one of the things that a lot of people have said about Beethoven is that there’s something almost revolutionary each time that he does something and it is not like anything that he has done before. And I write poetry, and I know that one of the things that I always do is I don’t want things to sound like a piece of work that I’ve written before, because I just don’t— I don’t know why. I just do not want something to just fall into a formulaic pattern where I am knocking something out, for want of a better phrase. So in terms of what Beethoven was trying to achieve then, presumably you can find from what the computer is deriving from the 10th Symphony some ideas, some direction that he was going.

Speaker A: Yes, so essentially we are building our work on the sketch materials. Beethoven was really an avid collector of musical ideas. He had piece of sheet music with him at all times. He would scribble down ideas as they came. Then he had an elaborate system to process these ideas, copy them into bigger books, revisiting them from time to time, and jotting down ideas about structure, form, and so on. And when you look at the materials for the 10th Symphony, it’s very striking. First of all, he’s starting work while he’s actually working on the Ninth Symphony. So often he would compose in pairs, and when he talks about the other symphony, he means the Tenth, and he often indicates ideas like mood or instruments that he might want to use or references to music by others. And you take all of this and it gives you a pretty good sense of the basic direction. And then of course the question becomes, okay, so let’s take this and fill it in with actual music that Beethoven wrote down for the 10th. And this is where the AI comes then into play. So you give the AI the actual music that Beethoven wrote down in the sketches, and it starts to create a thread of music out of that. And it has a very good sense now of the overall form that it should create, and the stylistic idiosyncrasies. So it’s quite interesting to see it work. There’s of course a lot of feedback that we give it. So you cannot think about it as you press a button and out comes a symphony. It’s rather, you know, you are in a relationship with the AI as a human. You give it feedback, you say, well, this worked good, this worked not so well, and then it progresses and it becomes better. So it is in fact a process in which there is human creativity and there is machine creativity, and the combination is really very, very interesting.

Speaker C: So how is this being received by Beethoven scholars? Presumably some of them have listened to it and helped you correct it.

Speaker A: Yeah, so we work very closely with experts from the Beethovenhaus, which is the museum and the research facility in the place that Beethoven was born. We work together with experts from the scientific community, and of course everyone is excited to be, you know, part of this huge challenge. Something like this has, on this scale, has never really been done. And so there is a lot of excitement. There is of course also a lot of skepticism at first, you know, First of all, is it possible? But what happens with the human genius? Why are we doing this? And so I think this project really goes to the core of some of the issues surrounding what it means to be human in an age of artificial, of machine intelligence.

Speaker C: That was one of the things I was going to ask you. And obviously, I assume you’re learning something about the way that Beethoven composed, but I assume also that you’re learning something too about the relationship with machines, this augmentary relationship that almost in a sense is taking us forward into the 21st century and what humanity will be doing in future with the machines is a partnership in a sense.

Speaker A: Yeah, absolutely. I think this project is about democratizing creativity also. When you think about the work and time and effort that has to go into one’s life, if you want to write a piece of music, you know, or write a poem like you do, then in the end, these are countless hours and hours and hours and days and years of studying before you can even begin to express yourself in music. Now, if you have a musical intelligence with your keyboard or any other kind of instrument that you play that can help you to become expressive and create something emotionally appealing for other humans without all of that work, then that allows thousands of people worldwide to actually express themselves. And I think this aspect of democratizing creativity is very powerful because You have that in music and in art, but you can also apply it in all other domains of human knowledge and think about what that can mean for innovation, progress in our world.

