Will AI be the brains behind your next memorable restaurant meal? That is the question posited in a recent New York Times article in which two prominent Chicago chefs — Grant Achatz of the Alinea Group and Jenner Tomaska of Esmé — detailed the ways in which they’ve been utilizing large language models to devise both recipes and dishware and inch their kitchens closer to the transhuman future.
Achatz revealed plans for an upcoming menu at Next for which he has used AI to flesh out (as it were) the resumes and backgrounds of eight fictional chefs who will contribute to it. One is “Jill,” a 33-year-old Wisconsinite who has managed to cook in the kitchens of a great modernist chef, a sushi master, and Auguste Escoffier — the legend who codified French cuisine and has been dead for nearly a century.
I don’t know if Achatz has written the bios of the other seven, but I have a couple of suggestions I fed into ChatGPT:
“Glen” is a 38-year-old incel who lives in his mother’s basement in Orland Park. He is a self-taught cook who screams at the TV during cooking competition shows and is an obsessive fan of Anthony Bourdain. He is given the chance to prepare one dish for Achatz to earn a spot in the kitchen.
I was not prepared for ChatGPT’s weirder than weird but compelling response:
“Glen, the 38-year-old incel from Orland Park with a diet of cooking competitions and Bourdain quotes, arrives in Alinea’s kitchen carrying all the red flags and none of the résumé. But against all odds — and perhaps thanks to a kitchen porter calling out sick — he’s given a chance to stage for one night. The dish that catches Grant Achatz’s attention isn’t modernist, foamed, or nitrogen-frozen. It’s a deceptively simple tripe stew.”
The AI invented a pretty funny backstory about Glen and the terrible tripe stew his mother cooked, but what really caught my attention was just how sophisticated the recipe was. It involved an apple cider cure, a slow braise in a broth of chicken feet, anchovy, and charred mirepoix. It was then served in a lacto-fermented tomato broth and set with fried gnocchi and “a whisper of mint oil to cut the funk.” Honestly, I’d upload my consciousness to the cloud just to eat this dish. Perhaps we needed to look at a more challenging cook to join the brigade.
My next chef candidate was Q’og, a time traveler from the toxic planet Zhuf who is disguised in human form and claims to have worked at the French Laundry.
In seconds flat I was looking at a multi-course tasting menu featuring the likes of an oyster leaf tart with fermented kohlrabi pearls served in a chilled dish “shaped like a spiraling tooth.” None of it sounded particularly edible, though I was amused by a grilled morel stuffed with chlorophyll custard and served with an admonition to not talk because the mushroom was listening. The meal ended with a digestif of “hemispheric broth” that was “warm, umami-rich and mildly hallucinogenic.”
As the recipes were created by AI, nothing was stopping me from using the same tool to write the review. I asked ChatGPT to review this food in the voice of John Kessler of Chicago magazine. No problemo. “I” found the cooking uneven: “A few bites transported me. Others made me feel like I was being punished for crimes I haven’t committed yet.”
That’s a good line and a sci-fi deep cut. Intrigued, I then asked the app to imagine a restaurant I’d love. Now, I don’t know if it was just blowing smoke up my USB port, but ChatGPT surmised I was a “sharp but fair critic [who] doesn’t sugarcoat flaws, but gives praise where due.” Further, I often admire “balanced, composed cooking, especially with some international depth (e.g. Korean, Japanese, modern Mediterranean).”
So my ideal restaurant is called Lamplight. It’s in West Town and helmed by a second-generation Korean American who trained at Avec or Lula Cafe before working in Seoul and Istanbul.
While this is all fun storytelling, the menu actually does sound amazing. Allow me to list some of the dishes on it:
- Smoked lake trout crêpe with pickled onions and black garlic aioli
- Spring pea mandu in preserved lemon broth
- Grilled chicken oyster skewers with charred scallion miso
- Koji-marinated pork collar with rhubarb mustard and butter beans
- Charred cabbage steak with anchovy vinaigrette and sunflower seed dukkah
- Roasted carrots in fermented chili honey with whipped tofu
Man, oh man. If anyone wants to make these dishes, my fork is poised and ready. While I appreciated the sense of machine-generated humor that accompanied these inquiries, I was honestly amazed by just how legit the cooking sounded. In the New York Times article, Achatz said he was taking the dishes suggested to him and refining them. It sounded so gimmicky at first but now that I’ve seen how well these large language models can learn from the multitude of restaurant menus and restaurant reviews they have access to, I’m intrigued if not a little creeped out.
Yet there was one thing that artificial intelligence got spectacularly wrong. It estimates a meal for two at this restaurant would cost about $120. Hah! In your electric sheep dreams, ChatGPT.