I don’t know about you but I’ve become a bit disillusioned with the internet of late. I mean, great, you can get a low-paid worker to bike a tepid Big Mac round to your house with a couple of swipes. It feels like a small consolation for the hours we have lost distracted and depressed; for fake news and sur - veillance capitalism; for Donald Trump and Andrew Tate; for Snapchat dysmorphia and Facebook politics.
So, I will admit I emitted a small squeal of delight when the first pictures from Dall-E 2 began to circulate online last year. Here, it seemed, was a glimpse of the wild, creative future we had been promised. Here, I thought, was something new. If you haven’t had the pleasure, Dall-E is one of a new wave of programs showcasing the wonders of ‘generative artificial intelligence’, the latest Silicon Valley craze. You can type just about anything into Dall-E’s text box and it will generate an image of it for you: ‘Banana eating itself.’ ‘Martian landscape in the style of Canaletto.’ ‘Bowl of soup that looks like a portal to another dimension.’
The results are uncanny and impressive; a bit like Google, but for the weird dream you had last night. Tom Furse, the digital artist who created the rather beautiful images here and on the front of this magazine, experienced a similar epiphany when he first used VQGAN+CLIP — another generative AI program that he describes as like Dall-E but ‘much weirder’ and ‘more incoherent’.
‘I heard about it and I just thought: “Maybe that would be a fun thing to do today,”’ he says. ‘And that was it: I became completely addicted to this magic, infinite image box.’ He has now dialled down his day job (he’s the keyboardist for the band The Horrors) in order to usher in our new digital future. He thinks generative AI is precisely the shot in the arm that the creative arts needed. ‘If you look back at the history of music, there’s so many points where a piece of new and usually controversial technology is the thing that spurs on creative change. I feel like we were overdue something like that.’
For it’s not only pictures that can now be created this way, but poems, stories, symphonies, animations, maybe even movies some day. ITV will shortly unveil Deep FakeNeighbour Wars, a sketch show that uses AI to draft Stormzy, Harry Kane and Nicky Minaj into service as comic actors. OpenAI, the company behind Dall-E, is also responsible for the text generator, ChatGPT, and a music program called Jukebox, which is a bit like Dall-E but for music. If everything goes to plan, we will soon be able to hear ‘All the Single Ladies’ as a fugue in D minor with a trumpet solo by Miles Davis simply by cutting and pasting that prompt.
Naturally, you can already generate custom pornography via another text-to-image interface, Stable Diffusion, if you find your way on to the right Discord server. Yes, some of the creatures in it have 13 fingers and frightening nipples, but these errors can be trained out with higher-quality data. And hey: whatever turns you on.
Still, after the smile comes the shudder, a little like the one experienced by the Google engineer Blake Lemoine when he became convinced the chatbot that he was working on had become sentient. Artificial intelligence and machine learning already permeates so much of our conscious and unconscious lives, from Google search results to online assistants, facial recognition software to Facebook newsfeeds.
Amp all that together and a still more dystopian future beckons: deep fake revenge porn, surveillance capitalism on meth, automated feels, political manipulation on an undreamt of scale. First the robots came for the factory workers — and I did nothing. Then, they came for supermarket checkout assistants and I did nothing. Now they have come for artists, musicians — and clearly I am screwed as all my editor will need to do is type ‘Richard Godwin article about AI Art’ into GPT-3 and I will become as useful as a MiniDisc. ‘I think the best case is so unbelievably good that it’s hard to... it’s like, hard for me to even imagine,’ Sam Altman, the CEO of OpenAI, said in a recent interview. The worst case scenario? ‘Like, lights out for all of us.’
But let’s not get too ahead of ourselves. The fact that OpenAI was recently revealed to be using a Kenyan datafarm to filter out various child-sex-animal scenarios from the Dall-E results suggests we should take that hype with a degree of scepticism. I recently watched a four-hour Zoom lecture on the prospects for artificial general intelligence (which Elon Musk has promised us by 2029) and the consensus of people who actually understand this stuff is that it’s not happening any time soon, if ever. For the moment, even the best text generators are full of silly errors, incapa - ble of common sense, abstraction, reasoning or many other basic components of human intelligence, and tend to exac - erbate our biases and flaws. ‘Is there anything of value in GPT-3? It’s hard to find anything,’ was Noam Chomsky’s verdict. Nick Cave, meanwhile, responded to a fan’s artificially-generated Nick Cave song as ‘a grotesque mockery of what it means to be human’.
Still, no one mistook a camera for a human — and it still changed the course of art. I was interested to see what an artist would make of Dall-E as a tool, so I turned to the writer and painter Yelena Moskovich, whose work often touches on digital technology. Unfortunately, the prompts that she came up with violated the program’s content restrictions. Dall-E doesn’t like suffering; it doesn’t like politics; and it won’t let you put any real people in your images either. With a bit of rephras - ing, we did manage to come up with: ‘Final thoughts of someone before they die, art brut’ and ‘Two friends in a future genderless society.’
