Artificial intelligence (AI) continues to be adopted in the drinks industry, with many large and some smaller craft brewers looking to adopt the technology in an array of business areas.

Some major breweries have taken interest. Earlier this year, Heineken launched its global Generative AI lab in Singapore. Following an initial pilot, one of the areas Heineken aims to be able to use GenAI in the Singapore space is in its marketing and sales teams. The group also has a GenAI-supported knowledge and insights management platform and also uses a GenAI chatbot called Hoppy to support its finance teams.

When it comes to production, uses of AI appear to have been more focused on limited edition product launches. Take the launch of Beck’s Autonomous by AB InBev and Molson Coors’ Atwater Artificial Intelligence IPA launched by its Atwater Brewery.

Back in 2017, Carlsberg partnered with Microsoft, Innovation Fund Denmark and two Danish universities to develop sensors that detect specific aromas in beer and develop a “flavour fingerprint” for various samples. The information was, at the time, expected to be able to be used assist with developing new beers.

At the time of writing however, Carlsberg confirmed to Just Drinks that project was no longer running.

Quality control and product and recipe development have been some key areas where AI has been used in terms of production but how are brewers still using AI in these areas and is it likely that uses of AI will scale in the production side of beer anytime soon?

Speeding up processes

One major global brewer that uses AI to assist with its beer production today is Kirin.

The company Kirin first piloted a system called Takumi AI in 2021. The system is presently used only in product development facilities in Japan but has moved from the piloting phase four years ago “to full-scale operation” today, Satoshi Okada, fermentation technology specialist, and Yusuke Kase, manager, at Kirin Brewery’s Institute for Future Beverages tell Just Drinks. .

Takumi AI acts as a beer recipe data generator. “It is a system that supports brewers throughout the entire beer product development process”, they note.

The programme refers to data which is input by senior brewers at Kirin. According to Okade and Kase: “Through Takumi AI, which incorporates the brewing expertise of seasoned technicians, we expect to promote the transmission of skilled techniques.” The technology is expected to help more experienced brewers pass on techniques to junior brewers, and help them identify recipes that will achieve the desired taste in a shorter period of time.

Through Takumi AI, which incorporates the brewing expertise of seasoned technicians, we expect to promote the transmission of skilled techniques.”

Satoshi Okada, fermentation technology specialist and Yusuke Kase, manager, Kirin Brewery’s Institute for Future Beverages

Speeding up the production process is also something being done by the firm Northwest AI Consulting, which helps businesses implement AI solutions into their operations, has been involved in.

While the group cannot disclose the names of the brewers it works with, its clients include US craft brewers and mid-sized brewers in the US as well as an international contract brewer in Europe.

One of the areas it’s been helping some of its clients with is reducing fermentation time.

“We start with their sensor and production data so things like temperature, pH, CO₂ levels, fermentation curves, tank usage history, and then build predictive and prescriptive models,” explains Wyatt Mayham, founder of the consulting firm.

“For instance, one client reduced tank residency time by over 30 hours per batch using AI to optimize fermentation curves [data visualisations of rate and health of fermentation]”.

In this scenario, the brewer was able to use a model that could recommend pressure adjustments, and ways to control temperature.

Quality control

Monitoring quality control in beer is another area where machine learning models can help.

“I think it’s almost just by the nature of what AI can do, where it’s going to come into its own is very much going to be the quality control side of things,” says Kevin Baker, global research manager for beer and cider at GlobalData, Just Drinks parent.

Improving flavours is one thing AI has been tested for in recent years, but it’s still very much in the research stage.

Last year, a team of Belgian scientists published research on AI models they had developed which they said can predict how a beer will be received by drinkers, as well as which aroma compounds brewers can add to improve a product.

The group chemically analysed around 250 beers, compiling data on taste profiles and aroma compounds in beers over a five-year period.

With enough data, the scientists found they could use machine learning to predict how a beer might taste and used it to develop models that could predict the flavours in beer by interpreting data on the chemical makeup of flavour compounds in the 200 strong beers they had chemically analysed.

Working alongside machine-learning students, they also managed to train a model that could predict how much a consumer might enjoy the specific combination of aroma compounds in a new beer that they haven’t yet tasted.

