There are two ways this can go: ugly or hard.
Ugly looks like life in the uncanny valley—color, but no cohesion. Aroma, but no nuance. Flavor, but no fizz. A glass of beer made by machines, just for you. A recipe designed by a computer, fermented in a lab, and marketed with brands and designs micro-targeted specifically at you.
Tastes fine, though. This Tears in the Grain IPA—okay, that’s the name ChatGPT suggested when I asked it for a name referencing Blade Runner and featuring a cringe-worthy pun—is pretty good actually. It’s just missing something. Hard to say exactly what. The human touch? A soul?
We are bombarded by AI. Every day, the intensity of the onslaught of buzzwords and startups and promises of disruption seems to intensify. It can be unsettling—a feeling of vertigo on the precipice of the AI revolution. We know it’s all going to change; we just don’t know which future we’ll get: the Jetsons (jetpacks! finally!) or The Matrix (oof).
So, about that hard way.
I don’t mean hard like the open warfare against the all-controlling AI as seen in The Matrix. However, AI’s ubiquity ensures it will change the beer industry. And change is always hard.
A New Way of Thinking
I don’t want to give the impression that AI is a new force acting on the brewing industry. As Thomas Schlechter writes in the 2024 study “Impact of AI on the Brewing Industry: A Comprehensive Summary,” AI technologies have influenced brewing for more than 30 years.
The current wave of attention to—and excitement around—artificial intelligence rose with the launch of OpenAI’s ChatGPT bot in 2022. In just three years, the agent has gone from a neat party trick to an indispensable tool used by hundreds of millions of people every week.
ChatGPT and the whole field of competing AI chatbots are just one flavor of AI. Known as large language models (LLMs), these systems digest examples of written content and metabolize the resulting massive dataset into a predictive model. John Hughes, a brewing scientist at UC Davis, likens LLMs to the predictive text function on our smartphones—they’re just much more advanced and complex.
Central to the theory of AI, and at the core of the LLMs themselves, is machine-learning technology. The field of study focuses on developing the algorithms behind all the intelligent systems. These algorithms apply statistical models to huge datasets and predict unknown values. Furthermore, the algorithms use these predictions to make decisions without human interactions. At the cutting edge of machine-learning technologies are concepts such as deep learning and neural networks, which further work toward making these machines more human-like in how they think.
“It’s a useful tool,” Hughes says, “but we are really early in the curve.”
It remains to be seen whether the LLMs will continue to improve predictive capabilities into something akin to comprehension, or whether they’ll drown in their own slop—might there a tipping point where the bots are trained more on AI-generated texts than human-written words?
Hallucinations in the Brewery
Homebrewers and pros have been asking ChatGPT for recipes since the service launched to the public—even if just to see what it spits out—and there are several commercial beers marketed as being based on AI-generated recipes. Boston’s Night Shift, Detroit’s Atwater, and Albuquerque’s Rio Bravo are among the companies that have released one.
Some intrepid brewers have even pitted their own recipes against the machine’s creations in blind taste tests. Curiously, few of the brewers who’ve made beer based on AI-sourced recipes wanted to talk about their experience. The volume of beers based on AI-created recipes also appears to have dwindled after the initial excitement over the tech. And if we’ve learned anything from 30-plus years of craft-brewing ups and downs, it’s that novelty is fun and exciting, but it eventually cools and doesn’t ensure longevity.
In Massachusetts, Tree House runs a YouTube channel that goes beyond the typical brewery marketing content. Videos often feature cofounder Nathan Lanier sharing insights on topics from beer education to homebrewing tips to detailed looks into the brewery’s technical and artistic processes. In March 2024, Tree House posted a video titled, “Can AI write a good hazy IPA recipe?” In the video, Lanier asks the Mistral AI chatbot to concoct a recipe for a New England IPA.
