Can The Perfect Algorithm Brew The Perfect Beer?

“Essentially, the beer is self-tuning over time,” say the creators of a new AI-backed brewing company.

Companies like Google are using machine learning to train AI to do everything from giving us better search results, to actually suggesting the words we should text to our friends. But how might we bring some of this technology to our physical world–to the things we touch and taste?


That was a question asked by the innovation and consulting firm 10x and machine learning specialists at Intelligent Layer, a London-based firm founded by an Oxford-trained machine learning specialist. As the two companies shared the same WeWork space where free beer flowed from the office kegs, they realized–likely over some impromptu happy hour–that they were already looking at the perfect product to hone via wave after wave of human feedback: beer.

And so for the last 12 months, they’ve developed a brew label called IntelligentX AI Beer. It’s a series of beers, classified simply as Golden, Amber, Black, and Pale, distributed at bars and stores across London. And its claim to fame is, rather than working from one recipe, the ingredients and processes to craft IntelligentX brews are constantly being rewritten by a learning AI.

“With beer, you can learn with each batch and brew another one,” says 10x CEO Hew Leith. “Essentially, the beer is self-tuning over time.”

To train the system, IntelligentX sells you a beer. Say it’s their Pale. You drink it, and then you log feedback into a Facebook Messenger bot. The bot starts by asking your favorite beer of all time–that gives the machine a frame of reference to your tastes–and then it drills down into qualities of the beer you tried, rated on a scale of 1-10. Was it too hoppy? How was the maltiness? All of this feedback is aggregated into the system, which thinks through how to solve the problems. For instance, too much hop flavor could be solved by changing the hops, or shortening the length of the mash, or using water with a different pH level. Whatever the solution may be, the computer spits out a new recipe to a brewer. And the brewer remakes the beer, using their own knowledge in the art and science of brewing to reject any bad ideas.

The system sounds a bit like IBM’s quest for cognitive computing, in which Watson, wearing his chef hat, generates recipes for a chef to consider and cook. But there’s an easier way to understand it, too: IntelligentX is really just building the age-old practice of focus group testing directly into their product development cycle, removing the middlemen statisticians, and putting the feedback onto the assembly line.

“Essentially what we’re doing now is putting all of our customers in the same room with the brewer,” says Leith. “We don’t believe algorithms are going to rise up and take all of our jobs. We believe AI will augment human skills. And that’s what we’re going to do here, to give the brewer super-human skills.”


Assuming the underlying technology is coded well enough, IntelligentX sounds promising, but the question becomes the same thing asked by any good designer looking over focus group data: Should they incorporate all feedback? Or more specifically, can so much feedback water down IntelligentX’s beer to the point that it’s no longer a set of unique craft brews, but a single crowd-pleasing Budweiser alternative?

Leith has heard this criticism before, and acknowledges the problem before I even bring it up. The solution, he says, is largely in the system’s “wild-card ingredient” they’ve programmed in. The AI suggests one of 1,000 different additives–the example he offers is grapefruit–so that, much like Dogfish Head does today, which adds ingredients like lobsters and peaches to their brews, IntelligentX can stay surprising through sheer ingredient novelty. That novelty goes hand-in-hand with the fact that IntelligentX vows to never brew the same beer twice, so the “Black” beer they sell today could taste vastly different when the recipe changes in a month. And in this sense, their product becomes one of guaranteed change, rather than consistent flavor.

“We’re not chasing the perfect beer here. We’re providing an experience to take our customers on a journey to try different things and really push people’s palettes as to what’s possible with craft brewing,” says Leith, of the thousands of liters of limited-batch beers they brew monthly. “We’re the opposite of a large corporate brewer that wants to brew one thing and sell it. We’re about brewing constant iterations and learning over time.”

In this regard, Leith likens his product to Pokémon Go, in which “everything’s collectible” by nature because the flavor is fleeting. The label’s next goal is to win a major beer competition, and if IntelligentX continues to do well, Leith imagines all sorts of other products could see the same AI treatment–including chocolate, coffee, and perfume.

Do you smell that, too? It might just be the future of product development.


About the author

Mark Wilson is a senior writer at Fast Company who has written about design, technology, and culture for almost 15 years. His work has appeared at Gizmodo, Kotaku, PopMech, PopSci, Esquire, American Photo and Lucky Peach