This article is taken from the November 2024 issue of The Critic. To get the full magazine why not subscribe? Right now we’re offering five issues for just £10.

“Be kind to all dumb animals,” advised the old American newspaper sage, The Cheerful Cherub:

And give small birds a crumb;
Be kind to human beings, too.
They’re sometimes pretty dumb.

Here’s a small source of consolation: machines are even dumber. Everyone of real intelligence should reject AI as a nonsensical notion and an impossible aim. Even the dullest of my students distrusts ChatGPT and Claude3.

There are AI engines that can paint by numbers, regurgitate language typical of uninventive writers, rehash bits of Mozart into a simulacrum of a symphony and perform other plausible acts of piracy and plagiarism.

Writers short of inspiration (or impresarios short of cash) may turn to robots for formulaic output, such as a screenplay for Midsomer Murders or an editorial for the Daily Express.

Machines, however, will never produce good original recipes: they can’t cook; they can’t taste; they can’t scent a truffle or squeeze a steak; they have no experience or sensitivity, no passion or judgement or imagination; they have never eaten anything.

They’re essentially disqualified. Yet AI recipes abound on the internet, and some slickers and shysters are making money out of them.

AI turns food into fraud. “Our advanced algorithm,” claims the first site I encounter, in the squirm-inducing language of a snake-oil salesman, “will instantly craft a completely unique recipe just for you.” No, it won’t. It will merely ransack recipes already on the web. It knows nothing about you, except whatever misleading data some spyware may have accumulated.

It knows nothing about the ingredients previous recipe-writers have deployed: their provenance; the influence of the season; their freshness, firmness, variety, dimensions and all the other elements that affect cooking and that only the senses can detect.

Flagrant lies breed in sites that sell AI cookbooks: robots misrepresented as cosy, unthreateningly pretty authoresses; recipes reported as tested when they haven’t even been proof-read. Literal nonsense typically distinguishes the verbal output.

Ralph Jones is a model of journalistic bravery who makes the front line in Gaza or Ukraine seem benign

Joanne Molinaro, real-life writer of a vegan recipe collection, found an astonishingly similar, artificially generated book attributed to the non-existent “Rachel Issy”. AI “vegan” recipes have included chicken and egg without casting light on which came first. In The Ultimate Crockpot Cookbook for Beginners, supposedly by the suspiciously prolific but humanly unverifiable “Teresa J. Blair”, 99 per cent of the recipes are for mojitos.

The aim of AI-exploiters is not to serve the public but to multiply “hits”. The recipes are “crafted” not “just for you” but in order to maximise popular search terms. It’s easy, in consequence, to find recipes for “Barbie-themed appetisers” and random combos masquerading as fusion food.

Ralph Jones, a Guardian reporter, tried the AI Cookbook’s “Shepherd’s Pie Sushi” and “Thai Green Curry Lasagne”. He is a model of journalistic bravery who makes the front line in Gaza or Ukraine seem benign. The story of a recipe for “aromatic water” that emitted chlorine gas is as widespread on the web as any urban myth, but unhappily it’s true. AI engines engorge the rubbish other robots generate, multiplying the likelihood of madly mismatched ingredients.

Mallory Arnold, a journalist who confesses to liking kimchi with fried eggs, seems an ideal client for these weirdly incautious machines. No self-respecting robot ought to have responded to her demand to pair an unspecified type of sausage with peanut butter: the only proper advice is to skip the meal.

But if, like Ms Arnold, I really had a craving for chicken breast with brown sugar and tomatoes, I wouldn’t do as her robot suggested — adding soy sauce. I might serve grilled chicken with a soffritto, including slightly caramelised Roma tomatoes.

Robot-generated recipes that aren’t bizarre tend to be banal. When I prompted “Buzzy” with the simplest ingredients — water, salt, pepper, olive oil and garlic — my computer begged me to wait whilst “Buzzy is thinking”.

The result of the intellectual effort was a suggestion for combining unspecified pasta with “chopped seasonal vegetables (like [sic] zucchini, bell peppers, or spinach)”. Buzzy’s second attempt was for similarly dressed quinoa with chickpeas, which he forgot to tell me to soak or cook, and red peppers, uncooked and unpeeled.

Buzzy and his cohort are dumb. But culinary failures don’t prove that AI is impossible: you can’t judge microchips by a skill that needs senses as well as sense, and emotion as well as emojis. In the unlikely event AI attains human discrimination, virtual chefs will still say, “Yes” to weird requests. But here’s the real test: sexbots will say “No”.