The Art of Computational CreativityShare
An edited transcript of a panel discussion on "Computational Creativity and the Arts" from the Fifth International Conference on Computational Creativity serves as the prompt to a discussion of meta-level issues in the field of computational creativity.
On the first day of the Fifth International Conference on Computational Creativity, in Ljubljana, Slovenia (June 10-13, 2014), there was a panel discussion on the subject “Computational Creativity and the Arts”. The first section of this article reproduces an edited transcript of that discussion. A second section offers reflections on this discussion, and adds some “meta-level” commentary on the art of computational creativity. While it is quite natural – and, it is hoped, thought-provoking – to read a discussion between artists and computer scientists, the second section of the article asks: what would be required for a similar dialogue to take place between creative computer systems? This is a theme that will hopefully be taken up – at least by the human participants – at next year’s ICCC conference, in Park City, Utah.
Panel session on Computational Creativity and the Arts
The participants in this panel discussion, held on June 10th, 2014, were Robert Seidel, Mitu Khandaker-Kokoris, Kazjon Grace, Anna Jordanous, Ed Key, and Mike Cook. The discussion was moderated by Professor Simon Colton.
Robert: I am an experimental film maker. I’ve started to do video installations bringing abstract worlds back into reality. “Abstract,” for me, means creating something that evokes different feelings, but that is not a reproduction of something. I find it somewhat confusing that 90% of artists are painters, when we have so many other ways to express ourselves.
Mitu: I am an indie game developer, but also a PhD student. My best-known game is “Redshirt”, which is a weird social networking simulation. It’s set up as a tongue-in-cheek sci-fi parody. You play a new arrival in a startrek-like space station, where everyone is obsessed with something called Spacebook. You climb the ladder via Spacebook. All the agents behave autonomously, and things will ripple out. This is how I like to think of games, as almost political, even if not overtly. Are games created entirely by computers a holy grail? I don’t know a lot about computational creativity, but I find myself a bit sceptical. I like the idea of using things in this field to understand human creativity. Games can do this too: we can understand ourselves through games. There’s a quote: “We are the mud that stood up” – maybe computationally creative systems can say something like that sometime, too.
Kaz: I came to computational creativity from design cognition. I’ve been attempting to build computational models of the way people do design. My PhD was about “how do we look at a model of analogy making”. The computer generated analogies and interpretations, and the research focused on reinterpreting, and the idea of finding something that wasn’t there at first. So, I’m interested in looking at processes like interpretation and expectation, which are two slightly different ways of looking at the idea of evaluation. Expectation is an observer-centric thing, that is, expectations come in part from the mind of the observer. One has consciously developed expectations that were developed prior to examining the artefact, and now, later, after you’ve looked at the artefact you think something different. This is moving away from computational creativity evaluation being about comparing artefacts to other artefacts, and toward comparing observer responses. So, thinking about the processes of response – this is not always cognitively inspired but it can potentially go in that direction.
Anna: I’m a computational creativity researcher. I originally started with musical creativity and musical improvisation, but about a year into my PhD research I was looking at ways to evaluate how creative the system was and I needed some methodology for evaluating this . Inventing a new evaluation methodology is the best way to either be unpopular or be invited to a panel. Doing the PhD distracted me from music and I went instead toward thinking mostly about evaluation of the systems that we were producing, and looking at how creative are we being. And now that that’s over, I can get back to some of the other ideas! What are some questions to discuss? To what extent does computational creativity help us understand human creativity? To what extent should computational creativity help us to understand human creativity?
Ed: I’m an independent game developer. I made Proteus, which was surprisingly on a screen out there in the atrium. In Proteus, music and landscape are linked; the landscape is generative and music is created according to a set of precomposed samples, and is mixable by rules. Proteus is perhaps interesting because it’s not really a game. Before I worked in it, I was coming from a game development technical background; and one of the things I became interested in is the idea of creative play and the player being one of the co-creators. The game is then a packaged version of “yourself” that is interacting with the player, and new things are being revealed that you didn’t imagine being revealed at first. So, how about players generating stories for themselves? The situation in which the player is both audience and performer.
