Artificial intelligence is revolutionizing the whisky industry by accurately distinguishing between Scotch whisky and American whiskey, relying on data analysis instead of human tasting. Recent research highlights that an AI-powered molecular odour prediction algorithm, developed by a team from Germany, achieved remarkable precision in identifying flavor profiles.
The AI, known as OWSum, was trained using detailed descriptions of various whiskies, and its capabilities were put to the test using 16 different samples—comprising nine Scotch whiskies and seven American bourbons or whiskeys. Impressively, the AI demonstrated nearly 94 percent accuracy in classifying drinks based on key flavor descriptors such as floral, fruity, woody, and smoky.
The complexity of whisky aromas emerges from numerous chemical compounds, prompting researchers to enhance OWSum’s understanding by introducing a reference dataset of 390 molecules typically found in whiskies. When combined with gas chromatography–mass spectrometry data, OWSum achieved a flawless 100 percent accuracy in differentiating between Scotch and American whiskies.
Certain compounds emerged as clear indicators; menthol and citronellol signaled American whiskey, while methyl decanoate and heptanoic acid were markers for Scotch. Furthermore, OWSum’s performance was compared against a neural network and human experts in predicting the top five aroma descriptors based on whisky’s chemical composition. OWSum scored 0.72, while the neural network scored 0.78, significantly outperforming the human experts, who averaged a score of 0.57.
The findings underscore the intricate nature of olfactory perception, highlighting that while machines show greater consistency, human expertise remains essential in training and refining AI systems. Researchers acknowledge that their models currently do not account for the concentration of molecules, aiming to improve accuracy further in future studies.
Grasskamp anticipates that such AI technologies could enhance quality control in distilleries, aid in the creation of new whiskies, and detect counterfeit products. The potential applications extend beyond whisky, with possible uses in food and drink production and the chemical industry, underscoring the versatility of this groundbreaking technology.
Topics:
