Chef Yuki Watanabe doesn't describe what she does as "using AI." She says she uses a system that suggests "ingredient adjacencies" — combinations that are chemically compatible but culturally unusual. The system told her to try white miso with dark chocolate and black sesame. The resulting dessert sold out at her Copenhagen restaurant every night for four months.
Flavor as Data
The premise behind culinary AI is straightforward: flavors have chemical signatures, and chemicals that share signature compounds tend to pair well. This is the logic behind the "food pairing" hypothesis that sommelier François Benzi proposed in the early 2000s — the idea that ingredients sharing flavor compounds would be harmonious together.
Modern food AI goes further. It cross-references pairing data with cultural context, seasonality, dietary restrictions, and textural contrast. The result is not recipes but suggestions — probabilistic paths through flavor space that a chef can accept, modify, or discard.
"The algorithm doesn't cook. It doesn't taste. It can't tell you if something is beautiful. But it can tell you things that a human palate, trained by habit and culture, will never think to try."
The Creativity Question
The more interesting question is not whether AI can suggest good flavor combinations — it demonstrably can — but what this means for the nature of culinary creativity. If a machine tells you what to cook, are you still the chef?
Most practitioners answer quickly: yes. The suggestion is not the dish. The technique, the execution, the editing — the decision of what not to include — remains entirely human. The AI is a library, not a cook.