It is May 2021 and you’re on your way to your first post-Covid dinner party — held at the home of an old friend. As you enter, you notice on the walls a good ten new bits of art: a few abstract paintings here, a couple of sombre portraits there, and, down at the end of the hall, some kind of an impressionistic landscape. “What’s the story with these? Part of the family collection?” you ask. “Oh no,” he says, “they’re all new: pieces entirely generated by AI — it’s all the rage at the moment.”

Sound far-fetched? Nope — it’s already happening. Just last month I stumbled upon the “Art AI Gallery”, a website boasting the world’s largest collection of AI-generated art for sale. By my calculation, they’ve already sold over three and a half thousand unique artworks in all manner of styles — “abstract”, “surreal”, “nature”, “humans”, “flowers”, “fauvism” — all of which were produced by just two algorithms working in tandem: one that analyses actual, historical works, and one that generates new ideas from scratch and then, by learning over time how to “trick” the first algorithm into believing what it produces is created by a human, refines them into something more formally conventional.

The Art AI Gallery claim to be ‘democratising’ art

There’s been a burgeoning market for this stuff ever since 2018, when a splodgy portrait called “Edmond de Belamy” became the first AI-generated artwork auctioned at Christie’s — going for a handsome $432,500. The following year, the HG Contemporary Gallery in New York ran the first ever solo exhibit by an “AI artist”, an algorithm called AICAN, with “portraits” going for between $6000 and $18,000. The Art AI Gallery sells its works for a more modest figure: somewhere between $50 to $100 per work (they claim to be “democratising” art, which sounds to me like a feel-good way of saying they’re going to put a load of struggling artists out of business). On their “About” page, they state proudly that their algorithms have “the potential of upending the art world” (this is supposedly a good thing), and, in a chummy introductory video, point out that — despite some understandable public opposition — AI artworks are, at the very least, “one heck of a conversation starter.” Sure — a bit like nuclear warfare.

Still, as much as I’d love to say it’s all complete rubbish, there is something undeniably compelling about a few of the pieces: there are one or two mesmerising portraits — spooky, warped faces of people that never existed smudged against an alien landscape — and some perfectly passable abstract works. And if nothing else, the whole thing helps to shine a light once and for all on Banksy’s secret identity: a giant supercomputer in Bristol being fed cartoons and GCSE politics exam papers.

What, besides loyalty, dictates that we should privilege the artworks of humans over algorithms?

Whether we like it or not, though, the intrusion of AI into the domain of human creativity is going very quickly to become a fixture of our lives — and not just in the visual arts. A company called AIVA (Artificial Intelligence Virtual Artist) has developed a programme that allows anyone to generate “emotional soundtrack music” — already allowing companies like Vodafone, Nvidia and TED to bypass human composers altogether. In September the Guardian ran a terrifyingly sharp piece (and not just by Guardian standards) written entirely by an algorithm called GPT-3. Its predecessor, GPT-2, has been used for all sorts of literary experiments — the artificial intelligence researcher Janelle Shane programmed it to generate the opening lines of stories, resulting in at least a few genuinely interesting starts: “I am, or was.” / “The black stone was aching from the rain.” / “I was playing with my dog, Mark the brown Labrador, and I had forgotten that I was also playing with a dead man.” Even this piece you’re reading right now was actually generated by an AI algorithm running on my laptop. (Alright, I’m joking about the last bit: this one’s entirely by me, and I probably fail both criteria for “artificial intelligence”.)

Naturally, all of this raises difficult questions, especially for the purists, like myself, who think art should be assessed on its own merits, not the merits of its creators. Yes, a lot of the AI stuff being churned out at the moment is, admittedly, pretty stodgy, but we’re naive if we don’t think it will get much better, and quick — the hope that many of us once had, that artificial intelligence would never match the quality of humans, is almost certainly long gone. What then, besides loyalty, dictates that we should privilege the artworks of humans over algorithms?

If there’s any hope, it’s this: AI-generated art will get rapidly better at two things — pastiche derivatives and strange, mangled amalgams of styles. Both are obviously fascinating in the immediate term — at least for their historical significance — even if they’ll likely lose novelty pretty quick. But there’s one thing the bots will never be able to do: make reasoned decisions about the direction art should go.

Take music. It’s not at all obvious that, if we’d had supercomputers and “neural networks” in — say — 1780, they’d have been capable of doing much more than generating passable pastiches of Baroque or Classical compositions: there would have been no way for them to tell, from the existing material of the eighteenth century, for instance, that a hundred years later we’d be listening to the dense, rich, strained chords of Mahler or Debussy. Similarly, if we’d trained these supercomputers on the visual arts, they would have struggled to anticipate, say, the impressionistic late works of J.M.W. Turner — derided at the time as “the imbecility of old age” and recognised today as the work of a prescient genius — not to mention the abstract works of the early twentieth century.

The only way artificial intelligence could plausibly predict the specific kinds of innovations humans would, in the future, want, would be if our very thoughts themselves — the reasoned decisions that direct the development of human cultures — could be observed physically. Machines learn — terrifyingly quickly — from what they observe, but as yet we know of no way they could possibly “observe” anything without a physical form. If there is any non-physical aspect to the human mind, then, this — at least — would remain invisible to any prying computer, and present a blind spot where human ingenuity could still flourish.

No matter how sophisticated AI becomes, it will always depend, ultimately, on human judgement

Of course, there are some who do believe our thoughts are entirely physical — that we are simply mechanistic meat machines, whose subjective, conscious experiences are wholly illusory, and whose every decision was predetermined at the very start of time. But the philosophical arguments for such a position are remarkably weak. In fact, there aren’t really any — aside from the non sequitur that, since an increasing number of things in the universe can be shown to be physical, it will inevitably eventually turn out that everything is. This is, of course, a classic case of the evidence (that we have subjective experiences others cannot access or inhabit — perhaps the most unshakeable intuition we all have) being made to fit a preconceived theory: that all things simply must just be reducible to pure matter. But while the last few hundred years have yielded remarkable scientific discovery in pretty much every other domain of life, they have brought us no closer actually to explaining the nature of subjective experiences. Yes, we have a far better understanding of the brain, and yes, we can observe that certain sensations correlate with certain physical changes, but none of this makes any better sense of what the internal feeling actually is. To paraphrase Nietzsche: we have come to describe in ever greater detail, but can explain nothing more.

Which brings us to another difference between us and AI: we reason, AI reacts. Sure, it can process its tasks with stunning speed, far faster than humans will ever be able to do: just last year, for example, researchers looking for disease-causing genes used an algorithm to analyse more than a trillion “genetic data points” — something that would take a human thousands of years to do. But no matter how powerful and sophisticated artificial intelligence becomes, it will always depend, ultimately, on human judgement. In the world of AI art, the algorithms have to be designed by humans, the artworks generated need to be sifted through and okayed by humans, and the final pieces need to be chosen, and bought, by humans.

Still, in the end, it doesn’t require AI to become entirely standalone, nor to be able to innovate, for it to completely upend the art world. It’s not especially obvious, after all, that among the general public, there’s much demand for genuine innovation: indeed, the opposite seems true. And if we end up provided with an endless mine of derivative works, it may well be that the incentives for artists to develop new ideas dwindle altogether, and the one thing the human mind truly is capable of — intentional reasoning — ends up being dispensed with altogether.