Speaker C: It’s music and the first 13 seconds of Beethoven. But is the rest real creativity? Some philosophers agree with John Searle, the disgraced former emeritus professor at Berkeley in California. He argued that if a computer ingests enough data, it can rearrange that data into new patterns, but that this is not creativity. Rather, it’s an extension of what Alan Turing, the 20th-century father of British computing, called the imitation game. John Searle maintained that if he, who does not understand a word of the Mandarin Chinese language, were put into a room containing enough phrasebooks and dictionaries, he would be able to answer questions written in Chinese by simply looking up the correct written symbols, even though he understood neither the questions nor the answers. This is one way to approach the question. Can AI be creative? By the way, Searle is no longer an emeritus professor because he infringed the university’s guidelines on sexual harassment. His Chinese Room argument echoes the view of Ada Lovelace, the 19th-century inventor of computer programming. Lady Lovelace wrote about her work with Charles Babbage in the calculating machine called the Analytical Engine. The engine has no pretensions to originate anything. It can do whatever we know how to order it to perform. Only when computers originate things should they be believed to have minds. But in the 21st century, researchers have taken up the challenge and tried to establish a test for creativity. They call it Lovelace 2.0. Professor Matthew Guzdial of the University of Alberta, Canada, explained how it works to Passwords Jane Wyatt.

Speaker D: Simon Colton and Geraint A. Wiggins are two of the sort of leading founding members of this computational creativity field, and they defined computational creativity as the philosophy, science, and engineering of computational systems which, by taking on particular responsibilities, exhibit behaviors that unbiased observers would deem to be creative. So that was their initial definition. And that sort of bakes into it, right, a definition for how we’d figure out if something is creative or not. If some AI has produced something creative, a quote-unquote unbiased observer just says so, whether or not it is creative. But there’s a little bit of a problem here, which is that if I produce something, if I’m a creative person, and I’m not saying that I am, but if I’m a creative person and I make a work of art, but then I tell you, oh, actually an AI did it, and you deem that to be creative, then, you know, great, I’ve passed the test, right? And obviously that’s not great, but there gets to be this weird gray area, which is if I tell the AI everything it needs to produce the creative work that I did, and it produces that creative work on its own, quote unquote, but with all the information given by me, is that still creative? And this becomes a much harder thing to argue about in terms of when we get into specifics like this. And that’s part of why Dr. Marko Rudel produced this Lovelace 2.0 test framework the way he did. Having a human user have to give the constraints on the AI, which sidesteps the ability of the AI’s developer just saying, “Oh, you know, just produce things like this. This is good enough,” or, “This is creative.” We’re trying to get the human author of the AI, not the human creators of creative things, but the human author of the AI out of this question., right? We’re not trying to measure the human AI author’s creativity. We’re trying to measure the AI’s creativity, which is sort of how we got to the Lovelace 2.0 test.

Speaker C: As Matthew Gustiel says, true creativity should be more than just a rehash of data that’s been fed into the machine. An early episode of the TV series Star Trek showed us how difficult that can be. “None of these Da Vinci paintings has ever been catalogued or reproduced. They are unknown works, all apparently authentic to the last brushstroke and use of materials. As undiscovered Da Vincis, they would be priceless.” Would be? You mean you think they’re fake? Science fiction like Star Trek is awash with fanciful forms of machine art, and computer-generated imagery can bring them to life on the screen. But traditional sonnets and symphonies are governed by rules and have a mathematical structure. Machine learning systems, such as generative adversarial networks or GANs, can train AIs to recognize their patterns and copy them. So what about painting? How can an AI replicate the splash of paint, the pressure on the brush, or the artist’s abstract Liat Graver in Berlin and her robot painter E-David are working on it.

Speaker B: So there is no interface that one can buy. It’s all custom made and built together with the collaboration with the computer engineers from the University of Konstanz and other collaboration that I’m working with. Part of them are companies as well, other universities. The interface is we have a visual feedback. It means that there is, after certain iteration of painting, the computer takes a picture of the situation, then we can kind of play how the computer interprets what happened, how the computer understands, how we extract this visual information into the creation of new one. Are we just interested in copying it in a pointillistic way, like point by point, like pixels? Or we want to try to teach the computer or to teach the system to see lines, to see area of paint, to see direction, to see expressivity, how we decode it and encode it again for the creation of the new strokes that the machine needs to operate according to the information that has been taken down and regenerated. So the interface itself right now is based a lot on basically code that we are writing and rewriting and retrying in different ways. So the visual feedback is one element, and the other will be using different kind of mathematical concepts that one can directly visualize. What usually we know today very well from a lot of like art-science collaborations, it’s kind of the emerging new aesthetics. But to take it and to instead of having this like really clean, high images that come out of it or visual information, how one would call it. I think this entire concept of what in German called the Bildsprache that was being discussed over the 20th century now comes to really a new understanding of how we analyze information and then re-execute it using a machine that has a different way of operating. And I would say like, thinking is not the right word, but like creating or executing actions that we will have, and we attribute it to our wishes.

Speaker C: Okay, so what you’re saying is that what’s happening is that the robot, David, is actually painting, and you’re trying to take various bits of information about how you do things and put that into David.

Speaker B: Exactly, there is kind of a back and forth situation. There is a lot of things that of course I cannot control, like when I’m doing my own painting or working in my own studio or even in front of a graphic program, there is a level of control, but we always adapt our way of working and our way of thinking according to the tools that we have. And our entire history is basically rewritten or created or been, or shaped by the tools that we have in our hands. So what the working with specifically with computer technology, and in this case as well, I give it more freedom for computer to make decisions for me and working with a machine that operates at a level that of course the simulation in the computer is very clear what comes out, but between the simulation and the real execution, things are changing, they’re going into different direction because of the nature of the mediums that we’re working with. This is paint, it drips, it’s not very clean. A new world of possibilities opening up, and that allows me as a creative person, as an artist, to kind of reconsider the decision that I’m making, why I’m choosing to paint the way I’m choosing to paint, and to really think as well on conditions of aesthetics that are being taught to me, for example. I am coming originally from Israel, so I grew up in the Middle East from like an Iraqi family, but I am now living in Europe because the Western aesthetic is so pronounced, it’s the status quo of how we create art today. And suddenly I’m getting to the point when you work with computer and you need to tell it and you need to teach it There is a room to explore new possibility of aesthetics. I’m not obliged to follow or I’m not dictated to follow by everything I’m being taught through my years in art academies, for example.

Speaker C: So what you’re saying is that you’re trying to develop a new— So where’s the piece of art in all of this then? Is the piece of art in this collage of experience and research and all of these other things? That you put together, where essentially the finished piece, the work of art, is just an expression of all that’s gone on. Where is the art?

Speaker B: Well, I think this is a very separate discussion about what art is or supposed to be or should be. This can go into a huge political and social, say like, takes political and social turns. For me, and I think it’s a very individual understanding of how I see it. The art is the experience that I’m doing and my ability to translate, to transmit this experience. And this is like multi-layer understanding of the world to the audience, to the viewer. So it could be that, you know, in a very 2D normal painting hanging on the wall, this entire information will be encapsulated. And sometimes I’m taking a different approach of doing an entire space installation when you have many robotic paintings with combination of simulated computer program that shows the process as well, or with video, or with light. There are so many different tools that one can create and add to it in order to kind of communicate this entire process. So I would not try to limit myself and say the work of art is the process, or it is a painting, or it is a space installation. I think in each project that I’m doing, I have a different goal, and hopefully the artistic intention is meant to go through to my audience.

Speaker C: The robot’s called David. I assume that that’s a nod to Michelangelo. What would you be doing if you didn’t have David? What would you be doing now if David wasn’t around?

Speaker B: First of all, David is not really just around because I am living in Berlin and David is located in the south part of Germany, so almost like 900 kilometers away from me. I am collaborating with different kinds of different institutions. I have to as well say that I am not trained in computer science. I am not trained in the technical domain. And my work is built upon a very tight collaboration with people that are coming from the technical background. So a huge element is going in this creation with the interaction, like the eDavid is not only about the robot, it’s about the entire department that stands behind it. It’s about the people that I’m working with. So even if I’m not there and I’m not directly working with the machine, I’m still exchanging emails or having conversation, how one should go about it. And of course, one can as well build a very like low— like this is a specifically like the IDA with the industrial robot, very like big and powerful, but there is no reason why cannot do it like with low-tech elements and build something very small at home or work with kind of between simulated and real in different kind of levels. So using different elements of technology that reflect or correspond to our contemporary matters of artificial intelligence or working with different kind of machineries as part of extension to our own body as we are living today.