But the gimmick didn’t last. ‘When I looked at a couple of images, it was exciting,’ says Moskovich. ‘But after a couple, you see the humans behind the algorithm more than some computer that’s come to life. If these were graphic designers, I’d say: “Okay, I get their style.”’
These limitations became even more apparent when I reached out to OpenAI. I was introduced to a ‘science com - municator’ named Andrew over Zoom. He was dead keen to show me things like ‘Tuscan sunset’ and ‘Corvette in a desert’. I was more interested in how the machine would handle something like ‘despair’. Andrew assured me that Dall-E could do despair — but when he typed it in, Dall-E spat out four images of Asian people with their heads in their hands. ‘How come they’re all Asian?’ I asked. Andrew was not sure, but he flagged this — presumably for the team in Kenya. Glitches like these are one of the reasons why these programs require extensive testing and training before they are released into the real world.
But here lies one of the main perils of artificial intelligence. It can only draw on the data it is fed and is thus subject to all our existing stupidities — and our crude attempts to remedy them. Galactica, Meta’s attempt at a large language model, reliably produced fake news articles and had to be taken down after three days. Dall-E reliably reflects the priorities and concerns of the San Francisco tech geeks who coded it rather than humanity at large.
It’s too late to go back now. I’ve accepted our fate and I’m here for all of it
Which isn’t to say that it’s lacking the essential spark of creativity, says Sacha Golob, reader in philosophy and aesthetics at King’s College London, whose job it is to figure out precisely what art is. ‘There are people who will say that a programmer has put in various specifications and the machine is just grinding through them all,’ he says. ‘I think that view is too hardline — it misses the distinct novelty. There’s a gap between the input that you can give Dall-E and the range of outputs that it can produce. And because of that gap, there is a kind of creativity in play.’
For all the promise of machine-learning to deliver some brilliant future, what it really does is trap us in the recent past — since it can only make its predictions from things we’ve already done. This is why Facebook is always trying to sell you the raincoat you bought last week. And it’s why Dall-E is really good at things like ‘The Last Supper but with Minions’, but not so good at imagining something brand new.
But then again: how many human artists are capable of coming up with something brand new? Not many. ‘I have a lurking suspicion that we haven’t quite understood the impli - cations of Dall-E 2,’ says the film-maker behind the BBC documentary series Can’t Get You Out of My Head, Adam Curtis, when I email him for his thoughts on Dall-E. His overriding feeling is that generative AI is a ‘quiet depth charge’ under - mining the idea we are all uniquely creative individuals. ‘It may mean that creativity just isn’t that difficult,’ he says. ‘That really it’s quite banal.’ We imagine that self-expression is radical, precious and will set us free. Really, when it comes down to it, most of us are sim - ply regurgitating bits of culture that we’ve imbibed along the line. ‘The machines are quietly working to undermine this idea, because they can see the dirty little secret that actually we are far more like each other than we think.’
Generative AI emerges at a time when ‘originality’ is at a premium anyway. The more digital media penetrates our lives, the more similar our tastes seem to become. Ever noticed how everyone posts exactly the same sunsets and breakfasts on Instagram? The people who produce this content are already slaves to the algorithm, however much we would like to think they’re not.
In which case, the derivative of art that Dall-E produces might just be the art we crave. As Golob points out, it’s not like most of us are hanging masterpieces on our walls. ‘Lots of people just like crap art!’ he insists. ‘They like clichés. They like sentimentality. And they’re going to like it whether it’s generated by a human or a chatbot.’
Maybe what we treasure in art isn’t individuality. It’s a shared experience. It’s something to make us feel connected and less alone. ‘Self-expression might have become the con - formity of our age, and far from being radical it actually dis - empowers us,’ says Curtis. ‘Because you know what tends to happen to artists in their garrets? They get lonely and isolated. Whereas if you really want to change things you have to come together in the mass and demand change.’
And it is precisely the flawed, imperfect, not-quite-thereyet aspect of generative AI that Furse likes. The singers we love aren’t the ones who are the most technically adept, they’re usually the ones whose voices crack and smear.
In the meantime, says Furse, there is work to be done. He conjures a future of bedroom avatars; of living artworks; of undreamed-of collaborations with dead artists. ‘I’ve always wanted to make an album with Chet Baker,’ he says. ‘That’s almost realisable now. Extrapolate that to every single artist, ever. It means your work need not die with you.’ And yes, awful things will happen, too. ‘But it’s too late to go back now,’ he says. ‘I’ve accepted our fate and I’m here for all of it. A wonderful, fresh new medium. A whole new playground to splash around in. What a rare occurrence!’
