Recently, a research team from the Fraunhofer Institute for Process Engineering and Packaging in Germany developed an artificial intelligence molecular odor prediction algorithm called OWSum, which successfully distinguishes between American whiskey and Scotch whiskey, achieving accuracy that surpasses human experts. The team trained this AI tool using flavor descriptions and chemical data of whiskey to explore its potential in whiskey identification.
Image Source Note: Image generated by AI, image licensed from Midjourney
In the study, researchers selected 16 samples, including 9 Scotch whiskeys and 7 American bourbons. OWSum analyzed keyword descriptions of these whiskeys, such as floral, fruity, woody, and smoky notes, and was able to distinguish between the two types of whiskey with an accuracy of nearly 94%. As the research progressed, the team further provided the AI with a reference dataset containing 390 common whiskey molecules, combined with results from gas chromatography-mass spectrometry analysis, which improved OWSum's distinction accuracy to 100%.
Through this data, researchers found that certain compounds like menthol and citronellol are distinctive features of American whiskey, while methyl decanoate and heptanoic acid are more commonly found in Scotch whiskey. Additionally, the research team tested OWSum against a neural network in predicting key whiskey aroma keywords based on chemical composition. In this test, OWSum scored 0.72, while the neural network scored 0.78, and human experts scored only 0.57. This indicates that while AI performs exceptionally well in such tasks, the complexity of distinguishing whiskey remains a challenge for humans.
Research member Sartan Singh noted that although machines perform more consistently, humans still play a crucial role in training the machines. In the future, the research team hopes to improve the model to consider the concentration of compounds, thereby further enhancing accuracy. Grasskamp mentioned that such AI tools can not only be used for quality control in distilleries but also help in developing new whiskeys and identifying counterfeit products. Furthermore, this technology has the potential to be applied in other fields such as food and beverage production and the chemical industry.
Paper: https://www.nature.com/articles/s42004-024-01373-2
Key Points:
🍸 The AI OWSum distinguishes American whiskey from Scotch whiskey with nearly 100% accuracy, surpassing human expert performance.
🔬 This AI identifies specific compounds to differentiate the two types of whiskey by analyzing aroma keywords and chemical components.
👨🔬 Humans remain important in training machines, with future aspirations to improve model accuracy for more complex tasks.
