clock menu more-arrow no yes
Amanda Northrop/Vox

Artificial intelligence can now make art. Artists, don’t panic.

From painting to poetry, music to math, we can view AI as a collaborator rather than a competitor.

Music lovers gathered in a London theater one night in March to take part in an unusual event: half classical concert, half futuristic experiment. Their task was to listen to music that had been composed partly by Bach and partly by artificial intelligence — and try to guess which parts were which. Throughout the performance, the audience members voted by holding up cards with a blue human face on one side and a red robot face on the other.

“It was quite shocking,” Marcus du Sautoy, an Oxford mathematician who masterminded the event, told me. “There were moments when I think Bach would have turned in his grave! Moments when Bach was playing and people were saying it was the AI.”

Experiments like these are becoming more common as researchers try to create AI tools that can generate music — and paintings, and poetry — that’s just as compelling as the human-made kind. They often rely on machine learning, a type of AI that involves feeding computers example after example of something until they learn to pick out the patterns in it and create their own version.

In the case of the London event, the AI had been fed enough of Bach’s music that it was able to learn and then mimic the composer’s signature style, fooling the audience.

The AI researchers designing these tools aren’t doing it for the fun of tricking people. They’re trying to prove that they can take AI farther than we’d previously thought possible — that they can make machines creative, just like human beings.

And some creative humans are happy to have AI join the art world. They see it not as a threatening interloper, but as a potential collaborator that can get them out of their usual ruts and spur them to think in new directions.

Du Sautoy is among those optimists. In his new book, The Creativity Code: Art and Innovation in the Age of AI, he examines the likelihood that AI will become creative in its own right, where creativity is defined as “coming up with something that is new, that is surprising, and that has value.” He explores the way AI is changing music, visual art, literature, and math.

Refreshingly, he ends up arguing that we should see the relationship between AI and humans not as adversarial, but as collaborative. As a novelist, I wanted to speak to him about what to do when we nevertheless feel unmoored by the computerization of everything. A transcript of our conversation, lightly edited for length and clarity, follows.

Pierre Fautrel, co-founder of the French collective Obvious, stands beside an AI-generated painting titled “Portrait of Edmond de Belamy.”
Pierre Fautrel, co-founder of the French collective Obvious, stands beside an AI-generated painting titled “Portrait of Edmond de Belamy.” It sold for $432,500 at auction. The signature on the painting is a mathematical formula.
Timothy A. Clary/AFP/Getty Images

Sigal Samuel

In popular media, the relationship between AI and humans is often pitched as adversarial rather than collaborative. Why is that?

Marcus du Sautoy

The public is in an uncertain place about where the future is going. One of the frightening things is that the AI revolution is very fast — within one decade, you’ll see a lot of jobs disappear. So everyone is going through a slight existential crisis. And the movies love feeding this dystopian Terminator image of AI.

But we’ve all felt the one thing that’s uniquely human is our creativity. Now we’re having to ask, what if AI can approach even that? That’s a real challenge.

Sigal Samuel

I tend to think that if we come to realize human creativity isn’t so unique, there’s actually something potentially very freeing about that. Just like the Copernican revolution made us realize not everything revolves around us and the Darwinian revolution made us realize we’re actually a lot like other animals — that can be very unmooring, but it can also take away a lot of human-centric ego, which could be a healthy thing for us and the planet.

Marcus du Sautoy

Absolutely! That’s one of the strands I wanted to pull out: People think art is something very mystical — that there’s something appearing out of nothing, the creative genius. I wanted to reveal that a lot of creative acts do have structure and pattern and algorithms and logic.

Especially with music. Many people think emotions are just being spilled out onto the page, but any composer will tell you, “I’m actually doing something very structured, and the emotion arises out of the controlled acts I’m using in creating a piece of music.”

Sigal Samuel

Do you think an audience would be less impressed by AI-generated music if you primed them in advance by telling them which parts were composed by AI?

I’m thinking of this part in your book where you describe a critic who reacted very negatively to Microsoft’s 2016 attempt to have AI create a painting in the style of Rembrandt. He called it “a horrible, tasteless, insensitive, and soulless travesty.” I had the feeling the critic wouldn’t have reacted as negatively if he hadn’t known the painting was made by AI.

Marcus du Sautoy

That’s what was interesting about Art Basel [a major contemporary art fair in 2016]. There, they didn’t tell people the art was created by AI, and the people had a positive emotional reaction to it — actually more positive than for the art produced by humans!

I think knowing [an artwork] is produced by AI does color your reaction to it. And perhaps justifiably so, because you want to feel that you’re connecting to another human being. And so if you find out there is no human being behind it, you do feel cheated.

Sigal Samuel

Should we be trying to override that part of our brain, though? The part that says, “Hey, I expected to be communing with another human’s soul here”? Part of me wonders if it makes sense to be super pragmatic about art and our interactions with it: Maybe if I enjoy art made by AI, I should just learn to value my positive emotional response and try to divorce that from any presuppositions about the intentionality behind the art. In literary theory, there’s this whole idea of the intentional fallacy — basically, the intention of the author isn’t actually what matters.

