Wine made by artificial intelligence is now a reality at this Wisconsin winery

ALGOMA - Does artificial intelligence have slightly oaky and spicy tones with dry fruit flavors a silky mouthfeel? Maybe AI is light-bodied with floral notes and tree fruit and citrus flavors?

That's possible at von Stiehl Winery, which is using AI to help decide the flavors and aromas of a groundbreaking new line of grape wines.

von Stiehl staff show off bottles of their new Artificial Intelligence V2.0 wines in their Algoma winery. From left are co-owner and winemaker Aric Schmiling, cider maker and vineyard manager PJ Koehler, assistant winemaker Chad Moore, cellar worker Eryka Hughes and winemaker Dave Pizzala.
von Stiehl staff show off bottles of their new Artificial Intelligence V2.0 wines in their Algoma winery. From left are co-owner and winemaker Aric Schmiling, cider maker and vineyard manager PJ Koehler, assistant winemaker Chad Moore, cellar worker Eryka Hughes and winemaker Dave Pizzala.

The Algoma winery's Artificial Intelligence White V2.0 and Artificial Intelligence Red V2.0 are blended wines that use AI based on feedback from consumers to shape the recipes for these blends.

While AI is seeing increasing use in growing and harvesting the grapes for wines, using AI to make a wine with flavors that are essentially crowdsourced directly from its customers is a method that hasn't been done anywhere else, as far as von Stiehl general manager Anthony Bilwen knows. Its a big, innovative step for Wisconsin’s oldest licensed winery, but Bilwen said it's what von Stiehl does when the opportunity arises.

"With von Stiehl, we've always tried to be on the cutting edge of the technical side of things," Bilwen said. "It's in our DNA, trying new things."

Here's how it works.

AI already is at work in the vineyards, but it's new to making the wine

Artificial intelligence is rapidly making inroads into the wine business, as numerous articles in wine industry magazines have attested in the past years.

But it's utilized mostly in the vineyards, where data from weather and growing conditions enables AI to determine irrigation needs, detect disease and predict the optimum times to harvest grapes. It's also being used in wineries' cellars to help wine makers monitor the fermentation process.

But AI also is making moves directly onto wine drinkers' taste buds.

One French wine maker, Aubert & Mathieu, fed information on their location and the types of grapes it had available to the popular AI chatbot system ChatGPT and asked it to create a red wine based on that, according to an article in Decanter magazine. ChatGPT responded with information on basic wine making and suggestions on not just a general wine recipe but also on the bottle, label, marketing and name (which is "The End").

And, an article posted in April on the BBC.com website notes that California tech company Tastry has software that can analyze the sugar levels, acidity, tannins and other chemical elements of a wine and then compare it to taste preferences of about 248 million American wine drinkers. The idea is to guide wine makers, especially those who produce for the masses and aim for greater consistency, to wine blends that will be popular with a broad audience.

However, as Bilwen said, von Stiehl's AI V2.0 wines are the first to use direct feedback from consumers to help determine their wine recipes. An online search didn't turn up any other wineries using von Stiehl's method.

It started with a taste and a QR code

von Stiehl started the process for Artificial Intelligence V2.0 wines last summer when it made and issued two Artificial Intelligence V1.0 wines, one a red blend and the other a white blend.

The label of each V1.0 wine bottle carried a QR code that those who had the bottle and tasted the wines could scan. The scan took the wine drinkers to a survey where they could give their tasting reviews: What they liked and didn't like about the flavors, aromas or mouthfeel of the wines, even the bottle labels.

Those responses were fed to the popular AI chatbot systems ChatGPT and Google Gemini to analyze all the data. von Stiehl's wine makers then asked for a paragraph from the chatbots summarizing the critiques of the V1.0 blends along with suggestions on how to best alter their recipes to satisfy those critiques.

von Stiehl Winery used reviews of a red blend and white blend wine fed into ChatGPT and Google Gemini to get recommendations on how to make their new Artificial Intelligence V2.0 blends in their Algoma winery.
von Stiehl Winery used reviews of a red blend and white blend wine fed into ChatGPT and Google Gemini to get recommendations on how to make their new Artificial Intelligence V2.0 blends in their Algoma winery.

