Could artificial intelligence surpass humans in the art of selecting a single malt? According to a study published on Thursday, machine learning algorithms have proven to better predict the dominant aromas of various whiskies than human experts. In our environment, most odors consist of a complex mixture of molecules that interact within our olfactory system to create a specific impression. Whisky's aromatic profile can be determined from over 40 compounds, making it challenging to evaluate or predict its aromatic characteristics based solely on molecular composition.
Chemists have achieved this feat using two machine learning algorithms, as reported in a study published in Communications Chemistry. The first algorithm, OWSum, is a statistical tool designed to predict molecular odors, while the second, CNN, is a convolutional neural network that helps uncover relationships within complex datasets. Researchers trained these algorithms with a list of molecules identified in 16 whisky samples, including brands like Talisker and Glenmorangie, along with aroma descriptors provided by a panel of 11 experts.
The algorithms were then utilized to determine the country of origin of each whisky and its five dominant aromas. OWSum successfully identified whether a whisky was American or Scottish with over 90% accuracy. The study suggests that these machine learning methods could be used to detect counterfeit whiskies or evaluate whether a blend will have the expected aroma, potentially reducing costs by limiting the need for human tasting panels.