Speaker C: You talk about this technology being augmentary, you talk about it freeing us, it emancipates us. Does this mean that somebody who would not have been what we would have considered traditionally as an artist can now be an artist? Does that mean that they can be augmented by the technology and become an artist?

Speaker B: I think that has nothing to do with the technology. I think it has more to do with us as a society and how we title who is an artist, who is not an artist, what is a traditional artist or not. I do think that it’s allowed for people to express themselves in different ways that before were not. And if I would maybe narrow down the concept of an artist, and let’s say specifically now working with the e-David as a painting robot, just to reflect on the concept, who is a painter? Am I a painter or was the machine the painter? Because the machine physically made the line on the canvas, or maybe the painter is actually the computer scientist who built the entire very complex architecture of information and code that created this work. So I think it opened up like not only the question of like, okay, who can be an artist, it’s how we understand how we title those people and what it means to us. I definitely think it’s allowing or giving more power to people that maybe would not be able to do so in different ways.

Speaker C: Talking to Liat Greather, I realized that what’s really creative here is not that a robot can make a painting, but that it is in itself a blank canvas. Until it’s been trained on some datasets, it has no preconceptions about beauty, truth, perspective, or any of the other painterly traditions. Its lack of creativity can become a new version of what we think of as art, a new aesthetic. The quality of the datasets, of course, is crucial, and ironically, if they are too clean, they cannot be creative. At the Zurich headquarters of the computer giant IBM in Switzerland, Stanislas Božniak is excited about a new development called spiking neural units. It brings together two rival schools of thought that diverged in the 1940s. One is classical artificial intelligence. The other relies on system spikes, like jolts from electric shocks, which produce actions and reactions in the human brain or in a neural network of machines. Dr. Božniak’s team have synthesized them and trained them to predict and create, for example, the next chords in a piece of music by Johann Sebastian Bach. Does this mean the spiking neural units are creative?

Speaker E: This is a very difficult question, and I don’t think that there is a very good answer to this because Defining a genuinely new artefact is a philosophical question. For instance, if you define a previously unseen combination of points on a canvas to be a creative picture, then already a simple mathematical random generator could be considered creative.

Speaker C: Okay, behind you at the moment there’s a picture on your Skype profile picture, there’s a picture of Van Gogh’s sunflowers. But when Van Gogh painted the sunflowers, he painted those because he saw sunflowers completely differently to anybody else. Surely when an AI system looks at a sunflower, it sees the same sunflower that another AI system sees. Well, not that the AI system sees anyway. There is a photograph that is taken that is exactly the same as another photograph that another AI system would look at and the AI system would say that’s a sunflower. They wouldn’t look at it in that different way that Van Gogh did to come up with his view of a sunflower.

Speaker E: How are you going to overcome that? There is a way how you can make the same network look differently on the same picture. And I go back again to the generative adversarial networks where part of the network input is noise. And this noise, you can put it along with your standard input. So you can imagine, okay, you have your standard inputs with the vision that you see a particular picture and the same network can receive different noise inputs. And these noise inputs can alter the way it interprets the picture.

Speaker C: So sorry, when you, and when you say noise, what you just mean is meaningless data, extraneous data. Things that the AI system has to filter out?

Speaker D: Yes.

Speaker E: So typically we consider noise as something that is not really useful and something that has to be removed. But noise has been shown to be very useful for some particular applications. And for example, these generative adversarial networks, they really, really rely on this noise. The noise, I could maybe somehow compare this to some touch of creativity because you’re mentioning this, how we could make these systems creative. So actually, if you have a fixed image, the computers are deterministic. So they would always see in the same way. But the AI systems can be designed so that they take the noise in as the input. And this noise actually impacts the way they see things. So in this way, they kind of move them away from the standard way of looking into things, and then the noise is useful. The question is also how you generate a very good noise. This seems trivial, okay, I need to have some noise coming into the system, but this is also a very profound scientific question. How random is your noise? Is it really truly random?

Speaker C: If you’re putting randomization into something, then somebody would say, well, that’s not random because you’re trying to encode randomness. So in the act of encoding randomness, you’re actually being unrandom. And the other, the other point that it’s making is, really, isn’t an AI system just copying human creativity? Is it really genuinely being creative? It doesn’t, it doesn’t think it’s being creative, does it?

