Not AI for a change

French scientists have developed a wine-tasting machine - a computer able to pinpoint exactly where the grapes in a glass of red were grown. The technology identifies wines from different Bordeaux estates with perfect accuracy, even when vineyards were separated only by a single country road.

The finding supports the long-held belief in “terroir” - the soil, microclimate, vineyard position and grower’s methods that shape a wine’s character. “Our results show that the terroir is expressed in a wine’s chemical composition, which was suspected to be true but had never been established on such a fine spatial scale,” says Alex Pouget, a French computational neuroscientist at the University of Geneva.

By confirming a bottle’s origin, the system could help combat counterfeit fine wines, though it might also enable imitation by revealing the molecular signatures behind great bottles.

The process begins with gas chromatography, which analyses volatile molecules - those that readily turn from liquid to gas. While producers traditionally measured only a few compounds, the new method uses machine learning to examine tens of thousands, exposing subtle, estate-specific patterns invisible to the human palate.

Trained on chromatography data from 73 wines of various vintages from seven Bordeaux estates, the system was tested blind on other wines from the same properties. In repeated trials, it identified the correct estate 100% of the time.

Some compounds, such as dimethyl sulfide (DMS), influence flavour in minute quantities, adding complexity when subtle but unpleasant aromas when concentrated. Pouget believes the technology could help smaller estates optimise blends, narrowing the gap with Bordeaux and Champagne’s most celebrated châteaux.