In other words, the consumers largely decided how this year's AI White and Red V2.0s would taste, smell and feel, using the V1.0 wines as a base. Which is what von Stiehl was aiming for.

"We collected all that data on customer feedback – is the wine too dry, too fruity, too floral? So what do we do to alter it?" Bilwen said. “While we’ve always valued our patrons’ feedback, the introduction of AI-driven wines empowers them to directly influence our winemaking process like never before."

But, AI isn't taking away the artisan side of wine making

Bilwen said the wine makers weren't given specific recipes by the ChatGPT and Gemini reports. Instead, the reports offered suggestions like "reducing sweetness" and "balancing acidity," for example, then offered ideas on how to implement them, including specific grape varietals that could be added to the blends to achieve those aims.

It was then up to the wine makers to decide how to solve the issues raised by those suggestions. So although these are AI-driven wines made from crowd-sourced recipes, they still rely at least a little bit on the time-honored art of wine making to come to life.

"The technology isn't smart enough yet to kick out a tried-and-true recipe (for a wine blend)," Bilwen said. "Unless there's a recipe out there, and most wine makers won't do that, won't make their recipes public."

This method of making wines may someday become the future of the business, but for 2024 the AI Red and White V2.0s are the only von Stiehl wines made this way, among the more than 40 wines and ciders produced by them.

"We're not going to do this for every one of our wines," Bilwen said. "That kinda takes the artisan side out of it."

So, what changes were made from AI V1.0 to V2.0? And how do the V2.0s taste?

According to von Stiehl's website, ChatGPT and Gemini returned similar summaries of the AI Red V1.0 to von Stiehl. Both noted that consumers' opinions on taste, flavors, mouthfeel, finish and even the label varied quite a bit. Consumers apparently liked the dark red color and readily noticed the dark fruit flavors and oaky tannins on the finish, although opinion apparently was divided on whether those last two were positives or negatives.

In general, the chatbots recommended making a slightly drier wine by reducing the sweetness while maintaining the fruity character, enhancing the body and depth and finding a better balance of acidity. They also suggested several grape varietals that could accomplish those goals.

The AI White V1.0 also received mixed reviews that were noted in the chatbot summaries, which reported consumers generally liked the citrus and pear aromas but were split on whether or not the taste and the finish were too sweet. As with the red, the recommendations for the white were to reduce the sweetness and balance the acidity, while ChatGPT again recommended specific grapes to introduce to the blend.

Also, ChatGPT suggested adding more information about the AI wine effort to the bottle labels on both wines, perhaps through another QR code.

The result, von Stiehl's website says, is that Artificial Intelligence Red V2.0 "lessens the sweetness and adds an oaky and slightly spicy character" compared to V1.0 while AI White V2.0 is slightly drier and a little more acidic than its V1.0 while retaining its lightness.

Where can I find AI V2.0 wines?

Samples of the AI wines are available in von Stiehl's tasting room, of course, but wine drinkers need to buy a bottle to scan the QR codes and be able to submit a review for consideration in future wine recipes.

Bottles are available at the winery and through its website, and Bilwen said they will appear in liquor and grocery stores that carry von Stiehl, such as Festival Foods, by the end of June.

Wait, did you says 'consideration in future wine recipes?'

Yep. This will be an annual thing.

Bilwen said the winery is planning to make new AI-generated blends every year, so people who get a bottle of the V2.0 wines can scan the QR code on the label and offer their opinions on the wine, which will be gathered during 2024 and summarized by AI for use by their wine makers in the V3.0 wines coming next year.

For more information

von Stiehl Winery is at 115 Navarino St., Algoma, along with its cidery and distillery. For more information, call 920-487-5208 or visit vonstiehl.com.

Contact Christopher Clough at 920-562-8900 or [email protected].

MORE: A Door County marina in business for almost 60 years is sold to a Texas-based company

MORE: Door County marine construction firm gets new owner, will move to new, bigger location

FOR MORE DOOR COUNTY NEWS: Check out our website

This article originally appeared on Green Bay Press-Gazette: AI used to help create wine blends for von Stiehl Winery in Algoma