Peut-on concevoir un menu dégustation de cinq services dans une vraie cuisine professionnelle, seul, grâce à l'IA ?

Le sourçage, le timing, la dégustation, le dressage, la gestion du matériel en panne. Les robots peuvent s'occuper de certaines parties ; personne ne gère l'ensemble du processus.

Background

AI and robotics have made strides in isolated cooking tasks such as chopping, sautéing, and sous-vide temperature control, yet none of these systems can independently plan, time, and execute a five-course tasting menu in a real working kitchen. Robotic arms like those developed at Moley Robotics demonstrate limited recipe replication but cannot adapt to on-the-fly decisions like adjusting doneness when a pan overheats or modifying plating when a component breaks (IEEE Spectrum, 2026). Current AI models such as those evaluated by IEEE Spectrum and profiled by Deloitte in 2026 act primarily as advisory tools, offering step-by-step guidance but relying entirely on human operators for physical execution and real-time judgment (IEEE Spectrum, Enriched May 9, 2026; Deloitte Insights, March 2026).

The working kitchen presents dynamic variables—equipment failure, ingredient variability, timing drift—that exceed the operational envelope of today’s automation. A human chef integrates sensory feedback (smell, taste, texture) with technical precision to adjust sauces, correct timing, and mitigate errors during service (Deloitte Insights, March 2026). While automated sous-vide circulators and precision temperature control devices reduce manual labor, they do not compose a coherent multi-course narrative or manage the orchestration across multiple stations (IEEE Spectrum, 2026). Research in culinary robotics remains focused on modular sub-tasks, with no system capable of running an entire pass, let alone a five-course menu, without human oversight (IEEE Spectrum, Enriched May 9, 2026).