Tikalon Blog by Dev Gualtieri

A.I., Wine, and Beer

June 10, 2024

Science and technology have enhanced the safety of our food supply in many ways, such as preservation by freezing and vacuum packing. Food variety has been enhanced, albeit with a significant carbon footprint, by air transport that brings us fresh seasonal fruits out of season. Food chemistry has also given us preservatives and other additives, including flavorings

I've never been a member of the avocado toast crowd, but the avocado is a good example of modern food distribution.

Once harvested, avocados ripen in one to two weeks at room temperature; so, reasonably rapid transport from Latin America to North American destinations is required to ensure freshness. Ethylene gas will hasten ripening, a technique that's common for other fruits.

(Haas avocado, grown in Colombia, Wikimedia Commons image by Iafar)

There's also a dark side to food science that includes the addition of too much sugar, salt and fat to make bland processed foods more palatable; and, often, addictive. More insidious is the use of genetic engineering in terminator technology that makes plant seeds sterile. Fortunately, such things have had considerable opposition.

The Bob New hart Show had a humorous scene involving technological excess in reducing the expense of food processing. In an episode of that television sitcom, Elliot Carlin, a serial entrepreneur and perennial patient of Newhart's psychologist character played by Jack Riley (1935-2016), brings Bob a bottle of whiskey. The whiskey was "Von Krueger's, the only whiskey aged in styrofoam kegs."

In my fifty years of monitoring the scientific literature, I've seen many studies that used scientific instrumentation, mostly gas chromatography, in attempts to find what distinguishes a superb wine from an inexpensive house wine.[1] A typical wine can contain more than 800 different aroma compounds; and, of these, just 10%-20% are known compounds.[1] A study by the Australian Wine Research Institute examined two Australian red wines made from the Syrah grape, a dark-skinned variety grown throughout the world.[1]

The object of this research was an attempt to synthetically recreate the aroma of these wines.[1] It was found that only about 50 aroma compounds could be detectable by smell, and only about 28 were present in large enough quantities to theoretically make a difference to the aroma.[1] However, it was found that a minimum of 44 compounds were necessary to approximate the smell, and that certain compounds had a major affect on the final aroma.[1]

Satyrs harvesting grapes in a scene from an Attic amphora. wine has been with us for at least three millennia. Homer's Iliad describes its use during the Trojan War, thought to have occurred circa 1194-1184 BC and confirmed by archaeological evidence. (Wikimedia Commons image by British draughtsman, Dugald Sutherland MacColl (1859-1948), and contained in Greek Vase Paintings by Jane Ellen Harrison and D.S. MacColl, T.F. Unwin Publisher (London: 1894).Marcus Cyron from a sketch by )

Some chefs think that "Everything is better with butter," a sentiment with which JJulia Child (1912-2004) concurred. Today's software mantra is, "Everything is better with AI (artificial intelligence)." It's therefore not surprising that artificial intelligence has been applied to identification of wine quality through determination of where the grapes were grown.[2-3] This research was done by scientists from the University of Geneva and the Institute of Vine and Wine Science at the University of Bordeaux (Nouvelle-Aquitaine, France).[2-3] This was no easy task, since minor differences in chemical composition will affect a wine's flavor.[3] Multiple factors, such as the soil characteristics, climate, the variety of grape, microbiology of fermentation, and the wine-maker's practices, will all affect the wine's flavor.[2]

This research was the first time that the exact origin of a wine was determined solely based on its chemical makeup.[3] To do this, the team applied gas chromatography and electron ionization mass spectrometry to the problem of finding Bordeaux wine chemical identities and vintages (harvest year).[2] They analyzed chromatograms from 80 red wines from 12 vintages in the years 1990-2007 across seven Bordeaux estates.[2-3] For data analysis, the researchers used the artificial intelligence technique of machine learning in which algorithms learn to recognize patterns in data to process the chromatography data for each wine, which could include up to 30,000 data points.[3]

It was found that the estate could be identified perfectly, and the vintage with up to 50% accuracy.[2] It appears that the chemical identity of a wine is not defined by just a few chemical compounds, but is a function of its full chemical complement.[2] What's significant is that the method analyzed the entire chromatogram, background noise included, and let the machine learning algorithm find the most informative parts of a chromatogram.[2-3] Each estate possessed an unique chemical signature, and the method would be an effective way to combat wine counterfeiting.[3]

A study similar to that for wine was done for beer.[4-5] A research group of Belgian scientists comprised of members from Vlaams Instituut voor Biotechnologie (Leuven, Belgium), Katholieke Universiteit Leuven (Leuven, Belgium), Leuven Institute for Beer Research (Leuven, Belgium), and AB InBev SA/NV (Leuven, Belgium) have developed models using artificial intelligence to predict how a particular beer will be rated by consumers, and what aroma compounds can be added to improve it.[5]

In the study, they combined chemical analyses and taste-testing data of a set of different commercial beers to predict taste, smell, and mouthfeel from compound concentrations.[4] Beer is one product that can be improved by modification, since it has thousands of flavor compounds derived by a mix of raw materials consisting of malt, yeast, hops, water and spices, and their evolution through biochemical conversions during brewing consisting of kilning, mashing, boiling, fermentation, maturation and aging.[4] The research study began with fewer than a hundred beers, but it was expanded to 250 commercial beers over 22 beer styles, for which more than 200 chemical properties were characterized and ranked by a sixteen person testing panel combined with more than 80,000 public consumer reviews.[4-5] The study took five years to complete.[5]

Light beer be damned! These correlations show a definite preference for ethanol and caloric value in beer. The Spearman Rho correlation coefficient is graphed. (Portions of fig. 4c of ref. 4, distributed under a Creative Commons Attribution 4.0 International License.[4] Click for larger image.)

It was found that an algorithm called gradient boosting led to accurate prediction of taste preference from chemical profiles.[4] Reverse engineering of the models identified specific and unexpected compounds as important to beer flavor and appreciation; and, adding these compounds to both alcoholic and non-alcoholic beers led to a greater consumer appreciation.[4] This modeling is a basis for quality control, including detection of food spoilage, product fingerprinting, and counterfeit detection.[4] The research team's future goal is making better alcohol-free beer.[5]

A.I., Wine, and Beer

References: