Food and wine pairings: artificial intelligence and a sommelier tie

Tuesday January 29 2019 by Vitisphere

- Photo credit : Pierrick Bourgault

On January 22nd, an unprecedented tournament in Paris pitted two international experts, sommelier Paz Levinson and wine critic Antoine Gerbelle, against Matcha, a system modestly referred to as “technologies to help sell wine”.  The first test involved suggesting the best wine for a classic, gourmet or “tricky” dish, whose ingredients make wine pairing extremely complex. The second test required responding to requests such as “I am looking for a bottle of white wine to open with Madame for our tenth anniversary. We enjoy organic wine and our maximum budget is €50”. The third involved designing a wine list for a restaurant based on a text summarising personality, dishes and location. The panel, composed of Michelin-starred chef Alain Dufournier, winegrower Clément Pinard and Paolo Bouca Nova, commercial director of Repaire de Bacchus which provided the wine database, rated the anonymous responses of the two humans and the virtual wine merchant.

When the verdict came in, it showed that in the first two tests, humans beat the “chatbot”, with a slight advantage for Antoine Gerbelle. Matcha's suggestions for complex dishes were quite honourable - except for one, which founder Thomas Dayras has agreed to correct: “I will teach the machine that this pairing is strange, because I agree with you”. For the third round, the panel preferred the wine list designed in just a few seconds by Matcha to the one Paz Levinson took an hour to prepare. The final result came in as a tie.


Copy the code :
© Vitisphere 2021 - Tout droit réservé