Artificial intelligence could potentially outperform humans in selecting a single malt, as machine learning algorithms demonstrated superior ability in detecting the main scents of various whiskeys compared to an expert, as per a study released on Thursday.
Most scents in our environment consist of a complicated blend of molecules that react in our sense of smell to produce a distinct perception.
Whisky, with over 40 compounds influencing its aroma, can also have numerous non-fragrant volatile compounds.
Evaluating or perceiving a whiskey’s aromatic characteristics solely based on its molecular composition can be quite challenging.
The chemicals have achieved this through two machine learning algorithms, as shown in a study published in Communications Chemistry on Thursday.
OWSum is a statistical algorithm created by the study authors to detect molecular scents.
CNN is a convolutional neural network that assists in identifying connections in highly intricate data sets, like those between key molecules and aroma characteristics in blended whiskey, as stated by AFP Andreas Grasskamp, a researcher at the Fraunhofer Institute for Process and Packaging IVV in Germany and the lead author of the research.
The researchers trained the algorithms by giving them a set of molecules identified through gas chromatography and mass spectrometry in 16 whiskey samples, including Talisker Isle of Skye Malt (10 years old), Glenmorangie Original, Four Roses Single Barrel, Johnnie Walker Red Label, and Jack Daniel’s.
Taste characteristics specific to each sample were provided by a panel of 11 experts.
The algorithms were subsequently employed to determine the nation where each whiskey originated from and its top five prominent flavors.
Detection of forgery
OWSum could distinguish between American and Scottish whiskeys with over 90% precision.
The identification of substances like menthol and cytoronelol was closely linked to categorizing the whisky as American, whereas detecting metyl decanoate and heptanoic acid was primarily linked to Scottish whiskies.
The algorithm found that caramelized notes are most distinctive in American whiskeys, whereas “maçã”, “diluent”, and “phenolic” notes are most distinctive in Scottish whiskeys.
The researchers later requested OWSum and CNN to forecast the scent characteristics of whisky by analyzing the identified molecules or their structural features.
Both algorithms outperformed human panel experts in accurately and consistently identifying the five main notes of a whiskey.
“Our algorithms were more in agreement with the panel outcomes than each individual member, resulting in a more accurate assessment of the overall odor perception,” Grasskamp stated.
Machine learning techniques could be applied to identify counterfeits or assess if the blended whiskey will possess the anticipated scent, ultimately cutting expenses by lessening the requirement for assessment panels.
It is theoretically possible to achieve comparable outcomes with wine by providing a list of detected compounds in the sample along with their descriptors, as stated by Grasskamp.
Determining if the wine aromas are distinctive enough for an AI algorithm is a challenge that lies in the subtlest details.