When machine learning invents an IPA beer and creates artificial hops

Blonde, white or amber, more or less hoppy … beers in all their diversity are fashionable in the land of chocolate. In a few years, the Swiss market for this ancestral drink has completely transformed with the proliferation of craft breweries. In 2000, there were only 81 breweries in Switzerland, according to official figures published by the Confederation. In 2015, there were 623. This number has now exceeded a thousand (1,132 in 2019). Note that these figures do not take into account producers who do not brew more than 400 liters per year.

Given the abundance of beer on offer in Switzerland, how do you stand out from the crowd? For MN Brew, the answer lies in two words: artificial intelligence. This Lucerne micro-brewery collaborated with the software publisher Jaywalker Digital and researchers from the Lucerne University of Applied Sciences to develop an IPA-style beer (Indian Pale Ale) based on a recipe proposed by the “Brauer” tool. AI ”, which runs on machine learning. To develop their algorithmic model, the researchers started with 157,663 recipes. This data has undergone a major cleanup. To reduce complexity, recipes with rare ingredients were put aside, as well as those that took too long to make and those that simply did not inspire the partner brewery.

Recognition of patterns in beer recipes

67,345 recipes were ultimately used to train Brauer AI, who learned the many ways to best combine 15 malt varieties and 1,648 hop varieties. “After a few hours, the AI ​​began to recognize repeating patterns in the beer recipes. By internalizing all the recipes, the program was able to generate new almost ready-to-use recipes, ”explains Marc Bravin, research assistant at the University of Applied Sciences Lucerne. In order for Brauer AI to generate a recipe, it must start from a style (IPA, white, amber, etc.) either indicated by the brewer or chosen at random by the AI. The latter then offers a list of malts and their proportion in the brew. Suggestions for suitable hop varieties and cooking times follow. The relevance of each recipe is verified by a human brewer, who will make adjustments if necessary. The project partners are therefore keen to emphasize that their tool does not consist in excluding humans from the brewing process, but in having them collaborate with AI to find truly unique recipes.

AI and synthetic biology to do without hops

However, using artificial intelligence to make beer is not a Swiss exclusive. In the United States, researchers in synthetic biology are going even further by using AI to find out how to modify molecules at the DNA level, in order, for example, to produce very hoppy IPA beers. .. but without hops. The team of scientists at Lawrence Berkeley National Laboratory developed the Automated Recommendation Tool (ART), which draws on machine learning and probabilistic modeling techniques to quickly predict the results of a change in a gene or a protein, without it being necessary to fully understand the complex intricacies of the biological system of the molecules concerned.

In the case of the hop-free IPA, the researchers used their tool to find out how to modify the ethanol-producing yeast used to brew beer in order to synthesize metabolites, linalool and geraniol, which give the flavor. hops. ART has been able to guide scientists to achieve a particular level of linalool and geraniol so as to match the taste of known beers. According to the researchers, their technique is economically advantageous because the cultivation of hops is intensive in water and energy, and its taste is very variable from one culture to another. It remains to be seen whether fans are ready to taste genetically modified beers …

Generating new beer flavors is just one example of applications of artificial intelligence in the field of taste. Not long ago, the Geneva company Firmenich has for example developed a flavor of grilled beef by relying on an AI.

Leave a Comment