The real ale developed using artificial intelligence
Artificial intelligence is transforming how real ale is brewed thanks to a cutting-edge start-up.
The UK’s real ale industry is growing, with companies such as BrewDog, Camden Town Brewery and Innis & Gunn generating millions of pounds through crowdfunding site Crowdcube.
IntelligentX, a joint venture between machine learning company Intelligent Layer and the innovation and creative agency 10X, is taking beer tech to a whole new level.
We spoke with Dr Rob McInerney (pictured left), co-founder and CEO of Intelligent Layer, about how he and 10X CEO Hew Leith (right) came up with the idea – and how it works.
Where did the original idea for IntelligentX come from?
Hew and I were sharing a beer in our co-working space in London. We both have a passion for craft beer and started talking about how technology could transform the craft beer production process. We felt that data could have a role, but realised the real key was using machine learning as a means to rapidly innovate and test new ideas. We saw that the opportunity was to create a whole new class of product - something that continually adapted and changed towards the dynamic preferences of engaged consumers.
Speaking to the layman, how does it work?
The algorithm, ABI (short for automated brewing intelligence), learns in much the same way a human learns to brew. It starts by observing the recipes that human brewers come up with, then once it is more experienced it starts coming up with its own ideas. Initially these ideas are tested on expert brewers, who either reward or penalise the algorithm based on how good the recipes seem to be. Eventually the algorithm is able to come up with ideas good enough to put in front of our consumers. The consumers can then give feedback which is interpreted by the algorithm into new beer recipes.
Speaking to a techie, how does it work?
We use a combination of Reinforcement Learning algorithms and Bayesian statistics. Reinforcement learning is the mechanism through which the algorithm learns from experience - if it does something positive, then it receives a reward and that behaviour is reinforced. Bayesian statistics represents a highly robust form of statistics that forms the backbone of how the AI understands the world. It essentially provides the rules of the AI's brain - the processes through which it understands and uses information.
How is feedback gathered/delivered?
On each bottle of IntelligentX beer is a URL which takes you to a Facebook Messenger Bot. The algorithm asks questions dynamically to each user, adjusting the questions based on the knowledge of the person answering and the information it needs at that moment to come up with the next recipe.
Which ingredients/aspects of the brewing process are altered by the algorithm?
We are ultimately able to alter all aspects of the brewing process using the algorithms. However, whilst there are only four key ingredients in beer - water, barley, hops and yeast - there are almost endless different varieties of each of these. It is infeasible to accommodate all of these, so instead we select a subset and use the algorithm to optimise flavour within that subset. Interestingly this means we could apply our algorithm in different breweries, using different subset and get quite different results.
How many types of beer do you produce in this way? And what are they?
We currently have 4 varieties.
Amber - derived from a British bitter, which has a darker appearance and stronger flavour, with a hint of grapefruit for a fresh taste of summer
Golden - the origins are from a classic British golden ale recipe featuring Styrian Golding hops
Pale - derived from an American pale ale, this beer uses significant quantities of Cascade hops to give a uniquely hoppy taste
Black - a real marmite beer. Originally derived from a classic porter recipe, this beer has an incredibly strong smokey flavour that some people love and others hate...
How many iterations of each beer have so far been brewed?
We are now up to 16 iterations.
How much of a gap is there between versions?
Sometimes the difference is quite subtle, other times the difference can be quite significant. At the moment the algorithms favours taking larger steps between versions as that way it learns more at a higher level. As we continue it will naturally start refining those recipes more precisely.
Do you find it breeds customer loyalty?
We believe this is the future of product development: consumers are demanding a closer relationship to the products they consume. If you've heard of 'the Ikea effect' - the way in which someone feels closer to the something they've built themselves - this is one of the key principles we operate under. By doing that, customers are naturally more loyal to the brand and the product.
How many beers do you produce?
This is changing rapidly, so I don't have an accurate number at the moment.
Is your beer the best in the UK?
Our algorithms are continually learning to improve the beer recipes; however, our whole ethos goes against the idea of a single 'best' beer. The amount to which someone enjoys a beer is highly dependent on them as a person, the moment in which they are in, the food they are eating, the people they are with and so forth. Added to that, taste is always changing - take for instance the recent influx of highly hopped American IPAs into the UK market. Our brand is all about change and dynamically adapting to consumer preference. Do we think we are the best at doing that? Absolutely!
How profitable is the business? Can you divulge figures? How do you see the business developing?
We can't divulge financials at this point. We have had huge international interest in the beer – from the UK , US, Japan , South America and Europe, for example. At the moment we are working out how best to scale to keep up with demand. We fundamentally believe that the future of product development is a closer relationship between consumer and innovation and already have plans to expand into other categories soon.