Whisky decoded: can AI outperform the human nose?

For centuries, whisky has been a drink of history and romance. Now, artificial intelligence is stepping in – not to replace tradition, but to enhance it by decoding flavors with unprecedented precision.

The history and craftsmanship that goes into whiskey is fascinating, dating back to 15th-century Ireland, where monks distilled grain into “aqua vitae” – today known as whiskey. This artistry manifests itself in a lot of complexity that drinkers appreciate when discussing the tones of the concoction.

After prohibition in the United States – 1920-1933 – disrupted the industry, groups began to become more structured, distillation techniques improved and aficionados began having more nuanced discussions about the intricacies of the drink.

Even now, whiskey provokes rich discussions among peers, with particular attention paid to safeguarding traditions while bringing new players to the table – such as Taiwan’s brilliant Kavalan whiskey.

However, limitations come in a few varieties. First, it’s quite common to be standing by a whiskey cabinet or in a bar, being asked what kind you like and simply not knowing – tempting to just say “smoky” for the sake of it. They might pour you a couple to taste, and you end up picking one almost randomly, like going left instead of right for the hell of it.

Whiskey connoisseurship can also come with exclusivity and snobbery, which risks alienating newcomers. Furthermore, too much emphasis is placed on rarity, as if something aged in a sherry barrel from 2022 on the Isle of Lochhobbit adds any worth. True appreciation often comes at the expense of status-seeking.

Limiting whiskey geographically to Scotland, Ireland, and the USA does a disservice to places like Japan, India, or even Kenya, all of which are emerging as important players in the world of whiskey.

Now, new research suggests that AI algorithms offer a deeper and more efficient knowledge, especially when determining whether a drink is of Scottish or American origin.

Profiling whiskey aromas can be particularly challenging due to the complexity of odours. Investing time and money in human experts is difficult, especially when disagreement is often involved. There are five main pillars of whiskey tasting, and I’ll be damned if I can tell the difference between fruity, floral, smoky, spicy, and sweet.

Different parts of Scotland, for example, are famous for distinct flavours, with the west coast specializing in a smoky taste. In a blind taste test, even experts might struggle to pinpoint a whiskey’s location or sensory nuances.

The answer lies in pure chemistry. This research, led by Andreas Grasscamp and his team at the Fraunhofer Institute, was published in Communications Chemistry. They examined the chemical makeup of seven American drinks and nine Scottish ones. Two complex algorithms were used. One algorithm achieved over 90% accuracy in identifying the drink’s country of origin.

What’s even more impressive is how the machine, OWSum, was able to identify certain compounds: menthol and citronellol were characteristic of American whiskey, while methyl decanoate and heptanoic acid were linked to Scotch. American whiskey was also described as caramel-like, whereas Scotch was more medicinal, smoky, and reminiscent of apples.

For distilleries, this means faster, cheaper, and more accurate analysis. As whiskey tasting could arguably be described as a science, reducing variables and errors would make for a tidier process.

The worrying aspect, however, is that it takes away some of the romantic connotations associated with human whiskey tasting. When visiting a distillery in Kentucky, would you prefer a robot pinpointing the caramel notes, or a slightly slurring human telling you a beautiful anecdote about how he met his wife at a whiskey tasting?

But then again, are we at the tiresome debate about whether AI will replace humans in this gig? Maybe they’ll work in tandem and still allow for nuanced discussions.

If two separate bots tell you to invest in Nvidia and not to invest in Nvidia, do you still have a choice anyway?

If a machine tells you there’s licorice to the taste and you disagree, doesn’t that nuance hit the right notes all the same?