[2] GASTRO-X: Can the senses make AI kitchen advice worth trusting?

On 21 September I started a two-year Marie Skłodowska-Curie fellowship at the University of Oxford. The project, GASTRO-X, asks whether sensory cues can make people trust AI recipe suggestions, and whether that trust helps them waste less food. It now has its own website: gastro-x.califano.xyz.

Posted on by Giovanbattista Califano

A trust gap

Households produce around 54% of the roughly 59 million tonnes of food wasted each year in the European Union (Eurostat). Most people say they feel responsible for it. Few change their habits, because the decision happens on a Tuesday evening in front of a half-empty fridge, with little time and no idea what to cook.

Generative AI is good at exactly this problem. Give it half a courgette, stale bread and two eggs, and it will suggest a meal in seconds. The obstacle is not the AI but us: people accept AI for technical tasks and grow sceptical when it comes to food. In earlier work at the Crossmodal Research Laboratory we found that recipes are trusted less when presented as AI-generated, especially when they are unusual (Califano, Zhang, & Spence, 2024). AI food images, often rated more appetising than real photographs, lose that advantage once people learn where they came from (Califano & Spence, 2024).

So the question for GASTRO-X is what would it take for people to follow AI advice?

Crossmodal nudging

Gastrophysics, the science of how all our senses shape what we taste, has shown that flavour is built by the brain from much more than the tongue. Make the sound of a crisp louder and people rate it as crisper and fresher (Zampini & Spence, 2004).1 People link round shapes with sweetness and angular ones with bitterness. What we see and hear sets expectations about what we are about to eat.

An AI recipe can be correct and still feel wrong if it breaks these expectations. A crossmodal nudge builds the right cues into what an AI assistant shows: the words it uses, the image of the dish, the sound of it cooking. It does not argue or persuade. It makes the suggestion feel closer to real food.

Why a separate website

These notes are about methods, and GASTRO-X deserves more than a single post. The project website at gastro-x.califano.xyz will collect news from the lab, the open tools as they are released (a benchmark for AI recipe generation and the Gastrophysics Prompt Handbook), the data and papers, and, in 2028, the demo itself. It is written for anyone who cooks, not only for researchers.

  1. That study earned Charles Spence and Massimiliano Zampini an Ig Nobel Prize. ↩
  2. For the record, the acronym stands for Generative Artificial intelligence for Sensory Trust and Resource Optimisation via CROSSmodal nudging. Yes, the X is doing a lot of work. ↩

Comments