Speaker E: Because it doesn’t think. AI, as you pointed out, relies on what it has seen somehow. It reflects the data that it has been trained on. So in this way, even if you see something novel, as we discussed, a new recombination of the facts, these were the facts that this network has to first acquire from somewhere. So in that sense, yes, it is a very deep question whether this is really creative. So if you’re a writer, you are recombining the letters, and the letters were invented by someone. The words were also given a priori to you, so actually you have also internalized these words through a process of learning. Then you also, when you’re doing the writing, taking some ideas from the past and very often you are recombining them. So in that sense, some people could say, yes, the AI we have right now has somehow this notion of creativity already realized in what it can produce. But on the other hand, as I said, this is a very deep philosophical question. So unfortunately, I don’t think I’m able to answer this.

Speaker C: Oh, hang on a moment. I have a robot vacuum cleaner and it’s just turned itself on.

Speaker E: I also have a robot vacuum cleaner.

Speaker C: They’re very cool. Well, I quite like mine, but it does sort of seem to— it’s one of those stupid things, isn’t it? You always say, you know, I nearly said it has a mind of its own. It doesn’t have a mind of its own. That was a stupid thing to say. Now, you talk about using AI systems to create music and that people need music. Surely a lot of people who make music say, why are you doing that? Because we make money out of making music. And if you get machines to make music, then we’re not going to be making money any longer.

Speaker E: I think that this is a very common worry that is raised around AI, that it might impact the jobs and might take some of the things that people used to do themselves. So first of all, about generating the music, in our work, we have presented a network that predicts Bach chords. So it tries to deviate as little as possible from the style of Bach. Therefore, if you would be a music fan, I don’t think that that would be really appealing to you because you would like to experience something completely new and creative. Then you would have to explore more the use of these generative adversarial networks and the noise, that we discussed previously to somehow put some more creativity into what is generated by these systems. So that’s the answer to the music that we have shown that we are able to predict, but predicting, as I said, doesn’t mean that you would really enjoy that this is a new piece of Bach. So I think that the musicians are quite safe there.

Speaker C: Dr. Stanislav Bozniak. This is Password on Resonance FM with me, Peter Warren, and after this, keep listening for DJ Ritu with A World in London. Earlier, we heard Professor Matthew Guzdial explaining how the Lovelace 2.0 test was designed to establish whether or not an artificial intelligence can create original new works of art. Guzdial’s own medium is video games, and he’s happy to have a robotic assistant on his workbench.

Speaker D: Before I can talk about that, I need to introduce something else, which is a kind of creative cognitive process called combinational creativity. It’s also sometimes called combinatorial creativity in the psychology field. The idea here is that this is the process that you or I do when we make something new by recombining old pieces of knowledge. The easy example here is if I ask you to come up with a new animal. Right? If I say, “Come up with a new animal,” you will almost always take pieces of an old animal and put them together or say, “Oh, it’s like a dog with wings,” right? That is combinational creativity. You’re combining old things to make something new. So what we’re doing with that and deep neural networks and deep learning is we’re looking at how we can use a computational process. So basically empower the computer to do this kind of cognitive process, recombine to be able to come up with new kinds of deep neural networks. So for example, let’s say that we have a deep neural network that knows about dogs and cats, but what we want is a deep neural network that can know about foxes. We can use the knowledge it has about dogs and cats and this sort of combinational creativity to recombine that knowledge to approximate, to get us something like the knowledge of a fox. And did it work? Oh yeah, it definitely works.

Speaker B: How could that be applied in the real world?

Speaker D: One of the ways that we found that it’s really effective is in so-called low-training data problems. So low-training data problems are problems where you don’t have a lot of knowledge, right? You don’t have a lot of data to train the machine learning system on. And these are very common. It’s more common to not have knowledge than it is to have knowledge. And we’ve found that what you can use this process for is recombine old knowledge you have in some entirely unrelated domain to allow you to approximate this, this case that you only have a little bit of knowledge for. So this could be helpful in all kinds of places. One of the places that we’re exploring right now is in automated personalization. So we have this deep neural network that can learn to do something working with a person, and then we have some new knowledge about how it could work with you, but we don’t have a lot. Typically we’d need something on the order of, you know, hundreds of thousands of data points, and a single person isn’t going to produce that many. But we can use all of our old knowledge of working with everyone else to approximate how this deep neural neural network could work with you.

