New artificial intelligence software writes wine and beer reviews said to be indistinguishable from those by human critics. Researchers hope this tool will help wine and beer producers aggregate more reviews, and also provide human reviewers with a template. But the software has a major limitation: it cannot accurately predict the flavor profile of a beverage, something that can only be sampled and described by human taste buds.
Keith Carlson of Dartmouth College, who co-developed the algorithm, said wine and beer are perfect for AI-generated text because of the variety of their descriptors. Fermentation style, growing region, grape or wheat variety and year of production are all mentioned in reviews, aiding the AI algorithm. Reviews also tend to use the same vocabulary — words like “oaky,” “floral” and “dry.”
“People talk about wine in the same way, using the same set of words,” Carlson said, adding “it was just a very unique data set.”
The research team — interdisciplinary scientists from Dartmouth, Indiana University and The Santa Fe Institute — used more than a decade’s worth of reviews (over 125,000) from Wine Enthusiast, and 143,000 beer reviews from the website RateBeer. Results were published in the International Journal of Research in Marketing.
Next, they envision using the software for other experiential products, such as coffee or cars.
Read more (Scientific American)