Is this by Rothko or a robot? We ask the experts to tell the difference between human and AI art – The Guardian

Is this by Rothko or a robot? We ask the experts to tell the difference between human and AI art – The Guardian

The year 2022 was when AI-generated images went viral. Online, you may have come across very realistic yet suspiciously improbable images of, say, an astronaut riding a horse through space or an avocado doubling as an armchair.

Numerous new generators – including Dall-E, Midjourney and Stable Diffusion – offer anyone with an internet connection the chance to conjure up their own strange apparition, simply by typing in a “prompt” for the AI. (For example, “astronaut astride horse on Mars”. Or, for this article, “Mark Rothko Abstract Expressionist oil painting” – yes, the image above isn’t a real Rothko.) The possibilities have been endless, the opportunity for meme-making infinite.

It should not be surprising that a great many artists who have spent a lifetime honing their skills are a little put out by this latest disruption. Are companies going to keep hiring designers when they can produce prototypes themselves for free? Will budgets stretch to include animators if their hand can be imitated from a simple text description? Advocates of AI have insisted that creatives should have nothing to worry about and can adapt their process to incorporate or work around technological advances, much like the modernists did with the invention of photography.

But if those historical greats were alive and working today, would they also be watching their backs? And could a computer ever hope to reproduce the emotional depth that gives great art its charm and meaning?

To find out, we set a challenge for three art experts: Bendor Grosvenor, art historian and presenter of the BBC’s Britain’s Lost Masterpieces; JJ Charlesworth, art critic and editor of ArtReview; and Pilar Ordovas, founder of the Mayfair gallery Ordovas. Each was invited to look at pairs of artworks of a similar style and period over Zoom to see if they could tell which was generated by a machine. All three admitted to finding it tougher than expected …

Nineteenth-century landscape

(Left) Homer Watson, Down in the Laurentides (1882). (Right) An image generated using Dall-E with the prompt “Landscape oil painting Constable Claude Corot”. Composite: Homer Watson/ National Gallery of Canada; Image generated by Jo Lawson-Tancred and Philip Booth

Bendor Grosvenor “When authenticating a painting, composition is usually the last thing I would look at, after brushstrokes and condition. The one on the left looks like New Zealand, with the cows a bit plonked in and the grass not particularly well painted – but I quite like the way the light falls on the hills. There’s something about the picture on the right that looks a bit too good to be true. It’s got the bright, contrasty clouds of a Constable and the winding river reminds me of the French Barbizon school. If you asked a computer to make a Constable, that’s probably what it would come up with.”

<p class="dcr-1…….