“You get quite close, actually, for certain aspects,” explains Kevin Verstrepen, professor at KU Leuven and director of the VIB-KU Leuven Centre for Microbiology and Leuven Institute for Beer Research, and who helped to develop the research project.

“For example, bitterness impression, or fruitiness you get to an accuracy of 80-90%. For appreciation, so that’s literally the sum of everything, and then asking your tasters, ‘How well do you like this beer?’, you get to 50-60% which maybe doesn’t sound very high but then you also have to take into account that appreciation is, of course, extremely personal.”

With the research, the scientists could also use machine learning algorithms to adjust existing beers to make them more enjoyable.

Verstrepen explains: “You can take an existing, quite good beer… and then [ask] the model, ‘Okay, which parameters, which aroma compounds could we tweak, could we increase a little bit, to make this beer even more liked?’ [It was] quite amazing that it worked out.”

AI is also being used to help some brewers improve their fermentation processes.

One of North West AI’s craft brewing clients has adopted “reinforcement-learning based [near infrared] spectroscopy systems” – a machine-learning supported technique that uses near-infrared radiation to measure specific molecules in the beer, like sugar, ethanol, proteins and flavour compounds. It can recommend particular tweaks to be made to the fermentation process in real-time.

“In essence, it’s a closed-loop control system that doesn’t just follow a pre-programmed recipe; it actively steers the fermentation in real-time to guarantee quality and consistency,” says Wyatt.

NPD: is it a gimmick?

While Kirin’s Takumi AI can help the company speed up the process of passing on brewing techniques to more junior engineers, it can also be used to create new recipes, with staff able to refer to a database of technical knowledge on correct ingredient quantities and brewing conditions.

The company does not disclose the specific beers it develops with the assistance of Takumi AI.

A clutch of brewers developed new products using AI a couple years ago, but new developments since then have been few and far between.

As GlobalData’s Baker notes, “certainly the recipe side of things tend to be a gimmick”. In other words, they’re not product launches that are here to stay.

For Baker, where AI becomes particularly useful is in processes where it can pull and analyse plenty of data. “As I say, and it’s not just because I work in that area, but I think, it comes into its own, because it’s basically a computer system. That’s where it’s going to have the biggest impact is [with any] thing that is computerised. That’s handling large amounts of data, essentially.”

Others however do see more promise in the new product development side.

UK-based cask beer specialist Nethergate Brewery launched its special edition AIPA, made with the assistance of ChatGPT, in January last year.

Pitching the product to on-trade venues was tricky at first, managing director James Holberry tells Just Drinks, but the positive consumer satisfaction received means developing another beer with AI in the near-future could be a possibility.

“Feedback on its taste, its quality, its flavour build was really, really good. And actually, lots of people have asked for it back since, and potentially [it] is something we might revisit again next year.”

Holberry says the group will look to make another special edition brew for next year’s cask rotation, using ChatGPT and other platforms to see how it could adjust the recipe, or whether it could be adapted to suit a bottled format.

The team were slightly hesitant to use AI at first, says Holberry, but became more comfortable with it the more it tested it. “Like a lot of technologies, the more you kind of jump into it, use it, play around with it, trial it, test it, the more confident you come become with it, and the better you become with it.”

 While Nethergate had a positive experience with the AIPA roll-out, it is worth noting it happened on a small-scale. The craft brewer supplies 250 pubs in the East Anglia region and is regionally focused.  

Benefits for non-alc

One segment of beer where researcher Verstrepen sees growing potential for machine-learning trained models in brewing is non-alcoholic.

Together with a group of other researchers, Verstrepen is working on ongoing research around how AI can be used to improve non-alcoholic beers. “We feel that that is really a goal that is worth chasing,” he says.

While Verstrepen can’t share the full details of the findings yet, he noted that machine learning is particularly beneficial for developing non-alc beer, as “it can grasp all the complex interactions”.

He added: “And so that is exactly where the machine learning shines, because it can grasp the full complexity of the aroma palate and predict what more complex changes to the beer would do, which opens up the possibility of better mitigating the effect of not having any alcohol in the beer.

