• Enter Now

Insights

Machine Intelligence and the Future of the Wine Industry

Photo for: Machine Intelligence and the Future of the Wine Industry

13/02/2023 Wine writer Charlie Leary explores the strengths and weaknesses of machine intelligence for the wine industry.

ChatGPT said California Riesling wines competed against French Bourgogne wines at the Judgment of Paris in 2002. Wine writer Charlie Leary explores the strengths and weaknesses of machine intelligence for the wine industry.

Machine intelligence also called Artificial Intelligence (AI), and a few other names, has recently captured the attention of the wine industry, mostly in the form of chatGPT. This online program will answer questions, write texts, and even create spreadsheets. The program was created by OpenAI, and funded to the tune of billions of dollars by Microsoft. Then, Google introduced Bard, a new chatbot whose underlying intelligence technology will become part of Google.com. The next day, Microsoft Corp. said it was integrating a cousin of ChatGPT into Bing, its own search engine. 

In January, wine expert and journalist Jane Anson published an article by technology expert Simon Pavitt on chatGPT, suggesting it could cause “an extinction event” for wine writers and sommeliers. In fact, most of the article was written by chatGPT, not Pavitt. It caused a sizable reaction among wine industry commentators. One suggestion was “The ability of ChatGPT to generate detailed and nuanced descriptions of wine can be used to create effective and compelling marketing materials, both online and in print. This can be particularly useful for small and medium-sized wineries that may not have the resources to employ a full-time sommelier or wine expert.”

To anyone who hasn’t tried out the new technology, you should do so, including asking questions about wine. This article explores where it may have beneficial applications in the wine business, writ large, and where machine intelligence seems to have some weaknesses and pitfalls. It will probably not decrease demand for sommeliers, who actually taste wine (something machine intelligence still cannot do), and wine writers and marketers can certainly use it for research, but carefully. 

Potential Uses in the Wine Trade

Machine intelligence programs must be trained on corpora of knowledge, that is, sets of information that form the basis for its answers and capacities. It can also learn from its mistakes and will reformulate answers if challenged or you point out something is incorrect. But you must have the wine knowledge to challenge it. Nicholas Thompson, the CEO of The Atlantic has pointed out that chat programs replacing search engines that then send you to authoritative websites or other information sources could be a problem. 

With machine intelligence, the authority becomes chat GPT or another similar program, which potentially damages the integrity, openness, diversity, and overall “health” of the Internet. For example, instead of asking a question about 2022 wine sales and being sent to an article by a wine expert on a reliable third-party wine industry website, you get the purported answer straight from the machine intelligence program, and they make mistakes, as seen below. Thompson also points out that actual costs—chatGPT is currently in a free trial mode—could be substantial at scale. And someone will have to pay. The technology will be commercialized. 

Machine intelligence can greatly increase efficiency. It can create wine-tasting notes, suggest food pairings, and write enough detailed text to provide a wine region profile. Thus far, however, the information is drawn from general knowledge and there is not yet a machine intelligence program specialized in wine. That will change quickly. 

Restaurant owners could develop applications where it will suggest specific wine pairings, possibly even taking into account the tastes of individual clients. It could be integrated into tablet-based wine list programs, for example. Sommeliers can use it to track consumer trends or even suggest top producers, including a description of their terroir and wine style. It has still yet unexplored applications for streamlining business processes. 

[[relatedPurchasesItems-41]]

Examples of chatGPT on Wine

Here are a few actual examples from using chatGPT related to wine:

ChatGPT quickly wrote two profiles of California wine regions (AVAs) with substantial detail. This is a “chat” program, so one must ask questions, view the answer, and ask more questions to refine the answers and gain more details. An initial answer to, for example, a request to describe the Napa Valley AVA will produce very general answers. In one of the two cases, chatGPT got an important detail wrong: it listed a prominent winery which is not true in the AVA. When challenged on the incorrect answer chat GPT admitted its error and corrected the information, but if you don’t know the answer is wrong, chatGPT “sounds” correct and authoritative. Currently, do not make the mistake of taking all wine information transmitted by machine learning programs at face value. 

