Info Pulse Now

HOMEcorporatetechentertainmentresearchmiscwellnessathletics

A.I. Crushed Experts In A Blind Taste Test Of Identifying Whiskey Flavors Around The World

By Cass Anderson

A.I. Crushed Experts In A Blind Taste Test Of Identifying Whiskey Flavors Around The World

In a recent test, a trained A.I. model was able to identify if a sample was American Whiskey or Scotch Whisky significantly better than human experts. Having been to countless guided whiskey tastings over the years led by some of the industry's most celebrated experts, these findings confirm everything I've suspected since I've entered these circles of experts.

The Fraunhofer Institute for Process Engineering and Packaging IVV's Andreas Grasskamp led this research by training a molecular odour prediction algorithm (called 'OWSum') on the flavor profiles of various whiskeys. This included American whiskeys and Scotch whiskies, the unique notes of each sample, and the identifiable characteristics of the spirits. From there, things got interesting.

Using 16 samples (9 scotch, 7 American whiskey), the team ran a test where the OWSum A.I. model was tasked with identifying where the sample was from after it was fed 390 flavor molecules (menthol, citronella, etc), Scotland or the United States, based on the flavor profiles. According to recent work published in NewScientist, the A.I. model was able to quickly identify distinct flavor characteristics to Scotland and the USA.

According to the findings, the A.I. model was 100% accurate in identifying if a sample was from Scotland or the USA when it was able to use the "gas chromatography-mass spectrometry" data. It found that the aforementioned notes of citronella and menthol were immediate tells that a sample was from the US. Conversely, "methyl decanoate and heptanoic acid" signaled it was a Scotch sample.

At this point, they had OWSum, a trained neural network, and human experts. On the task of predicting the top 5 notes on a sample with '1' being perfect and '0' being worst, the OWSum model achieved a score of 0.72, neural network achieved 0.78, and the human whiskey experts were only able to identify the 5 most prominent flavor characteristics at a score of 0.57.

Sending these findings around to the BroBible team, my colleague Connor mused that his favorite line during guided tastings from so-called whiskey experts is when they ask participants what notes they smell/taste and say "there are no wrong answers." My dude, if there are no wrong answers than what is the point of being an expert? Can we all just make it up by saying whatever comes to mind?

I'm no stranger to winging it during a tasting. When it comes to wine I always taste dark red fruits. When it comes to bourbon the vanilla pops out to me, as does sweet caramel most of the time if it's present. Beyond that, I usually try and keep a list of potential flavor characteristics in front of me so I can cross-reference the potential notes *while* I'm tasting otherwise my brain is simply incapable of pulling those out of thin air.

I've always been told to just spend more time in the grocery store and at the farmer's market walking around smelling things. Your nose/brain need training just like every other tool your body has at its discretion. The whiskey experts among us are doing this regularly. I'd suspect that if you grabbed a regular drinker off the street and put them to these same tests their score would be about 0.12 but obviously more data would be necessary to determine that.

Previous articleNext article

POPULAR CATEGORY

corporate

9808

tech

8831

entertainment

12396

research

5854

misc

13000

wellness

10208

athletics

13170