AI can tell Scottish and American whiskies apart

Artificial intelligence (AI) can be used to tell the difference between American whiskey and Scotch, and identify their prominent aromas, potentially better than a human expert.

More than 40 different compounds with diverse structures determine the aroma profile of whisk(e)y. Chemical interactions of these compounds in the olfactory system make it very challenging to assess or predict the characteristics of a whisk(e)y’s aroma based solely on its molecular composition. Human panels are widely used to evaluate flavours, however, because olfactory perception can be subjective, such evaluation methods require a significant amount of time, money and often, trained panellists.

Researchers in Germany therefore set out to see if machine learning methods could be used to assess the molecular composition of seven American and nine Scotch whiskies. To do this they used two machine learning algorithms; OWSum, a molecular odour predictions algorithm developed by the researchers, and a convolutional neural network.

The team analysed the whisk(e)y samples with GC–MS and automatic compound detection analysis. They also collected sensory data, generated by a human expert panel using an evaluation method called ‘rate-all-that-apply’ (RATA), to determine the top five odour descriptors per sample.

‘Caramel-like’ was the most characteristic odour descriptor for American whiskey, whereas ‘apple-like’, ‘phenolic’ and ‘solvent-like’ odours were more pronounced in Scotch samples. This, the researchers said, ‘show that there are clear relationships between volatile molecules as well as olfaction with the type of whisky’.

They then used the algorithms to find out whether they could predict the smell of a whisk(e)y, represented by the top five odour descriptors and found that they were able to do so with ‘promising accuracy’; OWSum was able to determine whether a sample was American whiskey or Scotch with an accuracy of over 90% and both algorithms were able to identify the five strongest notes of a specific whisk(e)y more accurately and consistently on average than any individual human expert.

‘OWSum not only offers a method to quickly classify whiskies, but also allows us to analyse their ingredients or characteristic features in one step,’ they said.