The LLM returns a serviceable if rote recipe for a haze-bomb, but there are enough odd details to give the brewer pause. It recommends an American ale yeast, then suggests London II as an example. It also suggests a post-fermentation cold crash to “clear” the proposed hazy beer. Lanier’s recap of the experience has a positive slant, but there’s a biting disdain beneath the surface. He calls the output “juvenile,” and “relatively lacking,” says he’s “not sold on AI” in the brewery, and reiterates that a chatbot’s output is simply “an amalgamation of the information that’s available to all of us.” He also calls out the “loss of humanity” in the output of the machine.
A year on from that video, chatbots have become even more prevalent, and their algorithms have improved. Yet their output hasn’t necessarily gotten any better. There appears to be an ouroboros effect happening, where the growing mass of chatbot output on the web means that chatbots are digesting more of their own content, and there’s less humanity in the output.
“I wouldn’t trust any recipe [an LLM] gives me,” says Hughes at UC Davis. He warns that chatbot output appears to increasingly contain obvious factual errors and even wholesale fabrications. Called “hallucinations,” the errors stem from the algorithms seeing patterns in data that don’t actually exist or that are not meaningful. (One explanation of AI hallucinations likens them to humans seeing animal shapes in clouds.)
I experienced ChatGPT hallucinations for myself when researching this article. I had avoided engaging with the AI directly—I was deeply uncomfortable with the tech—but I rolled up my sleeves and sat down to “chat” with ChatGPT. I tried to get its perspective on how AI can help craft brewers.
The first three people whom it suggested I talk to for this article didn’t exist. They appeared to be confabulations of several people wrapped up into one name. Several companies that it cited as examples of AI integration in the brewing space were either defunct or entirely fictional. It wasn’t a reassuring experience.
A More Focused Tool
Many brewers do use, and even rely on, LLM chatbots. But none of the ones who spoke to me are asking AI for recipe advice. These brewers are using it like most people seem to be—for mundane tasks such as scheduling, creating job descriptions, and writing SOPs.
Bob Kunz, founder of Highland Park Brewery in Los Angeles, says ChatGPT is like a personal assistant, speeding up research projects or streamlining the kinds of complex bureaucratic tasks that come with running a business.
There is at least one alternative to off-the-shelf chatbots that’s more brewer-focused. ProBrewerGPT is an attempt to package the power and utility of ChatGPT specifically for the brewing-industry professional. ProBrewer cofounder Pat Hagerman says they developed the AI agent to help expose 30 years of ProBrewer content and community knowledge in a more user-friendly way. Using retrieval-augmented generation (RAG) to narrow the scope of the information indexed by the LLM, ProBrewerGPT is focused on technical issues and the kind of minutia that brewers have discussed on troubleshooting message boards over three decades.
“As a publisher, it’s our job to expose people to facts, information, and opinions, and we’ve built an amazing archive of content,” Hagerman says. Between editorial writing, the ProBrewer discussion boards, and even the classified ads, traditional search methods don’t surface relevant details as well as the custom AI agent.
Placed between the user’s questions and the LLM chatbot, the ProBrewerGPT agent works to constrain and guide the AI. “It’s like giving ChatGPT a job description, fencing off a designated area for it to search, and making sure it prioritizes factual content.”
The LLMs are like djinn who won’t go back into the bottle, and brewers inevitably are learning how to make their own best use of the tools.
However, LLMs aren’t even the most powerful incarnations of AI at work in the brewery.
An Industry Infestation
AI is already embedded in our pockets and desktops, and it feels impossible to avoid. AI is just as prevalent in breweries—but it’s most visible in all the same ways it’s visible in any business. It’s helping workers to answer questions, off-load simple tasks, and manage administrative chores.
Yet AI’s impact goes beyond user-facing chatbots. Machine learning is increasingly embedded in any equipment that’s wired to the internet. Anything that uses digital sensors to collect data is a target for AI integration, and these systems can be found throughout many breweries and taprooms.