ChatGPT created wine list recommendations for a 30-seat restaurant within minutes, including, when asked, specific brands and vintages:

“Here are some specific wine recommendations in each category:

1) Sparkling Wine: a. Prosecco: Bisol Jeio Brut Prosecco NV b. Champagne: Veuve Clicquot Brut Yellow Label

2) White Wine: a. Sauvignon Blanc: Kim Crawford Sauvignon Blanc b. Chardonnay: Sonoma-Cutrer Chardonnay c. Pinot Grigio: Santa Margherita Pinot Grigio

3) Rosé Wine: a. Zinfandel Rosé: Beringer White Zinfandel b. Provence Rosé: Miraval Provence Rosé

4) Red Wine: a. Pinot Noir: Meiomi Pinot Noir b. Merlot: Columbia Crest Merlot c. Cabernet Sauvignon: Silver Oak Cabernet Sauvignon d. Syrah/Shiraz: d'Arenberg The Footbolt Shiraz

5) Dessert Wine: a. Moscato: Castello del Poggio Moscato b. Riesling: Dr. Loosen Riesling.”

Readers can draw their own conclusions regarding its choices. It provided vintage recommendations after a subsequent request. 

In another instance, chatGPT was asked to name wine grapes “native” to Germany. “Native” refers to indigenous origin. The program took a more liberal definition of native and answered: “The most common grape varieties that are native to German-speaking countries and regions include: Riesling, Müller-Thurgau, Spätburgunder (Pinot Noir), Grauburgunder (Pinot Gris), Gewürztraminer, Silvaner, and Scheurebe.” When asked where Spätburgunder originated, suggesting it was not in Germany, it said: “You are correct, Spätburgunder (Pinot Noir) is not native to a German-speaking country. It is believed to have originated in the Burgundy region of France. However, it is widely grown and highly regarded in German-speaking regions, particularly in the wine-growing area of Baden.” Too, when asked about Graudburgunder, it admitted: “Grauburgunder (Pinot Gris) is believed to have originated in the Burgundy region of France. However, it is widely grown in other countries as well, including German-speaking countries, where it is known as Grauburgunder.” To clarify further, it responded to the question, “So it is not truly native to a German-speaking country or region?,” with “No, Grauburgunder (Pinot Gris) is not considered to be a truly native grape variety to German-speaking countries or regions. It is believed to have originated in France, but is widely grown and highly regarded in several countries, including Germany and Austria.” 

The exchanges above show how chatGPT operates. One must be very precise and also very circumspect of its answers regarding wine and grape varietals, for instance. Only through a series of questions and answers can one drill down to truly correct and precise information. 

Finally, it made some egregious wine errors, all in a very authoritative tone. In 2002 there was a famous blind wine tasting where Gruner-Veltliner wines competed admirably against French Bourgogne wines. Jancis Robinson MW has written extensively about this, as have others. In terms of the wine world, it was a historic competition that truly enhanced Gruner-Veltliner’s profile. ChatGPT, however, said, “The grape that competed against Burgundy Chardonnay-based wines in the famous 2002 blind competition was Riesling.” What’s more, it went on to add: “The competition, known as the Judgment of Paris, was a tasting event in which a panel of wine experts compared California wines to their French counterparts. In the competition, California Riesling wines were pitted against Chardonnay-based wines from Burgundy, and the Rieslings received high marks from the judges, demonstrating their quality and competitiveness with some of the world's finest wines. This competition helped to establish Riesling as a world-class grape variety and put it on the map as a serious contender in the world of wine.” 

The problem is, this was not the Judgment of Paris, which occurred in 1976 and did not include Riesling wines from California. The event included California Cabernet Sauvignon and Chardonnay wines competing against famous French counterparts from Bordeaux and Bourgogne. Chat GPT did not understand, in fact, that the point was to compare apples with apples, not apples with oranges. The Judgment of Paris is an extremely well-known event in the wine world, with lots of information freely available, but, as Nicholas Thompson has put it, “we know machine intelligence can hallucinate.” 

A unique opportunity to present your wines to America's top sommeliers. The wine scores are benchmarked for on-premise channels by top sommeliers, master sommeliers, wine directors and restaurant wine buyers. Submission deadline is April 30, 2024.