AI outperforms human tasters in wine identification challenge

On March 15, an artificial intelligence named Deep Red, developed by the Lyon-based company M&Wine, competed against seasoned wine tasters in a blind tasting event. The challenge was to identify the country of origin, region, and grape variety of twelve wines selected by a court-appointed official. The results were clear, but not without some nuances.

Deep Red successfully identified the country of origin for all twelve wines, while the human tasters averaged six correct answers. For the region and grape variety, the AI correctly identified ten out of twelve, whereas the tasters averaged three correct answers for these categories. Philippe de Cantenac, the organizer of the French blind tasting championship, acknowledged the AI's success in processing and recalling data from its database. However, he noted that the wines were selected under conditions favorable to Deep Red, involving single-variety wines from recent vintages, as the AI can only analyze wines produced from 2020 onwards.

Despite its impressive performance, Deep Red has limitations. It can only recognize wines included in its database, which currently holds 35,000 references. This limitation was evident when Deep Red failed to identify a wine from Jura as poulsard, a grape variety underrepresented in its database. M&Wine plans to expand the database to include older wines and more diverse references. Co-founder Théodore Tillement emphasized the importance of contributions from wineries and cooperatives to enrich the database, allowing the AI to learn more about different wines.

M&Wine views Deep Red as a demonstration of AI's potential to decode the mineral signatures in wines, which remain stable over time and serve as a unique fingerprint for each wine. The company aims to offer this technology to the wine industry, highlighting the valuable information that can be extracted from a wine's mineral profile.

Philippe de Cantenac, however, cautioned against drawing hasty conclusions from the experiment. He pointed out that while AI can process and recall vast amounts of data, it cannot replicate the sensory analysis and enjoyment that human tasters experience. The goal of human tasters is to find pleasure in the wine, appreciating its uniqueness and evolution, which AI cannot do. Despite this, de Cantenac did not rule out the possibility of organizing another blind tasting with AI, but under different conditions, suggesting that the results might differ significantly.