Mike: I’m very interested in collaboration between people “like X” and other people “like Y.” For example, how can we get researchers interacting with people who are already making games?
[Simon is fielding questions from the Twitter audience.]
Simon: I won’t ask all of you to answer all of the questions. Are these two opposing camps?
Kaz: Opposing camps? No, different but not opposing.
Simon: How about your perspectives on humanity? I’ve read most of the papers for the conference, being the programme chair, and we talk a lot about human painters, human artists. Using the word humans seems to me to dehumanise people. Are we really entitled to treat humans as a different species? We talk about “artists do this, game designers do that” – but actually when people say this, they mean “I’ve done a little bit like this.” So there is no excuse to say things like this without knowing more about what people do when they create. So, a little thing that happened last week – a snippy woman from the US contacted me about one of the works produced by the Painting Fool, and said to me, “we want to use this piece in our sociology textbook.” It was a 150K print run, but I said no, because what they were going to do would have taken this work out of context and used it to get the reader worrying about “being replaced.” Actually I think this is a very bad word choice. When a new game designer comes out out they don’t replace an old one. People either move away or form a collective – it’s all solved amicably. Maybe if software is producing games, etc., there would be fewer people who make them – but people write this sort kind of thing only because of the scare factor. Is Arnold Schwarzenegger coming along and replacing them? Is Picasso saying that we’ll never have a chance to create again? So, I’d ask, if we achieve our goals, will people really be replaced, or will it enhance the field?
Mitu: I generally subscribe to the “technology as extension of ourselves” view. I see what we do with computers as an extension of our own creativity. Every time we create new technology we misunderstand these forces as alien to ourselves. McLuhan talks about how, actually, it is about extending the things we can do. And I think this is absolutely fine, if a bit cyborgian. Actually, neither can exist without the other.
Ed: I’m not sure I know what creativity is anymore, or computational creativity. But I don’t feel worried. It’s always going to be about creativity.
Simon: Funding agencies are very interested in this stuff. Various members of this community were in Luxembourg for big discussions on the future of ICT and creativity, and Tony Veale got up with a passionate speech about what it means to be meta-creative. I had to follow that up with a discussion of generative file types. But meta-creativity is really exciting stuff, you can become a teacher and teach software to be creative on your part. Artists going from creativity to meta-creativity is really a very different discussion from the idea of building “automated artists.”
Mitu: My game, Redshirt, is about being in a post-capitalist society, but you’re still trapped in a world of social capital, of schmoozing with people, and so on. So there are ideas related to this which I tried to express, but there are also procedural, or emergent, game play things, and this emergent stuff can influence the game. There are elements that can engage in accidentally sociopathic ways. For example, one time an an NPC1 dating figure got stolen away by someone else, and their ex started harassing them to get back again; they ultimately relented and then they were in an unhappy relationship. Is this something we’d call creativity? I do think it’s fine for computers to create, or for us to generate through computers – the path of ANGELINA. And I also like to see games as something that express ideas and human conflicts. This can co-exist.
Simon: You mentioned emergence: are we getting to a point where designers will only be happy with a game if it surprises its designer?
Mitu: Well, some people have that feeling and some don’t. It’s definitely important to me. And some people want to make games as a purely personal experience.
Mike: Speaking as a devil’s advocate, I am actually certain that some people will be replaced. What’s interesting about the people that Ian Gouldstone has invited to participate in the art exhibit here is that these are people do this as a form of personal, artistic expression. But in large businesses, if they can replace 500 people or get rid of James Cameron, they would. In the business world, people will be gotten rid of. At the same time, many more new ideas will be planted. And people like James Cameron will find other ways to be creative.
Anna: My brother, who’s a musician, thought I was making a robot that would replace him. He asked me if I was making terminator yet. I told him, I wasn’t making terminator – but other people here might be! And what about all of these people coming out of music college? Rather than being replaced by these new people, he just goes off and practices more, and gets to know them more. As an end result of their being more competition, the good players get better.
