A lot of people rely on wine reviews to evaluate wine. The reviews provide us not only with objective information, such as price and alcohol content, but also subjective information, such as the wine’s quality, tasting characteristics, and mouthfeel. Wine reviews can be powerful, persuading us to buy the wine and evoking memories and emotions. You believe and sometimes even feel connected to the reviewers, who are sharing their personal experiences smelling, tasting, and enjoying the wine.
But what if it was artificial intelligence that wrote that wine review you’re reading, and not a person?
Investigators from Dartmouth College have created a software program that can generate wine reviews with just a little bit of data, such as the grape varietal and the rating score it received. The algorithm was trained on 125,000 wine reviews from Wine Enthusiast Magazine. The artificial intelligence used that information to generate its own wine reviews.
The researchers then tested the computer-generated wine reviews against ones written by expert wine reviewers. Most human test subjects could not tell which reviews were machine-generated and which were written by people.
Here are just a few wine reviews created by the software:
· This is one of the best California rosés I’ve ever had. It’s so dry, so rich in fruit, it’s almost like a dessert wine, except it’s totally dry. The flavors are enormously complex, ranging from cherries and raspberries to white chocolate, and the finish goes on and on.
· While demure on the nose, the palate of this off-dry Riesling is chock full of luscious white grapefruit and tangerine flavors. It’s feather light on the palate, yet deeply concentrated, with a long, mineral-tinged finish.
· There are some good flavors of blackberries and cherries in this dry full-bodied wine. It’s a little rough around the edges, but it’s a nice sipper.
· If you’re looking for a dry, crisp white wine, try this one. It’s filled with citrus, peach, wildflower, and vanilla flavors, with a rich, creamy texture and a long, spicy finish.
To compare, here is a wine review from a recent issue of Wine Enthusiast Magazine, which identifies which taster wrote the review:
This study, entitled Complementing human effort in online reviews: A deep learning approach to automatic content generation and review synthesis was recently published in the International Journal of Research in Marketing.
The researchers claim that using artificial intelligence to write wine reviews would help wine reviewers, restaurants, and wineries by creating first drafts, and that the software is not intended to replace humans.
But what if the wine reviewer, restauranteur, or seller doesn’t carefully check that first draft for accuracy? And what’s to stop someone from passing off a computer-generated wine review as a human-written one?
Does this mean that we can no longer rely on wine reviews? The artificial intelligence used to write wine reviews can’t smell or taste what it’s reviewing. How can you trust it?
I always thought that if I couldn’t discern the cola or pomegranate notes that a wine review said were there it was because I didn’t have the most sophisticated palate. But maybe those notes really aren’t there at all?
The study's researchers acknowledge the ethical concern that a consumer may assume that a review is based on the reviewer’s personal experience rather than machine-generated.
If the use of artificial intelligence to write wine reviews becomes more common, I’ll be second-guessing the wine reviews I’m reading. Just who – or what -- is reviewing this wine? And do they know what they’re talking about? It takes a little of the magic out.
We hope that this study on how artificial intelligence can write wine reviews is a bit of an eye opener.
How much do you rely on wine reviews? Would that change if you thought that a wine review was written by a computer? Send us a message at firstname.lastname@example.org. Always feel free to reach out to us with any questions or feedback.
If you enjoyed this post, visit our website and take a look at some of our related articles: