Wine
and AI

Last May, news broke that Artificial Intelligence or AI was used to write a convincing wine review.  The news, reported in Scientific American and Food & Wine magazine quoted two reviews, written using “deep learning” algorithms, went thus:

“This is a sound Cabernet. It’s very dry and a little thin in blackberry fruit, which accentuates the acidity and tannins. Drink up.” 

Another said that the wine was “Pretty dark for a rosé, and full-bodied, with cherry, raspberry, vanilla and spice flavors. It’s dry with good acidity.”

If you read this, would you believe that it was a human review? I would.

I am caught in a dilemma.  One the one hand, I subscribe to the growing antipathy in wine circles over the pretentiousness of wine descriptions, over the esoteric terms popularized by Robert Parker.  You know the kind I mean? Descriptions that talk about cat’s urine, tar, wet leather, and the insides of a man’s shoe (not a woman’s shoe but the more robust version that comes from a male chromosome) in an attempt to illuminate a Northern Rhone.  And such overwrought prose is supposed to entice you to sip the said bottle.

Robert Parker, the revered and reviled American wine critic is often blamed for these long and often meaningless descriptions.  His descriptions include words like “a sweet nose of creosote, asphalt (has anyone smelled creosote and if so, what is it?)”  Other Parker classic is to say that a wine smells like sweaty saddles, rubber, a cigar box, pencil lead, sea spray or an array of berries. Such overwrought descriptions may be specific but are useless because they don’t aid olfactory memory.

Ever since Ann Noble, a professor emeritus at the University of California, Davis, came up with an “aroma wheel” to describe the flavors of wine, people have been using it to entice people to buy wine. Today, people are using Artificial Intelligence to do the same.

So what is AI? Whether we know it or not, whether we like it or not, AI is already part of our lives.  AI is how Alexa or Siri work to understand the context of our lives.  AI or deep machine learning is how we get shopping prompts in our Facebook or Instagram feed.  And AI could be used to customize our wine preferences.  At least that is what the wine industry is hoping.  Consider these scenarios.  An AI robot roaming the aisles of your wine store and asking you what tastes you like.  You input “spicy” or “fruity” and voila, the robot churns out a list of wines that you might like.  Similar, AI-based apps can predict what type of wines you gravitate towards and suggest “similar” bottles.  AI can be used to power drones over vineyards which will tell the vineyard owner exactly which vines need to be pruned or watered.  But AI as a wine critic? Now that is a hard pill to swallow. The problem is not that AI cannot come up with descriptions.  In fact, the opposite.  The “hive mind” of the Internet can trawl through the internet and come up with all the adjectives that apply to a particular varietal.  Is that good or bad? Well, let’s take it one step at a time.

What are the words you use to describe wine? Grippy is one I use to describe the sandpaper like taste.  Austere is one that seems structured, even too structured– and I use this mostly for white Chablis wines. Flamboyant is a California Cab.  

Some wine descriptions make sense. You drink enough Zinfandel and you will taste the thick, viscous, fruity taste that is often described as “jammy,” by aficionados. Australian shiraz is indeed spicy and peppery. And New Zealand’s Sauvignon Blanc does have that herbaceous flavour that reminds one of cut grass.  The Syrah in its birthplace, France, does not have the spicy and peppery flavour of an Australian Shiraz.  Minerally wines remind Indians of the water we drink from copper pots.

Some descriptions just don’t make sense to certain cultures. What does “chalky” taste like? Do you have to lick chalk to figure this out? Some descriptions try to be overly helpful by listing a wide range of berries that the wine is supposed to taste like. Having never tasted a linden berry or even a raspberry in its natural just-picked state, my palate has no clue how to process this information. Why not use people-friendly adjectives? A young Burgundy for instance can be tight and surly, like a glowering teenager that needs to mature.  

The problem is that wine descriptions tend to have a formula, which is why it is easy for AI software to replicate.  Consider this bit, written by a piece of software: “While the nose is a bit closed, the palate of this off-dry Riesling is chock full of juicy white grapefruit and tangerine flavors. It’s not a deeply concentrated wine, but it’s balanced neatly by a strike of lemon-lime acidity that lingers on the finish.” Pretty good, isn’t it? So how did the AI do this?

According to the study description, the algorithm trawled through a decade worth of reviews from Wine Enthusiast magazine– about 125,000 in total.  This helped it learn the general structure, tone and style of a reivew.  Then the study-authors gave the algorithm some basic information: winery name, style, alcohol percentage and price point.  The algorithm then searched for existing reviews using this basic information, and collated everything together.  It came up with the most frequently used adjectives for a particular wine and strung them together using the structure-formula.

The study authors say that such algorithymic descriptions will help small wineries come up with descriptions of their products, particularly if they have no time to write it themselves or have no knowledge of English.

As for the rest of us, we had better be careful when reading description on wine labels, websites or product reviews.  Who knows? It may have been generated by a machine.  As one study author said, humans are incredibly easy to manipulate.



Subscribe to my newsletter