Ed: Maybe human creatives can become more human through computational creativity.
Simon: Royal College of Art graduates have something important: humanity. What we’re beginning to realise with computational poetry, is that we actually want to make a human connection.
Kaz: If we enable humans and computers to collaborate, we get some artistic genres that didn’t even exist before. For example, YouTube poop, a new absurdist and particularly silly form of expression. New things are possible with co-created systems, it can be hard to imagine. Michelangelo could not have predicted YouTube!
Audience member: Interesting technology will replace humans to some extent, and it will make humans unemployed. If you don’t want that to happen to you, the question to focus on is which part of what you do is the hardest thing to formalise and computational.
Ed: Consciousness. And we do that without even thinking about it. Also, like Simon said, making a human connection through a work. If something is created by software, who or what are you making a connection with through that work? The idea of the sublime? You are looking at a produced artefact and it is triggering feelings in you – you bounce yourself off of the artefact.
Simon: We are talking about producing things for other people. But how about getting software to consume what was produced?
Ed: Well, software that would debug itself automatically would be nice… although maybe I’m being a bit flippant. But I think it means something for the artefact to have come from another conscious entity. This doesn’t apply to mountains: still you bounce off of mountains in a particular way. Maybe computational creativity’s creations will be more like mountains.
Kaz: So far this discussion is about us as humans reflecting on computationally-created artefacts. Well, what about systems reflecting on their own artefacts? For example, per Donald Schön’s thinking about, e.g., designers – when you’re being creative you’re externalising something, or crafting mental imagery, and you create something related to what you’re thinking about right now, and then you see something that you didn’t put there intentionally, that that creates a further opportunity for creative behaviour. It can change your search trajectory. This idea of a catalyst for transformational creativity is a compelling hard problem that a lot of us could benefit from thinking about. In order to go one step further, I think we should be talking about intentionality – in a preprocessing step, although this doesn’t yet get at reflection in action. Still, changing your intentionality can really be the key thing.
Simon: Haven’t we already done that. For example, with the Painting Fool, we get the software to do portraits, then it looks at what it made and it says, no I didn’t achieve my goal – or I did – and then it shows what it means by that. So I am definitely with you on the reflection process. This could also get us into the topic of computational lying.
Kaz: I wasn’t talking about going back and pretending the goal was different than it was. I mean, even outside of the creative process you engage in intentionality, generative processes, and so on. But I think it needs be coupled with the creative process.
Robert: I think you’re all forgetting one thing, which is that an artist struggles in real life, and the idea I have in the beginning of a project is actually something I can never fulfil. There are many things that can’t be solved. Why do people try to analyse a result without trying to understand how that particular style was developed? For example, why do some people choose to paint in watercolor in the first place? A lot of these comments and questions seem to be coming from a place where everything is settled.
Simon: The standard thing in AI is the problem solving paradigm, but we’ve been rebelling against the idea of “solving the problem of writing a sonata.” People paint to express themselves – or to find out what’s inside themselves.
Audience member: I’d like to say something quite trivial. One place for cooperation between humans and machine is personalisation and reaction to the specific individual. This is something that the human normally cannot do on their own, so it requires an autonomous capability on the part of a creative system. Of course, this doesn’t apply everywhere. I think personalisation is related to Kaz’s earlier points about “expectedness”.
Kaz: I think that personalisation is something that we haven’t quite explored yet, and there are immediate commercial ideas that come to mind, for instance, personalised advertisements, personalised copywriting on websites, and so on. And other things that could be possible with a computational system that we don’t have the capacity to imagine yet.
Ed: Thinking of personalisation, for me as an outsider, the thing that springs to mind is Google Search autocompletion and filter bubbles. You get results tailored for you because you have certain online behaviour. And in this way technology helps us become “more like ourselves,” which I find worrying. It’s not provoking people outside of their prejudices.
Mike: Google’s not interested in doing that, but once you understand what a person is, you can understand how to challenge them. Personalisation of ANGELINA games could happen. But expressing a message is something else again.
Audience member: You have artists creating a variety of artefacts, but the job doesn’t end there. That’s somehow where things begin. Submitting films to festivals can be a full-time job. What do the artists on the panel do to make sure that their works are looked at in the right way? In my experience working within the games community, you have to add a lot of theatre around the actual artefact.
Ed: Things like creating trailers and contacting the press, tweeting the screenshots – maybe an AI agent could be helpful for that. Self-promotion is something you might delegate to a robot who is your biggest fan. I quit my job to work full time on Proteus, but even then 80% of my time was spent doing business stuff, talking to the press, etc.
Audience member: How do you get people to respect the work that you’ve made? The artefact needs to be in a place where it will be respected.
Ed: Sharing screenshots early on will help people get excited, it will create a build up for it. It shows that thing is being formed somewhere.
Mitu: I was lucky to have a publisher who did the marketing and business stuff so that I could ostensibly get on with development. But I still had to do things to get people excited about it. Maintaining a game that you’ve created after you’ve released it, for example. People who play it will have opinions about it and some people will never be happy with the game. People will offer suggestions about things that they want changed. So you have to listen to feedback and think about what’s important. That’s a process that is important. I spent months doing patches and improving certain bits. There were particular aspects that were triggering for someone. This was important, but people didn’t understand why I would respond to that. Games are intrinsically capitalistic products. That is a weird tension that you can’t get rid of.
Ed: That reminds me of point systems, and so on. A lot of reductive games simply about making a number go up, achievements and so on. But it is possible to subvert that in interesting ways.
Kaz: A lot, or most, work in indie games tries to rebel against the idea of games as a Skinner box – this behaviourist reinforcement learning thing, say for rats. Making games that avert or subvert this — means reacting to and engaging with the creative community that exists around your artefact. This could be all kinds of things, sharing your work in beta, making screen shots etc. – and this is the rest of being a creator. Kyle Jennings and Rob Saunders talk about how the values of creators can be grounded in communities, and how different people can play roles of thought leaders, gate keepers, etc. – our computational systems don’t yet engage with the creative community on anywhere near that level. So this is a whole realm of creativity that is, as of yet, closed off. But, say, a Twitter bot might start doing something because other systems are jumping on a bandwagon – I think it’s at that point that we could say that we have an artistic practice.
Simon: Is there a general consensus that says we can achieve a computational evaluation of artefacts that works the same way as our own?
Robert: Never, because there are a lot of things that contribute to your personal view on aesthetics. There are no rules for aesthetics – images won’t fit into all of these categories.
Simon: With aesthetics it changes quickly, you’re never going to write down a set of rules – but I can imagine machine learning look at some rules that apply for the next 10 minutes to achieve a very personalised and local evaluation.
Anna: In answer to that question, tomorrow I will talk about some recent results, where we will attempt to do this exactly, looking at how a community makes aesthetic judgements about something, in the context of electronic music.
Audience member: Philosophers in the 19th Century, like Kant, have made a distinction between mathematics and aesthetics, as something subjective. If computational creativity is going to deal in a satisfactory way with aesthetics, you have to have the computer be subjective. So, my question is, where is subjectiveness in your work? Is it even possible?
Kaz: I don’t think we can algorithmically model aesthetics – I think we should experientially model it! In other words, I think it should be based on experiences of prior stuff.
Audience member: Your brain is doing both kinds of tasks, logical and aesthetic tasks. If you are just a neural network… well, why shouldn’t we be able to do that computationally.
Mike: I have some work on this that wasn’t quite ready to present here, but, yes, aesthetics are going into ANGELINA – but the only premise I have for the system is that ANGELINA should be able to defend aesthetic judgements in an argument with you. ANGELINA might say, I like paintings with a purple pixel. An aesthetic is only convincing if you can defend it! There are more interesting ones that partition the space in more interesting ways. These partitions should look like aesthetics, anyway. Removing randomness and replacing it with subjectivity will happen.