Speaker B: What kind of stage are you at?

Speaker D: We have a few experimental sort of tools that we’re working on right now, some of which we’ve published in the past. So at CHI, which is the biggest human-computer interaction conference, so a place where scientists who are interested in humans and computers working together meet, we demonstrated a co-creative level design tool. So a level design tool where a human and an AI work together to be able to create a level from scratch. And through the process of working with you, the AI actually learns to adapt to your specific video game level design style, even if you didn’t know you had one.

Speaker B: So in a sense here, it’s a little bit like a musician and an instrument?

Speaker D: So we’re a little early in this sort of co-creative field. And as I’ve used this metaphor in the past, but we’re not sure right now if the AI— say if you’re a chef, we don’t know if the AI should go in your cooking utensil or if it should be a robot chef helping you. We don’t know what’s the best way to support human creativity. That’s part of the research happening right now. As it is right now, I’ve been leaning towards the AI as being a partner, but even then, people people perceive the AI differently. So in this level design co-creative tool, for example, we had 12 industry level designers use the tool. They each saw the AI as taking a different role with them. And this came actually with the name of the paper, which was “Friend, Collaborator, Student, or Manager.” So the users of the tool, these level designers from industry, either saw the AI as being a friend, so someone they were just sort of messing around with, a collaborator, sort of an equal partner that would equally do as they did, a student, so someone who should just sort of do whatever they wanted. That’s what students do, right? And then some even wanted it to be a manager where it was sort of giving them feedback on their level design process.

Speaker B: And was it all of these depending on the individual, or did one kind of model for the relationship win out?

Speaker D: The key here is that the AI we designed could only act in certain ways. It learned from the human user, but it never learned, say, to— it was never able to give feedback to the human on how they were doing. It was never able to have some notion of fun, right? As if it was a friend messing around with you.

Speaker C: With you.

Speaker D: These are things that we didn’t author into the AI. We didn’t give the AI the ability to do these things, but people still perceive these things in the AI. And this has led to future research we have in terms of, okay, how can we take advantage of the fact that people are coming to this AI with different expectations in terms of how it should act?

Speaker C: In the multimillion-dollar video games The commercial benefits of working with AI are clear. But that makes me wonder why on earth today’s computer experts should want to replicate Shakespearean sonnets. There’s no money in poetry, as I know only too well. Yet Dr. J. Han Lo at the University of Melbourne, Australia, is a member of the international team behind the sonnet-writing Deep Spear experiment. Testing whether the computer-generated poetry could fool people into thinking it was written by a human scribe.

Speaker F: Essentially, it’s just a game. We basically pitch two poems to a user. These are crowd users that we recruit online, and then we ask them to— and one of them is a human-written poem and one of them is generated by DeepSpeech— and we just ask them to guess which one is written by the human user. That’s all. And the idea is if the human users are able to guess consistently the correct poem, then they’re That means the poetry written by the machine is distinguishable from the human poetry. But what we found, that was not the case. People were basically getting accuracy of like 50%, essentially just a pure guess. So this sort of reinforces the belief that at least the poetry that the machine, that DeepSpeech, has generated is kind of like human, and crowd users, lay users, are not able to distinguish it from from human poetry. Now, that was good and that was all happy and that was one of the reasons why we can publish it, because the results are good. But we also did another, a second assessment where we asked a literature expert. So one of our co-authors is an English professor at University of Toronto. And so he was involved very early on because he’s giving a lot of advice when it comes to language and all that. So we got him to basically look at these poems.. And when he was looking at the poems, again, he doesn’t know the source of the poem, whether it’s generated by human or generated by DeepSpeech. And then we just ask him to assess a few things like the fluency of the poetry, the emotional impact of the poetry, the pentameter, and the rhyme, basically. And the funny thing that we found is that the expert is able to tell which poems are generated by the machine, because he thought actually even though the poetry is quite fluent, but it lacks the emotional impact, it lacks the narrative, the coherent interesting storyline that human poetry has. So to him, he thought it was very clear as a literature expert which ones are fake.