In Bend, Oregon, Deschutes adopted a machine-learning system to increase efficiency in the cellar. Using their historical, manually gathered dataset of tank residency times and temperatures, the innovation team trained a machine-learning algorithm to predict the precise length of each phase of fermentation. The end result: Their beer spent less time in tanks, which meant they could turn each tank more often. They made more beer in the same amount of time with the same capacity, and the brewery rerouted funds allocated to growing the cellar back into the innovation project.
Other labor-intensive brewing processes are targets for AI intervention. From making cell counts in the lab to optimizing mash temperatures, machine learning is applicable whenever a process generates data. Anywhere that automation exists in the brewery, AI isn’t far behind. Collect a large enough dataset, and the algorithms can predict results—even as variables change. Or, at least, that’s the promise of the technology—but a brewer is equal parts artist and scientist, and placing trust in machine-made decisions can feel counter to the craft.
Speaking of art: The image-creation and design capabilities of generative AI systems such as MidJourney promise to “democratize” the artistic touch and allow anyone to create photo-realistic images simply by describing what they should look like. The technology is all over social media and ecommerce sites, and—much like chatbot-generated text—the images have a hard-to-describe quality that is obvious. AI labels and can designs are certainly out in the marketplace, but they’re still relatively rare. Craft beer trades on a sense of authenticity, after all, with a handcrafted appeal. Slapping AI art on the packaging may cost a brewery more in authentic appeal than it saves by not paying a designer.
Meanwhile, the biggest impact of AI and machine learning on the brewing industry is happening outside the breweries. From the barley fields and hop yards to the delivery trucks carrying kegs to the local pub, AI is reforging the entire supply chain.
There are algorithms that can decide which strain of barley should be planted in each micro-plot. There are computer-vision systems that can grade the harvested barley. And there are automated kilns that can process the malt with less energy. AI is informing hop-breeding programs, yeast design, and hybridizations. The presses, extractors, and membrane systems that produce flowable advanced hop products or pull the ethanol from beer also use AI to increase quality and yields.
“We use AI as a thought partner, but not in a groundbreaking way—at least not yet,” says Nick Harris, cofounder of Berkeley Yeast in California. “I’m excited for the technology to improve, though. In the future, I imagine it will save a lot of time when optimizing secondary metabolism or selecting mutations for protein engineering.”
AI, Breaker of Headwinds
AI’s influence on the beer industry probably won’t be in the invention of the perfect beer style or in the most compelling level ever seen in the history of packaged beer. The most important changes will be those countless tiny improvements across the supply chain, within the brewery, and in the marketplace. A couple hours shaved off fermentation time isn’t a big deal over one batch… but over 100 batches at a brewery in a year, it adds up to meaningful savings, across the whole industry.
Back when the rising tide was lifting all the craft-brewing boats—and even middling breweries with questionable business practices were growing unchecked—operators could be a little loose with the details. But the tides have shifted, and craft beer has more competition than ever. Breweries have to trim the sails and tighten the ship if they want to weather the headwinds. Inefficiencies will sink a business that ignores them in the face of shrinking margins.
The trick to succeeding in the face of AI-driven changes will be to find ways to leverage the precision and inhumanity of the AI systems without sacrificing craft beer’s foundation of authenticity. That’s the hard road. It could become easier to give the reins to AI, as the tech improves. But I don’t think craft beer’s core customers would stomach the loss of humanity in the glass. Nor would craft breweries be eager to become de facto contract brewers for AI “brewmasters.”
Machines are learning how to read, write, and speak more like humans—they’re getting better every day, and the hardware to allow machines to see, smell, and even taste is likewise continually advancing. But when an algorithm samples a beer with a gas spectrometer and a silicone-based “e-tongue,” is that really tasting?
Will a machine know the joy of a ray of spring sunshine glinting off a perfectly poured pilsner, Saaz aromas leaping from the foam to your nose, crisp-cracker malts sliding into zesty bitterness, and the holistic experience of that sip that echoes back to 19th-century Plzeň?
Or, will all those moments be lost in time, like craft beers amid a torrent of AI-generated lager-slop? Because that would be the ugly.