There was a moment when an algorithm recommended a wine I never would have picked myself. A Savagnin from the Jura. Oxidative, headstrong, completely outside anything I was used to.
It was exactly right.
The question I can’t shake
Taste lives in memory. In the smell of a kitchen on a Sunday morning, or in the first glass that truly clicked. It’s personal in a way that’s hard to put into words, but that you recognize the instant it’s there.
A machine understands none of that. But it does something else, something that both draws me in and unsettles me. It spots patterns I can’t see myself. It connects flavors straight across categories, and it carries no preconceptions about what something is supposed to taste like.
What I’ve learned to see
The recommendations keep getting better. The stories keep getting worse. An algorithm can tell me I’ll probably enjoy the Savagnin based on my profile. But it can’t tell me about Jacques Puffeney out plucking his grapes at first light, while the mist still hangs over the hills.
That isn’t just sentimentality. It’s a loss of information.
Here is what does work: human expertise paired with machine pattern recognition. Not one or the other. The combination turns out to be more interesting than either on its own.
And here is what almost no one asks: which wines actually went into the training data? Which regions, which producers, which price brackets? The algorithms aren’t neutral. They reflect the choices of the people who built them. That makes taste, whether we like it or not, a political question too.
Back to the Savagnin
I drank it while writing this piece. The algorithm was right. It was headstrong, complex, and a perfect fit for what I needed in that moment.
But I still don’t know exactly why it clicked. And maybe that not-knowing is the most beautiful thing about wine.
This is the first part in a series on technology and taste. The next installment asks whether you can understand Burgundy without ever having been there.