HomeThe New IntelligenceAI Beats Human Experts at Identifying Scotch vs American Whiskey

AI Beats Human Experts at Identifying Scotch vs American Whiskey

German researchers built an algorithm that identifies Scotch from American whiskey better than trained experts – without tasting a drop.

AI can identify whisky better than human experts 1AI and computer scienceGerman researchers found that artificial intelligence is better at correctly identifying whisky. (Science Reader)
German researchers found that artificial intelligence is better at correctly identifying whisky. (Science Reader)
Share
The New Intelligence · Explore this series
December 20, 2024
Key Takeaways
  • An AI algorithm identifies Scotch from American whiskey with 100% accuracy.
  • The OWSum system analyzes 390 molecular compounds, not taste or smell.
  • Human whisky experts scored 0.57 versus the AI's 0.72 on aroma prediction.

AI achieves perfect accuracy identifying whiskey origins

Researchers at the Fraunhofer Institute for Process Engineering and Packaging in Germany have developed an AI system that can distinguish between Scotch whisky and American whiskey with remarkable precision.

The algorithm, called OWSum (Olfactory Weighted Sum), achieved 100% accuracy when analyzing chemical composition data - significantly outperforming human whisky experts.

Key figure

100%

Accuracy identifying whiskey origin from chemical data

The team, led by Andreas Grasskamp, tested OWSum on 16 whisky samples: nine Scotch whiskies and seven American bourbons or whiskeys. When given only flavor keyword descriptions like "flowery," "fruity," "woody," or "smoky," the AI correctly identified each whisky's country of origin with 94% accuracy.

But the real breakthrough came when researchers fed OWSum chemical data from gas chromatography-mass spectrometry analysis. With access to information about which of 390 molecules were present in each sample, the system achieved perfect identification - correctly classifying every whisky as American or Scotch.

What is gas chromatography-mass spectrometry?

Gas chromatography-mass spectrometry (GC-MS) is a laboratory technique that separates and identifies the individual chemical compounds in a mixture. A sample is vaporised and passed through a column that separates its components by how quickly they travel; the mass spectrometer then identifies each one by its molecular weight. In food and drink science, GC-MS reveals the precise chemical fingerprint of a product without anyone having to taste or smell it.

Chemical fingerprints reveal whiskey identity

The AI identified specific compounds that serve as telltale markers for each whiskey type. Menthol and citronellol were dead giveaways for American whiskey, while methyl decanoate and heptanoic acid pointed reliably to Scotch origin. These compounds appeared consistently in one category but never in the other across all samples tested.

The research, published in Communications Chemistry, also tested whether AI could predict a whisky's dominant aromas based purely on its chemical profile.

On a scale from 0 (completely wrong) to 1 (perfect accuracy), OWSum scored 0.72 when predicting the top five odor descriptors for each sample. A neural network performed slightly better at 0.78.

Human whisky experts? They managed only 0.57.

[The results] underline the fact that it's a complicated task for humans, but it's also a complicated task for machines - but machines are more consistent than humans

Satnam Singh, lead author, Fraunhofer Institute

Practical applications beyond the tasting room

The researchers suggest their AI tools could transform quality control in distilleries, help develop new whisky varieties, and detect fraudulent products.

Because the system analyzes chemical signatures rather than actually tasting the spirits, it provides objective, reproducible results that don't vary based on the taster's mood, health, or training.

Current limitations include the system's focus on molecular presence rather than concentration - it knows what's there but not how much. The team hopes future versions will incorporate quantity data for even greater accuracy.

Beyond whisky, Grasskamp notes the technology could apply to "anything that smells" - from food and beverage production to chemical manufacturing and quality assurance across industries.


Sources

  • Original research: Singh, S., Schicker, D., Haug, H., Sauerwald, T. & Grasskamp, A.T. (2024). Odor prediction of whiskies based on their molecular composition. Communications Chemistry, 7, 293. https://doi.org/10.1038/s42004-024-01373-2

Fact Check: Claim-by-Claim Verification Verified

The recap closely tracks the Communications Chemistry paper and New Scientist coverage, with only minor simplifications and no substantive misrepresentations.

1 Verified
The study used 16 whiskies (9 Scotch, 7 American) analyzed by GC–MS and evaluated by 11 experienced panelists, as described in the Communications Chemistry paper and press materials
2 Verified
The OWSum algorithm achieved about 94% accuracy using top-5 odor descriptors and 100% accuracy using molecular (GC–MS) features to classify samples as American vs Scotch in leave-one-out tests
3 Verified
Menthol and citronellol were key markers pushing classification toward American whiskey, while methyl decanoate and heptanoic acid were associated with Scotch in the model’s feature importance analysis
4 Verified
Machine-learning models (OWSum and a CNN) predicted top odor descriptors with performance measures that matched or exceeded inter-panelist agreement, effectively outperforming individual human experts

Commentary

  • The “perfect accuracy” claim is correct but applies only to this small dataset and cross-validation setting; the paper explicitly cautions that results may not generalize to all Scotch and American whiskies.
  • The recap rounds the reported performance metrics (e.g., F1, MCC, ROCAUC summarized as a single 0.72 vs 0.78 number), which is acceptable for popular coverage but glosses over that multiple metrics were used.
  • The description that certain molecules “never” appear in the other category reflects what was observed in this sample set; the original authors frame these as characteristic markers under dataset limitations rather than universal rules.
  • The practical applications (quality control, fraud detection, broader use for “anything that smells”) are in line with how the authors and institutional press releases describe potential uses, but they remain prospective rather than demonstrated at industrial scale.

Sources used for verification

Share
Related Articles
AI In Science Connects the Dots, But Only In Fields That Are Fragmented

An analysis of 80 million papers shows AI boosts originality where knowledge is scattered and connections are weak, but contributes little novelty in structured science.

"Keep Humanity Safe From AI," Urges Pope Leo XIV

Pope Leo XIV's first encyclical reaches the same verdict on AI as the labs building it, then parts ways over the meaning of human limits.

AI Solves Erdős Math Problem: What's Next for AI in Mathematics?

An AI solved an 80-year-old Erdős math problem by walking a path mathematicians had collectively avoided.

Is AI Making You Dumber? Not If You Challenge It

Cognitive debt is the cost of letting AI think for you. New research shows the difference between healthy and harmful AI use comes down to one habit.