Connect With Us










 

Machine-learning helps start-up brew better beer

Steve Rogerson
July 19, 2016


 
The world’s first beer brewed by artificial intelligence, which improves itself from customer feedback, has been launched by a British firm. Intelligent X uses a complex machine-learning algorithm to determine what consumers like about its beers, then brews new versions that are more finely tuned to people’s tastes.
 
After trying one of the company’s four bottle-conditioned beers, consumers give their feedback on the flavours to the firm’s algorithm, which is called ABI for automated brewing intelligence.
 
The algorithm uses this customer feedback data, which are provided via a Facebook Messenger bot, to tell the master brewer what to brew next. ABI also has a bank of wildcard ingredients, such as adding fruit to a recipe, in a bid to create beer that pushes the boundaries of what’s possible within craft brewing.
 
Using a combination of reinforcement learning and Bayesian optimisation, ABI learns from experience by being rewarded when it does something good. The algorithm's ultimate objective is to keep learning and win a major beer competition, such as Camra’s Champion Beer of Britain. Camra is the UK consumer group Campaign for Real Ale.
 
Intelligent X is s partnership between Intelligent Layer and 10x, two companies that met while working in an east London WeWork co-working space. Intelligent Layer is a machine-learning firm, founded by Rob McInerney, a former Oxford University machine learning PhD. While 10x is an innovation and creative agency, founded by Hew Leith, a former M&C Saatchi director.
 
The beers, which have evolved eleven times so far, have been such a hit with customers that the two firms decided to spin Intelligent X out into a separate company, with dedicated staff.
 
The bottle-conditioned beers, which come in exclusive pirate black bottles, are available for £4.50 each at UBrew, Bermondsey. They come in four styles:

  1. Golden AI – the origins are from a classic British golden ale recipe featuring Styrian Golding hops.
  2. Amber AI – derived from a British bitter, which has a darker appearance and stronger flavour, with a hint of grapefruit for a fresh taste of summer.
  3. Pale AI – derived from an American pale ale, this beer uses significant quantities of Cascade hops to give a uniquely hoppy taste.
  4. Black AI – a real marmite beer. Originally derived from a classic porter recipe, this beer has an incredibly strong smoky flavour that some people love and others hate.
“Contrary to popular belief, we don’t believe AI is going to take everyone’s jobs,” said Leith. “We believe the future is a place where AI augments humans’ skills. In this case we’re using AI to give our brewer superhuman skills, enabling them to test and receive feedback on our beer more quickly than ever before. This means we can respond to consumers’ changing tastes faster than traditional brewers.”
 
McInerney, whose machine learning PhD thesis was on decision making under uncertainty, said: “What Google’s Deepmind has achieved with AlphaGo is extremely impressive. However, what happens if you don’t have millions of data points to train a deep learning algorithm? Clearly we can’t make a million beers, so we need to carefully manage uncertainty in the model so that what information we do have is used very efficiently. Instead of deep networks we use a Bayesian non-parametric approach, which is better suited to this type of problem. We also work very hard to learn as much as possible from the brewer directly, so this really becomes a joint effort between man and machine.”
 
Intelligent X plans to open-source every recipe created by the algorithm, allowing others to recreate their favourite beers at home. As each batch will be unique, there’s no need to keep the recipe under lock and key.
 
“Our beer is one of the world's first uses of AI to improve physical products,” said Leith. “We think it’s an area that has huge potential. Imagine emotive products like perfume, coffee or chocolate which are finely tuned to people’s tastes by machine learning algorithms."