The Creativity Code: Art and Innovation in the Age of AI

"I should add a disclaimer, which is that I know Marcus relatively well. I used to work for his outreach group in Oxford and he’s been a bit of a mentor for me, so I’m slightly biased. Before I read The Creativity Code , I didn’t know much about the current state of AI. This was a bit of a shortfall as it’s a big topic at the moment. It’s increasingly coming to the fore and Marcus deals with it in a really interesting way. He takes us through lots of different areas where AI is being used to mimic creativity, which some people would argue is the essence of being human. It’s interesting to see how close to being human AI is getting. He goes through the notable successes, but he also talks about the failures. The failures are often very funny. In literature in particular, AI is not making great strides. There’s an AI-authored Harry Potter book and the title is Harry Potter and the Portrait of What Looked Like a Large Pile of Ash . It’s clearly bonkers, but it’s quite close to a real Harry Potter title. There’s the idea of the ‘uncanny valley’, where things get so close to being human that they actually feel much further away. They’re so close but there’s something that’s quite clearly wrong and it freaks us out. AI is getting to that stage in some areas. But in other areas, it has completely surpassed humans. Definitely not Shakespeare just yet, but Marcus provides some evidence that AI is producing symphonies that can fool music critics. When they’re given blind tests, the critics say that they like the AI more than the real thing, but then when they’re told, they dramatically reverse their opinions and are freaked out by it. I actually looked up some of the images that AI has produced—the Rembrandt-esque ones are actually pretty convincing. I’m not an art expert, but they look Rembrandt-esque to me. So in those areas of creativity, it may be getting closer. But the areas where AI is perhaps making the biggest strides is in rule-based subjects like the games of chess and Go, where deep learning has become a hot topic. Previously, machines like Deep Blue—which was the first machine to beat a chess grandmaster—were written in a style of coding which meant the machine considered all the possible options. It won by being able to do that very, very quickly. “The areas where AI is perhaps making the biggest strides is in rule-based subjects” But the deep learning that AlphaGo—the machine that learned to play Go—implemented is very different. It plays many games against itself, or against other computers, and learns from them. And so it is able to do things that weren’t explicitly programmed into it. There’s this nice part in the book where Marcus talks about a battle between a Go Grandmaster and the AlphaGo machine. The AlphaGo program puts a tile down somewhere where no human would ever put one at that stage in the game. It would be suicidal. But then later on, as the game progresses, it turns out to be a completely inspired move. So it’s learning and creating things that humans, who have been playing this game for millennia, never came up with. It’s actually surpassing humans in that respect. That’s the area quite related to games, because math is like playing a big game. You have these axioms and these laws, and you have to play the game according to those rules. That’s why mathematics is potentially so susceptible to AI overtaking it, because you can tell AI these rules and just let it let it go and see what it comes up with. There are now more and more computer-aided mathematical proofs. Computers can produce these hundreds and thousands of page-long proofs, which prove things that mathematicians have proved in the past. It’s not done in a particularly elegant way, or in a way that a human would perhaps be proud of, but nevertheless they’re able to do it just by making logical deductions. It’s what we were talking about before: in math you don’t need to use inductive reasoning you just have to start from the rules and build up from there. “To some extent, a mathematical proof is just about whether you can convince other mathematicians that what you’ve done is correct” These AI proofs are not very readable and raise an interesting philosophical question, which is what we actually mean by a mathematical proof. To some extent, a mathematical proof is just about whether you can convince other mathematicians that what you’ve done is correct. The computer—with these hundreds of thousands of pages—will never be able to convince a human being that what they’ve done is correct, even though the formal logic that they’ve applied is completely right and they do end up showing that they have actually proved this theorem, because it’s not readable by a human. The communication of the idea is of paramount importance in mathematical proof. I don’t think Marcus is genuinely worried about his job, but he’s seeing that there are areas that are being eroded and could potentially be taken over by machines. It’s a little bit scary. Math may be the first place but it will perhaps spread out further. Yes. Marcus is a pure mathematician by training, but he’s always been interested in artificial intelligence and the brain. He’s clearly outside his main area of expertise, but you can’t really tell. He deals with the subject in a very authoritative manner. Support Five Books Five Books interviews are expensive to produce. If you're enjoying this interview, please support us by donating a small amount . What is so impressive about the book is that it’s across so many different disciplines scientifically, but he also knows his classical music. He knows his art. He knows his literature. He knows all these areas that the AI is trying to venture into. Marcus is not just this one-dimensional automaton that mathematicians are often painted as being. He has different musical taste to the taste that I have—I don’t like classical music particularly—but he is able to write across so many different areas. It’s a really comprehensive book. Yes, we need to be thinking about how AI is going to encroach on our society. We’re talking about having driverless cars, we’re talking about replacing people’s jobs with machines or with algorithms. The ethics of AI is something we really need to think hard about at the moment. Yes, it’s Machines Like Me by Ian McEwan, who is my all-time favourite author. I’ve read all of his books. Machines Like Me segues quite nicely from Marcus’s book because it’s all about artificial intelligence and a counterfactual universe were these first human-like automata have been put up for sale. There’s a number of these models, and the book focusses on Charlie, an ordinary chap who inherits some money and decides to buy one of the male automaton, which are called Adams. The book is all about the power dynamic between them. This AI human is essentially a slave; he has a ‘kill’ button behind his neck so he can be turned off. But as the AI human learns, that power dynamic starts to shift. He learns from his creator. He learns all kinds of things from reading the internet; he learns that he doesn’t want to be turned off with this button behind his neck. So he learns to disable the kill switch and he learns—and we learn in the course of the book—that he’s actually more physically powerful than his human owner. McEwan’s book is a really interesting counterfactual look into what could happen if we were to have these AI servants in our lives. Get the weekly Five Books newsletter There’s a lot of good math in there as well. McEwan talks about something which I write about in The Maths of Life and Death as well, which is the P versus NP problem. He also has this counterfactual about Alan Turing, who didn’t commit suicide and goes on to be knighted and is the head of various artificial intelligence agencies. We meet him and we meet his partner who’s a famous physicist. I really enjoyed that aspect of the book because Alan Turing is also one of my heroes. Having him re-imagined as being alive and what he would have gone on to achieve had he not been so poorly treated that he felt the need to kill himself is really nice to see. Yes, before this novel, he wrote Saturday , which was about genetics, Solar , which was about physics, and Sweet Tooth , which features a not very good mathematician. In almost every other book he has a science theme going on. In the quote about my book he wrote, ‘for those like me who have little math, it’s delightfully readable’. So I don’t think he has a science background, but I think some of his family are scientists and he’s obviously very interested in science. He’s the ideal type of person that I wanted to read my book: someone who maybe doesn’t have much mathematics, but is interested and wants to learn. I think I’m a bit unusual. None of my family are scientists. My dad did English at university, as did my sister. I’m sort of the black sheep in that respect. Whenever I go home, they’re always talking about the latest novel by whoever-it-is and I have to try and keep pace with that. Reading novels is what I was raised on, so to have the chance of combining these two passions—math and writing—in The Maths of Life and Death has been an absolute delight."