New book explores AI and the secrets of human creativity
Cracking the creativity code
When we look at a great work of art, or are moved to tears by a piece of music, art seems like something deeply, uniquely human. Spiritual, even.
Maybe that's why many of us are uncomfortable with the idea that artificial intelligence could be creative. And yet, increasingly, there are examples of AI producing work that people have trouble distinguishing from art produced by human hearts and hands.
But could AI really go beyond works that can pass a sort of artistic Turing Test to be genuinely creative, producing something truly new and valuable?
Marcus du Sautoy is a Professor of Mathematics at the University of Oxford, a best-selling author, and creator of a hugely popular TED talk on symmetry.
Du Sautoy spoke to Spark host Nora Young about the growth of AI in the arts, and a new model of co-creation.
Over the course of the book you talk to artists as well as curators and music critics. How did they feel about working with AI, or observing what it came up with?
I think the practicing artists are finding this a very exciting period. They see this as a tool, or a collaborator, at the moment and they don't feel a sense of competition. I think in the public at large there's a real dystopia about AI. We have these movies in Hollywood which are kind of making it look as though AI is going to just wipe us out as a species. And I think those who actually practice their art realize that well we're not at that level yet.
One of the interesting examples I have in the book is a jazz musician. There was an AI that trained on the way he played jazz, but then pushed it into new realms and he was like: 'Well I recognize that sound. That's my sound world, but the computer is doing things I've never even thought of doing'. So I think AI can really help us perhaps to behave more creatively as humans again.
When you look at something like that improvising jazz musician, what in that case is AI? Is it the medium? Is it a co-creator?
I think actually we have to get away from this idea of the creative genius as one person creating it all. I quote Brian Eno in the book, and he likes the idea of scenius: that every sort of creative act is actually a result of many people's works and ideas feeding into that. And perhaps that's when it comes to legal ideas of copyright — who we're going to give copyright to in a piece that's co-created by AI, a coder, somebody who's provided the data set. I think we're going to have to have a much broader idea of ownership of creativity.
Is there a type of creative field that seemed to produce better results?
There's been a real breakthrough with machine learning on the visual world. That was always a great hurdle — getting a computer to recognize what was in an image — but that's really changed. So I think the visual world is where actually AI has been hugely successful in the creative realm. Music as well, because music is full of a lot of patterns. So that, I think, is quite amenable for AI.
The place where AI was having a lot of difficulty was the written word. We've got a lot of examples of things which can generate small amounts of text but it very quickly loses the plot. I mean quite literally! There's this example I have where they tried to create a sort of eighth volume of Harry Potter and it's quite fun to start with, but after a while it just isn't coherent. There is where A.I. is having a lot of difficulty creatively.
You're a mathematician. You're also a trumpet player, you're in an experimental theatre group. As you went through this whole journey, how did you feel, just personally, about the role of A.I. in art?
Actually that was the motivation for writing this book, because doing mathematics is a highly creative subject. It's a lot about storytelling, about taking my fellow mathematicians on a journey of surprise, taking them somewhere new and exciting. And I always used the game of Go as my 'protective shield' about why computers couldn't do maths. So when I saw code [AlphaGo] playing this game — which also has a lot of intuition, creativity, pattern recognition — I realized there was this phase change.
So one of the strong themes in the book is looking at my own world of mathematics and seeing 'well maybe that is something incredibly logical, and seems like something a computer could obviously do'. But by the end I'm quite relieved that computers still don't seem to be able to do mathematics in the way that I do it. And I think it is that it's still very difficult to tell stories in a meaningful way. They can generate text but they can't tell stories. They can generate equations, but they can't tell stories with those equations yet. So I'm not feeling too threatened.
This interview has been edited for length and clarity. Click the listen button above to hear the full conversation.