Wine Tasting: Man Against Machine

First, there was Kasparov against Deep Blue. A test of whether the subtlety and complexity of human thought can outperform the sheer speed and parallel processing power of silicon. As a game based in mathematics, or at least mappable using mathematical principles, it’s not surprising that eventually we got to the point where the machine can trounce or at least draw the grand master consistently. But wine tasting? Intuition would say that the complexities of flavor and quality would be much harder to predict via computer, and certainly humans would have an edge.

A group of wine experts was put up against a recently developed prototype wine analysis machine from Aalborg University, Denmark in a competition to judge the region of origin, the rough quality level, and the price of the wine. The results were surprising. The machine came very close (we are not told how close) to the experts when it came to identifying the region and judging the overall quality of the wine (though we are also not told what measure of quality was used). Interestingly where the machine fell short was in the prediction of pricing. The experts (we are also not told who they were) apparently weren’t all that good at predicting the price of the wine either, but they were better than the machine.

This sort of thing raises the hackles of many a wine lover. The notion that something as subjective as wine quality can be measured by machine leads to the implication that it might be able to be engineered by machine as well. In this potential many people see the devaluation of the role of winemaker, and the industrial evisceration of something magical.

I certainly understand and to a certain extent share this point of view. When great wines start getting made by computer, we will have lost something special. On the other hand, I’m totally fascinated by the idea that this might be possible. Call it the inner mad scientist, but if we can figure out quantum mechanics and string theory, it seems pretty likely that we might be able to one day fully understand and model what makes for taste.

Read the full (but short) article.