Marcus du Sautoy

It’s worth remembering that for the moment, AI is producing things that still connect to our human creativity because it’s learning from our own art. So we are still having a reaction to human creativity but through this new filter. So even though you might feel cheated when you realize something was made by AI, you’re responding to something that ultimately had its source in the human.

Sigal Samuel

That’s true. I think the problem arises when we think we’re having a human interaction and then we find out it’s not that and we feel tricked. But if we know from the get-go that the art is generated by AI, maybe we can enjoy it for what it is.

I actually had this experience of feeling tricked when I was reading your book. You sneakily put in a 350-word paragraph that’s written by AI, and only later cop to the fact that it was AI-written. I knew when I was reading it that something was off because it had typos and it didn’t fully make sense. And then later, when you explained why, I felt so prickly about it! I yelled, “I knew something was wrong!” and felt so vindicated.

Marcus du Sautoy

Interesting! Yeah, I had to fight to get my copy editor to not correct those glitches.

Sigal Samuel

Aside from playing those little tricks on your readers, you do sound in the book like you sincerely believe machine learning can revolutionize creativity. As a mathematician, the work you do is fundamentally creative (even though math isn’t classically seen as one of the creative arts) and you talk a lot about how you’ve been feeling nervous as you watch AI horning in on the territory of creativity. Did you write this book because you were having an existential crisis?

Marcus du Sautoy

I did, yes! The whole thing was initiated by this sense that the new AI that’s emerging may well be able to do the thing I never thought it would be able to do, which is to create mathematics. That word “create” is what I always thought protected my subject from computers. I didn’t regard computers as at all creative, and creativity is a huge part of doing mathematics.

But then I started to see AI being creative in the game of Go [in 2016, when Google’s DeepMind team challenged a human champion to a series of matches against its AlphaGo program]. I watched the Go matches on YouTube obsessively because that game was always one of my protective shields against computers doing math — everyone always said computers will never be able to play Go. As I watched AI besting humans and actually showing them how to play the game in a new way, I thought, this thing is creating its own genuinely surprising moves. This is a significant moment.

Sigal Samuel

For now, though, it seems like a lot of codes attempting to make art are pretty good at mimicking human artists on the local level (say, a few notes at a time), but they fail at generating a larger structure that feels satisfying (like a complete work of art). Why is that?

Marcus du Sautoy

The way machine learning is working at the moment is, it’s doing quite a local analysis on text or music or pixels. So, for example, the jazz Continuator [an algorithmic composer created by François Pachet that learns your musical style as you play and continues your melody in real time] hears a few notes and then produces sound based on an analysis of the music it’s heard up to that point. That level of coding doesn’t seem able to get at a global arc or narrative structure.

I think that’s why when you turn to literature, where AI has been most successful is in poetry. Because poetry is a gnomic form. It’s got a lot of pattern in it — rhythm and rhyme — but also I have to do a lot of work as the reader, I have to bring my understanding to the poem. AI is quite successful in poetry because it’s able to create something that leaves enough ambiguity so the reader can use a lot of their creativity to bring the poems to life.

But when you turn to [longform] literature, AI has been quite unsuccessful. It can do pretty good short-form prose, sound pretty convincing with a bit of Harry Potter, but not longer prose. I don’t think that means it’s unsolvable; it’s just a harder thing to tease out. We know there are formulas for narratives in film, and archetypal stories in the novel, so in theory, why couldn’t you take one of these templates and use a good text generation algorithm to fill the gaps?

Sigal Samuel

Yeah, and it’s actually been said there are really only six main story arcs in all of literature. Given the possibility of AI mastering them, do you think that as a novelist, I’ll be out of a job in a few decades?

Marcus du Sautoy

Rather than both of us being out of a job, what I hope is that maybe we’ll be able to push ourselves in interesting ways as the AI becomes a partner or tool to extend our own creativity. I get so stuck in ways of thinking and sometimes I need something to kick me out of that. AI can help us behave less like machines and more like creative humans. That’s the most exciting thing.

Sigal Samuel

Do you think machine learning will prove an important new theorem in the next few decades?

Marcus du Sautoy

I actually think it’s way off. By the end of [writing] the book, I was encouraged because I felt I’m still very much in the game. For AI to actually come up with new ideas in math — I think we’re still a long way from that. DeepMind actually released a paper a couple weeks ago where AI had been trained on mathematical ways of thinking and then took a school-level paper and it didn’t even get a pass on it. But we’ve all got to start somewhere.

What you do as a novelist is very interesting to me because I think we actually come much more hardwired for language than we realize. The evidence suggests that a child learns language with exposure to very little data, which chimes in with Chomsky’s idea that we do come preprogrammed. And that preprogramming is millions of years of evolution. Whereas we haven’t been through such an evolutionary process in creating our mathematical language. So math is something that AI could perhaps fast-track, much quicker than being able to tell the stories that you tell.

Sigal Samuel

You mean my job is a bit safer than yours?

Marcus du Sautoy

I think it might be, yes.

Sigal Samuel

Great. Let’s stop there so I don’t have to have an existential crisis.


The US government’s battle with Chinese telecom giant Huawei, explained

Future Perfect

A celebrity biohacker who sells DIY gene-editing kits is under investigation

Recode Podcasts

How Amazon turned its darkest time into runaway riches

View all stories in Technology