“In other words: machine learning can help us predict how varying multiple aroma compounds can help recover the aroma that gets lost or modified when alcohol is removed.”

Human touch

When it comes to production, for now, for now at least, AI is by no means able to perform the same role as brewers, with machine-learning supported models still needing to be monitored and adjusted by real people.

As Kirin’s Okada and Yusake note: “Final decision-making regarding beer flavour requires brewing technicians.”

Nethergate also made regular adjustments to its AIPA during in its development, says MD Holberry. 

“Ultimately, it was a learning journey for me and the team and there [were] elements about it that we found incredibly useful. There [were] elements about it that it probably needed a lot more guidance and steer and influence. I think, like a lot of these technologies, often it’s only as good as the input.”

He adds while the technology was able to create a successful product for the business, it cannot perform the same level of quality testing as brewers.

“What ChatGPT or any artificial intelligence for that matter, will never be able to do is it’ll never be able to smell the quality of that yeast. It’ll never be able to see the quality of the hop and the malt that goes into it. It’ll never be able to taste the flavour or the quality of the water that’s used, and ultimately, it’ll never be able to taste the pint at the end.”

Hesitation still exists but curiosity has definitely passed initial scepticism.

Wyatt Mayham, founder, Northwest AI

Beer’s ties to tradition are also what puts some brewers off from adopting AI, with some convinced it will destroy the essence of the product.

As Verstrepen notes, when his research has been presented to brewers,  “you have quite a few brewers, especially often the craft brewers, who say, ‘Ah no, we are making a traditional product, and it’s all an art, and it should stay that way.’

“A big fear of, say, the more traditionalist brewers is that the machine learning is going to turn beer brewing into computation, and that the whole art is going to be lost, and that also the nice diversity that we have now in beers will be lost because the computer will predict a few ideal beers.”

Mayham at Northwest AI agrees that hesitancy is still around but there are signs that opinions are slowly changing.

“Yes, hesitation still exists but curiosity has definitely passed initial scepticism,” he says. “Many brewers worry about the ‘human art’ being lost, but once they see the data and outcomes, they welcome AI as an assistive tool, not a replacement for people. It’s a gradual shift: once quality improvements and waste reductions are proven, they become advocates.”

According to Verstrepen, while AI models are offering brewers recommendations, the brewing staff themselves still need to make these specific adjustments. The process isn’t automated and it still requires a high level of skill.

He adds, “to make that link between how to implement this in the brewing process, you need an extremely well-educated brewer… The signs of brewing becomes even more important in a way.”

What next?

How far are we from seeing AI being rolled out at scale in beer production? It might be more of a long-term development, and it isn’t something that’s being shouted about by the industry at present.

Taking Kirin for instance, beyond Takumi AI, it is unclear whether Kirin is continuing to develop further models to assist it with brewing. Kirin did not disclose if it was testing out other AI models or programmes in beer manufacturing at the moment.

According to GlobalData’s Baker, brewers’ uses of AI more broadly across their businesses appears to have reached a bit of a stagnant point.

“AI, generally in society, it’s probably still quite high profile. But it definitely seems like in the beer industry, there was a buzz about two years ago, and it’s kind of subsided. And I say really it’s because, you know, the industry can be quite faddy.”

Some however remain optimistic about AI’s future potential, especially in production itself.

AI, generally in society, it’s probably still quite high profile. But it definitely seems like in the beer industry, there was a buzz about two years ago, and it’s kind of subsided.

Kevin Baker, head of beer and cider research, GlobalData

Mayham at Northwest AI says he sees “a growth trajectory” for AI’s role in beer production.

“The most popular platforms so far involve sensor integration, ML-driven predictive control of batches, and reinforcement learning to maintain quality.

“As tools mature and integration costs drop, I expect more brewers (including smaller craft ones) to adopt flavor-prediction systems and fermentation optimisation tools.”

As Holberry at Nethergate argued above, AI will never be able to smell and taste beers to determine quality, but, he adds, this doesn’t mean it’s out of the question: “Maybe one day it will be able to assess the quality of that malt sack and the quality of that ingredient.

“Maybe one day it will be able to taste a pint of beer and be able to compare that in flavour and strength and hoppiness. Maybe one day, it’s still very early.”