Artificial Intelligence and Sensory Evaluation

Can AI be expected to take over Flavour Assessment?

Last month I was consulting the web-site of a leading brand of malt whisky to find out about a new expression and was appalled by the shallowness of the tasting note. We whisky writers are all too familiar with the pathetic attempts of PR agencies to describe the character and flavour of the products they are hired to promote without resorting to hyperbole, but this example was of no help whatsoever.

I typed the brand-name into ChatGPT and found an almost identical description – of course, the whisky was so recent that the only tasting note available to AI was the one put out by the company’s PR agency. Unless it was the other way round, and the agency had resorted to AI to do their job for them…

I then tested the robot with a few other whiskies I knew well. The results were mixed – from downright inaccurate to thorough and exhaustive. Dependent, of course, on the number and quality of the tasting notes available to the machine, they were generally helpful. But can/will they ‘take over’?

Sensory chemists discern two broad approaches to sensory evaluation, ‘objective’ and ‘subjective’.

The first is analytical: it sets out to describe only ‘what is there’, limiting, as far as possible, the interpretative faculties of the assessor. This is the kind of analysis employed in laboratories and is based on the fact that smells have an objective existence in volatile odour-bearing molecules which can be assessed either by scientific instruments such as Liquid Gas Chromatographs or organoleptically by what is sometimes referred to as ‘the human instrument’.

Assessors are highly trained; assessments take place under lab conditions; the vocabulary used, generally founded in chemistry, is limited and for ‘internal’ use only, to assess quality and consistency. Although the vocabulary used might vary from company to company, the key factor is that all the assessors are ‘singing from the same hymn-sheet’. While such an objective approach might well be usefully adopted by AI it is of doubtful use since much of the language is likely to be incomprehensible to consumers.

By contrast, subjective analysis gives free rein to the experience, imagination and recollections of the individual. The language is descriptive and figurative; it abounds with similes and metaphors; it can employ abstract terms; it can be pejorative without fear of censure; it may be rhetorical and allusive, although the effectiveness of the rhetoric depends upon the audience understanding the allusion. It is referred to as ‘hedonic’, as in ‘the pursuit of pleasure’.

How we react to smell and taste is deeply personal, since it is directly influenced by memory. Smells are registered by that part of our brains which also the seat of long-term memory and emotion. But our subjective reaction to smell and taste is also influenced by factors such as mood, situation, occasion, company, perhaps background music, maybe what you ate last night… Furthermore, aromas change in the glass as the liquid is exposed to the air and allowed to ‘breathe’. This is called ‘aromatic development’.

So many variables. In truth, there is no such thing as a definitive tasting note, and although experience heightens awareness and expands vocabulary, any descriptor is valid. Practice does not make perfect, although it helps!

Being strictly objective/analytical, AI programmes can’t deliver subjective/hedonic assessments. Since ‘whisky tasting’ is ultimately about enjoyment and enjoyment cannot be quantified, who needs the help of AI?

Robert Burns would reply: “Tak aff yer dram”!