Tasting a sample and naming what is in it · Contested (since 2023)

Osmo and the Monell Chemical Senses Center publish a principal odor map on 31 August 2023: on odorants never smelled before, the model's description lands nearer the mean of a trained panel than the median panellist does, for 53 percent of the molecules.

The map does not smell anything. It reads a molecular structure and predicts the words a trained panel would reach for, which on molecules nobody had described before lands nearer the panel's average than its median member does, more than half the time. Chemistry works the same seam from the other end: gas chromatograms of eighty Bordeaux reds give up the estate in ninety nine cases out of a hundred, and the vintage half the time, from bottles no instrument ever tasted. What holds the line is the ruler. Both results are scored against a panel of people, because nothing else says what a thing smells of, and the day the panel disbands the model has nothing left to be right about. Identifying is also the smaller half of the job. A cupper at the table is deciding whether to buy the lot, and that decision sits under taste as a commitment.