Simon: OK, we’ve got to be wrapping up for now. In 20 minutes there will be 30 minutes of Tony, presenting Robot Comix.
It is possible to reimagine the discussion we’ve just read as a metalogue2, after Gregory Bateson. It is only necessary to ask: what would be required for a conversation like this to have taken place between computer programs? Certainly, natural language is an obstacle, but perhaps there are some suitable work arounds. Is it possible for computers to talk together in any meaningful way about their experiences, their achievements, their concerns, and their ambitions? What assumptions about the nature of dialogue might we have to let go? Simon Colton’s question, “how about getting software to consume what was produced?” seems indicative of the primative state of development of the field of computational creativity at present, but it is a place to start. Already in 1951, Alan Turing suggested that computers should “be able to converse with each other to sharpen their wits” – and other fields, including computer Go3 and argumentation have their own dedicated servers and protocols for exchange. Can computational creativity to move in this direction?
Many points and ideas raised by discussants in the panel point to potential answers to the questions listed above. Robert Seidel’s idea of abstract art as a domain that works with evocative simulacra seems to invite the creation of computational systems that have “feelings” that can be “evoked.” This potentially outlandish-sounding notion was explored by Marvin Minsky at length in his book The emotion machine , and the ideas formed the basis of a realized system developed by Minsky’s final PhD student, Push Singh. Singh’s system, EM-ONE, featured agents with a certain degree of social awareness that interacted in a simplified physical simulation to perform a collaborative task.
Mitu Khandaker-Kokoris’s Redshirt seems to present another example of the sort of social protocol that agents could implement in order to communicate and work, whether individually or collaboratively, towards shared, or competing, goals. Spacebook doesn’t require complete shared understanding; communication is, rather, indicative.
There is a very limited range of things through which to assert an NPC’s personality. Even though I might want a particular NPC to be a really selfish character who doesn’t care about the others and is a bit mean, if for whatever reason the player doesn’t interpret a certain status that way … People behave a certain way but how someone else perceives them is down to their own personality as well.4
Perhaps a system like this could be the work-around we need until we develop better computational language capabilities. A potentially bigger obstacle than language is the “intrinsically capitalistic” – or at least highly individualistic – nature of the reputation game that is academia.
The ability to “defend aesthetic judgements in an argument” mentioned by Mike Cook assumes that discussants are operating in a domain with a shared resource gradient, or some other means by which to judge the value of a given way of partitioning a space . This fitness landscape may itself evolve, if slowly – “like mountains” to adopt a turn of phrase from Ed Key. At this level, features like novelty rise to the fore and value seems to become a secondary concept.
However, even here, following Anna Jordanous’s thinking , the emphasis is no longer on how we evaluate individual computer systems, but how we evaluate the computational creativity research community. When would we be prepared to say, in a suitably reflective mode, that we, as a community, are being “creative”?
These are issues of political importance – and surely not just in Luxembourg. On the one hand, there is the question of making systems that realise Turing’s vision and that are “worth attention to the same sort of extent as the output of a human mind.” Will we be prepared to call these systems people? And on the other, there is the question of the contribution of our field to the betterment of the community that we know, for now, as mankind. It is my contention that we should be brave enough to evaluate our work according to standards like these:
If it does not turn out better, nobler, truer, men and women, – if it does not add to the world’s stock of valuable souls, – if it does not give us a sounder, healthier, more reliable product from this great factory of men – I will have none of it. I shall not try to test your logic, but weigh your results–and that test is the measure of the stature of the fullness of a man.5
NPC stands for Non-Player Character.↩
“A metalogue is a conversation about some problematic subject. This conversation should be such that not only do the participants discuss the problem but the structure of the conversation as a whole is also relevant to the same subject.”↩
Anna Julia Cooper, “A Voice from the South” (p. 283).↩