Speaker B: And as researchers, what have you gained from all this? How does it help us to understand the role of AI in our lives at present and in the future?

Speaker F: It was really funny because I did this work when I was still at IBM as a research engineer before I moved back to academia. And when I first sold this idea to one of my managers and said, I wanna do, you know, creative poetry generation. And you know, I mean, they gave us a lot of freedom to do research. It’s a great environment, but they really struggled to understand what is the commercial value of a poetry generator. And I think I struggled a lot to try to sell them how this is gonna work. Profit IBM. And I don’t think we ever convinced them that it’s going to be profitable. But for me, it’s largely scientific why we do these things. We just really want to understand whether machine can produce a creative storyline essentially by— because machines are very good at pattern recognition, okay? You can get them to do speech recognition, you can get them to classify images, they can do all these pattern recognition thing. But the curious question is, can they actually, after reading, say, lots of books, create a completely new storyline that is coherent? It’s a very difficult problem, but that’s something that we are starting to scratch the surface. That’s all I can say when it comes to why we are doing this. It gives us a window to understand potentially how human cognition works when it comes to creative writing. Right.

Speaker B: Is there one piece of work by DeepSphere that you’re particularly fond of or pleased with?

Speaker F: I mean, I have a couple favourites. I mean, but I have to make a very clear disclaimer: I’m not a literature expert. I find it very interesting. It may not be very good quality if you’re someone who reads a lot of literature. So there’s one that was quoted quite a lot by newspapers, and then there’s a second one that I quite like. So here is this 4-liners that I quite like. Shall I behold him in his cloudy state? For just but tempteth me to stop and pray. A cry, if it will drag me, find no way. From pardon to him who will stand and wait.

Speaker B: And that’s it.

Speaker F: I don’t know if it makes sense, but when I read it, I just thought it was something that created a story for me.

Speaker C: Jehan Lowe reading from Deepspear. And if an AI can write poems, why not a children’s book? Max Deutsch has programmed his algorithm to produce an extra chapter in the Harry Potter series by J.K. Rowling. Here’s an excerpt: Harry was watching him. He looked like Madame Maxime when she strode up the wrong staircase to visit himself. I’m afraid I’ve definitely been suspended from power. No chance indeed, said Snape. He put his head back behind them in red groups as they crossed a corner and fluttered down into their ink lamp and picked up his spoon. The doorbell rang. It was a lot cleaner down in London. Harry collected fingers once more with Malfoy. Why didn’t she never tell me? She vanished, and then Ron, Harry noticed, was nearly right. Now be off, said Sirius. I can’t trace a new voice. He punished Uncle Vernon so loudly that she could barely use it with Victor Krum. These faces in one side of their previous year. Mr. Dursley again was his champion. He was about to give the Remembrall there. Every time he picked it up, the marble staircase toward Fred, who was rolling with pain. Stinking cowardly scummy thing too calm too terrible. Mumma covered in weird ink. Well, I am not a great Harry Potter fan, so I may be mistaken, but that sounds like utter garbage to me. J.K. Rowling need not worry that a computer programmer is going to steal her characters and make a half-decent 8th book in the Hogwarts saga. Still, the question of who owns the rights in AI-generated creative works is an interesting one. Cue Jane Lambert, a London barrister and expert on intellectual property law. She has a clear legal response to the question: can AI be creative? No, that’s their answer. The work is owned by the human who set up the arrangements of the creation according to the Copyright, Designs and Patents Act 1988, which came into force in August 1989. The United Nations is currently holding a global consultation on the question, but that’s the current state of the law. The AI is just a tool in the hand of the creator. If we consider its work is original and has artistic merit, that’s a symptom of Eliza syndrome, the behavior humans exhibit when they want to believe that a machine is capable of emotion or even empathy. It’s named after an early artificial intelligence chatbot that was built to help psychiatric patients by listening to their fears and obsessions and providing pre-programmed responses.

