The game of go, in which the computer program DeepMind beat the champion among humans, created a kind of confusion for Marcus du Sautoy, a mathematician and professor at Oxford University. “I've always compared math to playing go,” he says. And it doesn't have to be a game that is so easy for a computer to play because it requires intuition and creativity. So when du Sautoy saw DeepMind's AlphaGo defeat Lee Sedol, he thought there was a change in artificial intelligence that would affect other creative fields as well.
The scientist decided to investigate the role that AI can play in our attempt to understand creativity and wrote the book The Creativity Code: Art and Innovation in the Age of AI, which was published by Harvard University.
Artificial intelligence and creativity: who wins?
The Verge discussed with du Sauto the different types of creativity, how AI helps people become more creative (instead of replacing them), and the creative areas where AI faces the greatest challenges.
Let's first take a look at what “creativity” is, or artistic creation. In the book, you talk about three types of creativity. What is it and what does it mean for the role of AI?
Many people think that artistic creation is an expression of what it means to be human, and if so, how can AI come close to that? I look at many artists and show that quite a lot of art has a pattern and structure that is very mathematical in nature. This is why I believe that artistic creation may be more about patterns and algorithms than we think, and very often these patterns are hidden. Perhaps the AI can detect this, since it is very good at finding hidden patterns.
There is research creativity that takes the rules of the game and pushes them to the extreme, as Bach did. There is combinatorial creativity where you take two ideas that have nothing to do with each other to see how associations in one can help stimulate new ideas in the other. The third creativity, which for some reason is the most mysterious, is those moments that appear as if from nowhere - something like a change in phases, when you boil water, water turns into steam and the state of matter completely changes.
How does AI fit into these patterns?
Each of these creative approaches offers different challenges for AI. Exploratory creativity seems ideal for a computer because it is capable of much more computation than the human brain. Combinatorial creativity is fun - AI can learn patterns and apply them to new areas. But the most difficult thing for him will be to create something new and break out of the system.
People usually thought like this: “How can AI break the rules? Isn't it stuck in the system because it is programmed to work in a certain way? How can he jump out? " But if the AI says you have to break the rules, that will also be the rule. You have meta code that tells the program to break the underlying code.
In your book, you talk a lot about creative AI projects. Which ones were of particular interest to you?
One of the most interesting was jazz Continuator, which took the music of a jazz musician, studied patterns and started playing on his own. The jazz musician's reaction was striking. He said, “I understand everything I hear. This is my world of music. He plays just like me, except for things that I have never thought of before in my musical world."
So I think this is one of the exciting roles for AI in the future. People often start repeating patterns of behavior. Oddly enough, we become more like cars because we just repeat something, so I'm impressed that jazz Continuator made the musician think a little about his machine behavior. He helped spark his creativity by showing that you could rearrange the ingredients he already had without him even thinking about it. I wanted to show that the role of AI in creativity is perhaps to enhance human creativity, that this is a partnership of the future, that together we can make things more interesting than if we worked separately.
Another interesting story that I think is important relates to the world of fine arts and Google's DeepDream. Google tasked its visual recognition software to look at a random array of pixels and describe what it saw. Through this, we learned a thing or two about how artificial intelligence was programmed and how it saw.
What's the point of that?
One of the problems with modern AI is that many machine learning programs create code, but we don't really understand how it works. The Google DeepDream Project helps us find a way to understand how this happens. Therefore, as for us - people - art is a way to penetrate the consciousness of another person, perhaps the art created by AI will help to penetrate the essence of the work of this code, which is very mysterious.
Take Microsoft's Rembrandt project, which creates AI-generated images in the Rembrandt style. One could say: “Why do we need another Rembrandt? Don't we have fantastic Rembrandts yet? " The bottom line is that all this helps to understand new things in works of art. When we look at Jackson Pollock's work from a mathematical perspective, we see new things that we missed before. So AI can play an interesting role in uncovering new structures that we may have missed in art and now take for granted.
This pattern search is not limited to the fine arts, right?
Well, in the movie world, you can take the Netflix algorithm, which recommends movies that we might like. He can share films in interesting new ways. Some of the groups we could identify as “all comedies together,” but sometimes films are grouped based on how people say “like” and “dislike”, and then the general theme eludes us. It looks like AI has defined a new film genre for which we don't even have a name. You could say that "there is a new fragrance that you need to name." Perhaps AI takes our creative works and sees in them something that we can express, but not realize. He could help us consciously articulate the essence of creativity.
There are many creative areas. Name one where AI has the hardest time?
One of the surprises for me was how difficult it is to write words. Artificial intelligence has so much written to learn. I was quite surprised that even though the AI is pretty good at writing short fiction, it is still unable to write for a long time. It doesn't have a good sense of the storyline, for example. I have not seen anything that would extend the coherent story beyond three pages. It may be very difficult for AI to formulate language constructs as sophisticatedly as we do. Maybe he needs to go through the evolution that we have gone through. And then the question is: how long will it take?