Artificial intelligence and art

There is a challenge at hand. Artistes would do well not to be complacent about the superiority of their creativity

SEVERAL AI COMMENTATORS, including Gary Marcus, professor emeritus of Psychology and Neural Sciences at the New York University, have recently remarked on the fact that Dall-e, Openai’s image generation technology, finds it easier to generate an image of an astronaut riding a horse than it does a horse riding an astronaut. In either case, the prompt is asking the AI to do something outlandish. But the second is somehow more bizarre for the algorithm that is Dall-e — it is, we might as well say, easily misunderstood. If you demand an image of a horse riding an astronaut, you will likely get an image of an astronaut riding a horse.

Is this Dall-e being human-like, defaulting into that which is arguably more imaginable from the spectrum of the bizarre; or is it just simply making a mistake that it is likely to? There aren’t many images of a horse riding a human — let alone an astronaut — in the training data most likely. So, presumably, Dall-e does what it can do.

One can, perhaps, venture from this to opine on the difference, as it stands today, between the machine and the human mind. I find it useful to speculate on what happens in the human mind (the abstraction as we know it, as against the biological brain and its electricities) when it is given the same prompt, that is when it is asked to imagine a horse riding an astronaut. What are the minutest steps in our mental processes, steps that we can identify as abstractions and thus name using the most common language?

There is that slight scratch of dissonance, I think, an acknowledgement that the prompt is bizarre, and then, along with a flash of understanding the mind at once constructs an image (usually hazy, grey, possibly fluid, with several undetailed archipelagos). Three identifiable stages, then, all traversed in the briefest time possible: Primary parsing; understanding; and imagination.

Subsequent production, the transfer of the mental image to a medium — paper or software — is slow; its quality is dependent on the imaginer’s talent with the medium; and its correspondence with the mental image is unknowable, or knowable only through the articulations or anguish of the subject, that is, to an unreliable limit.

Of these, which ones does Dall-e lack in the context of the prompt?

On an abstract level, at which we understand anything, Dall-e merges the first two— parsing is understanding — and the last two — production is imagination —and it merges them to a mechanic utilitarian degree, one that makes us wonder if it has any understanding or any imagination. Its horses do not ride astronauts.

Generative AIS’ lack of understanding and imagination, as evidenced in their failure to tackle the truly bizarre, or their failure to produce original works of beauty (CHATGPT is awful at poetry), or their failure to appreciate the laws of physics in the images or videos they produce, is of some solace to artistes. In interviews, when asked the mandatory question of the threat AI poses to them, writers tend to emphasise Genai’s lack of “lived experience” and “imagination”, its definitive unoriginality, and, occasionally, the fact that art is not so much in the artefact as it is in the conceptual is at ion and development of the artefact.

Though there is nothing wrong in these remarks, they contain, to my mind, a misreading of the threat. GENAI attacks not art at its apogee but the economies of the artefact. Writers are not facing the possibility of CHATGPT producing an extraordinary and original story, but the prospect of it producing an infinity of bad, unoriginal, plagiarised or near-plagiarised stories. It may not have experience and understanding and imagination, and it may not have a window into what it feels like to write a story, but its ability to produce artefacts at the click of a button is an upending of things as they exist. In literature and cinema, we already see a winner-takes-all economy, where a few select works, whether of high quality or mass appeal, whether by accident or by design, gather a lion’s share of the revenues generated. As the reproducible is reproduced at the click of a button, as the price of production nears the asymptote of zero, and as the zone of human originality is challenged further and further (surely GENAI will also get better at aesthetics and originality), all human-created literature and cinema become luxury items. The upshot: Winnertakes-all intensifies exponentially, a lesser and lesser number of people get to live the experience of making art, the figure of the bold amateur vanishes from the scene.

The year 2024 is crucial for Openai because of the lawsuits it is facing across geographies. The organisation sometimes sounds rattled. It has called outright plagiarism a “bug” and has suggested that strict enforcement of copyright laws will be an existential threat to it. Artistes of any kind — even those who today use GENAI to aid their work — must read between the lines here. The algorithm intends to swallow everything you have made — for it cannot function without it — and it intends to swallow everything for no consideration to you. You’d do well right now not to bask in the nuanced superiority of your humanity, of your ability to imagine a horse riding an astronaut. Things are heating up. Think of what verdict might help sustain your one glorious and artistic life. Ask loudly for it wherever you can.

Solanki is a writer, most recently, of Manjhi’s Mayhem