Speaker G: Advances.

Speaker C: A more healthy attitude to creative technology comes in the form of a kit for making things at home. During the 2020 coronavirus lockdown, there was a new wave of interest in handicrafts. It came with furlough, unemployment, and a desperate need to create new home-based products and business startups. That’s propelling the Maku multi-purpose manufacturing gadget to a new a new niche market. Co-founder Alex Smolenski.

Speaker G: There’s a famous quote, “We shape our tools and thereafter our tools shape us.” And so we can’t escape the effects, the intended effects and the unintended effects of the things that we build. And so as technology that enables us to express ourselves in different ways emerges, like AI, like 3D printers, like computers, everything, then we’d start doing different things with it. Those different things express themselves in culture and then shape the rest of us and the environments around us that then shape the tools, and it’s kind of— it feeds into itself. And so as it becomes possible to make things locally, to make things that are as advanced as the things that you can build in a modern factory from a— from home, that democratization will lead to unimaginable ends. It’s what’s happened with music production. It’s what happened with photography. It’s what happened with many industries that have been made, the production of which has been made more accessible to independent creators. It really, this project was born, my business partner Benjamin Redford, he initially had this idea when he was at university, when he was reading a lot of Karl Marx, and it was this, this idea about seizing the means of production. You know, if you have access to the means of production, you have an inherent power, and that power of making is something that we want to bring to more people.

Speaker C: Cool. So essentially what we’re saying is, in the 21st century, Everything is going to be challenged. So the way that we’ve worked in factories— nobody wanted to work in a factory anyway— that’s being challenged. Robots are now going down the mines. We didn’t want to go down the mines anyway. So what we’re saying is, no, don’t be scared of the future. Embrace the future because it’s making your world better. It’s going to be a better place because we’re going to have choices in lots of different ways. Those choices that we’re being faced with are the ones that we really want. We like to be creative.

Speaker G: Yeah, and I don’t think it’s all, you know, I think there’s a sort of a techno-optimism that needs to be taken with a pinch of salt in the sense that when new tools emerge, they don’t just have positive consequences. So we can’t really control that and we need to embrace change as it’s happening. And steer it towards a way of living which is in relation to the environment around us. And I think part of that, if we have a chance of living more sustainably, part of that will be reducing the number of miles that things travel before they’re used. And local production is part of that, whether giving people the power to create things things is purely positive is only yet to be seen, but I do believe that overall it’s a net benefit to society to give everyone access to the tools needed to create the things in the world around us.

Speaker C: Yeah, I mean, I suppose one must acknowledge that things have happened like you can 3D print a gun from the internet, so, and that might not be a good thing.

Speaker G: Exactly, yeah, if you can make anything anywhere. Everybody has access to that. And so that has to be— it’s kind of then you start getting into all sorts of questions of moral philosophy about which I probably am not the right person to comment on about how to think about the benefits and the negative aspects of that to culture and society. But I do believe in access to tools., and I don’t believe in walled gardens that are only made accessible to large corporations. I do think that if you give creative people tools to build things, that overall that’s a positive force.

Speaker C: We’ll have to wait and see about that and come back to it in a future edition of Password. This one was produced by Blue Buffery and Jane Wyatt wrote the script. We’ll be back next September with more musings. Meanwhile, you can hear Beethoven’s 10th on the Deutsche Telekom website, and E. David’s paintings will be displayed in an exhibition in Berlin in September. Maybe I am suffering from Eliza syndrome, but I do believe some computer-generated artworks do have the power to stir human emotions and even move us to tears. So I’m not going to say goodbye, but handing over that role to the time-expired replicant Roy Batty from the Blade Runner film, played by the late Rutger Hauer. I’ve seen things good people wouldn’t believe.

Speaker B: Attack ships on fire off the shoulder of Orion.

Speaker C: I watched sea beams glitter in the darkness at 10 hours a day. All those moments will be lost in time like tears in Right.

Speaker B: Time to die. This program has been brought to you by Resonance FM.

Speaker A: If you like what you heard, please support our work by making a donation at resonancefm.com/donate.