AI vs. Human Experts in Whisky Distinction
The study, led by Andreas Grasskamp and his team, involved training an AI algorithm, OWSum, to differentiate between nine types of Scotch whisky and seven types of American bourbon or whiskey based on flavour descriptions. These descriptions included keywords like “flowery,” “fruity,” “woody,” and “smoky.” Impressively, OWSum was able to distinguish whiskies from the two countries with almost 94% accuracy, using only these flavour descriptors.
However, the breakthrough came when the AI was fed chemical data from gas chromatography–mass spectrometry (GC-MS), a technique that analyzes the molecular composition of liquids. With a reference dataset of 390 common whisky molecules, the AI’s ability to tell Scotch from American whiskey improved dramatically, reaching 100% accuracy. Specific compounds such as menthol and citronellol were associated with American whiskey, while Scotch was identified by the presence of methyl decanoate and heptanoic acid.
AI vs. Human Performance in Aroma Identification
The research team also compared the AI’s ability to predict the top five aroma descriptors (e.g., floral, fruity) based on chemical data to that of a neural network and human experts. Using a scoring system from 0 to 1 (where 1 signifies perfect accuracy), OWSum scored 0.72, while the neural network achieved 0.78. Human whisky experts, however, scored only 0.57, highlighting the AI's superior consistency and reliability.
Satnam Singh, a team member at Fraunhofer Institute, emphasized that while the task of aroma identification is complex for both humans and machines, AI offers greater consistency, though human experts remain crucial for training the AI.
The Science Behind the AI’s Success
One of the key reasons AI outperformed human experts is its ability to focus on the molecular composition of the whiskies. Whisky’s complex aroma is determined by the presence or absence of a variety of chemical compounds, which the AI can analyze without human sensory bias. The researchers acknowledged that neither the AI nor the neural network took into account the concentration of molecules, which is an area for future improvement that could enhance accuracy even further.
Potential Applications of AI in the Whisky Industry and Beyond
The potential applications of this AI technology extend far beyond distinguishing between different types of whisky. In the whisky industry, AI tools like OWSum could be used for:
- Quality Control: Ensuring the consistency and quality of products during production.
- Product Development: Helping distilleries create new whisky blends by understanding the chemical compositions that make certain whiskies unique.
- Fraud Detection: Identifying counterfeit whiskies by analyzing their chemical signatures.
Moreover, the technology’s use of molecular data makes it applicable to other industries, including food and beverage production and the chemical industry. If it can differentiate whiskies with such precision, there’s potential for AI to be used in the production of any product with a distinct smell or flavour profile, from perfumes to chemicals.
Conclusion
AI has proven itself a powerful tool in the whisky industry, offering superior consistency and accuracy in distinguishing between American whiskey and Scotch whisky based on both flavour descriptions and chemical data. While human experts are still needed to train these systems, AI’s ability to reliably identify and analyze complex aromas opens up exciting possibilities for improving product quality, development, and fraud prevention. As the technology continues to evolve, it may have applications far beyond the whisky industry, revolutionizing how we approach the production and analysis of any aromatic product.
_page-0001.jpg)
0 Comments