Prestige or plonk? AI develops a nose to sniff out fake vintage wines

Wines on shelves in wine shop
L'Intendant, one of the most prestigious wine shops in Bordeaux - DAVID SILVERMAN/GETTY IMAGES

A red wine’s provenance is key to its value, taste and prestige, and now a machine may be able to protect the finer variety from pale imitations.

Scientists have found a standard piece of chemistry equipment and a machine-learning algorithm can take any wine and show, with total accuracy, which estate made it.

A trial on 12 vintages from seven different Bordeaux chateaux found gas chromatography was able to perfectly determine the geographical origin of a glass of French red.

Dr Alexandre Pouget, the co-author of the study from the University of Geneva, told The Telegraph that researchers had long tried to use gas chromatography to find a “fingerprint” for wines.

However, previous attempts had focused on a handful of specific chemicals and were unable to accurately discern between vintages, grapes and regions.

The new research, published in Communications Chemistry, took a different approach and used the entire, unfiltered chromatograph.

This “noisy” data is normally undesirable but when it was fed into a machine learning algorithm, the technology was able to process the complex information.

Elite wine makers took part

The seven estates that provided wine for the study remained anonymous, but all are members of the “grands crus classés” that marks them as elite wine makers.

While the system was able to tell with 100 per cent accuracy which estate grew a wine, it scored only 50 per cent at identifying the specific vintage.

It is hoped the technology could be used to detect fraudulent wines and improve the overall standard of wine-making.

“People who work with gas chromatography usually focus on a few peaks in the chromatogram because they have reasons to believe they are important,” Dr Pouget told The Telegraph.

“Our statistical technique makes no such a priori assumption. It just learned which part of the chromatogram matters, for instance, for identifying estate identity.”

Different vintages of wine from the same chateau cluster together on a graph, he added, which can act as a chemical hallmark created by a combination of the grape, soil chemicals, and microclimate.

The data from the trial is now held in a database run by the Institute du vin et de la vigne (ISVV) in Bordeaux, and the scientists hope to test the computer system against a professional wine taster to see which is better.

But they say it could cost up to £1.6 million to conduct such a trial in a scientifically valid way, and they hope to find a wine-loving billionaire keen to fund it.

End to false advertising

The work offers hope that cheap wines could no longer be able to get away with falsely advertising as a more illustrious brand.

But the technique could also be used in crafting wines and improving the overall standard. Assemblage, for example, the blending of certain grapes make the best wine possible, could become cheaper and more accessible.

“This is central to wine-making in Champagne and Bordeaux and it is an art,” Dr Pouget said. “The biggest chateaux pay big bucks to hire the best oenologists in the business. Having an app like ours could help other less wealthy estates to produce optimal blends.

“We found that what defines a wine is not a few chemicals but the relative concentration across the entire chemical spectrum. That should be music to the ears of the winemakers.

“You can’t produce a top notch wine by simply adding a few chemicals here and there. You have to somehow extract the best of your terroir through your wine making technique and savoir faire.

“Our study takes a little of the magic away but there is plenty left!”

Broaden your horizons with award-winning British journalism. Try The Telegraph free for 1 month, then enjoy 1 year for just $9 with our US-exclusive offer.