Scientists have figured out how to use AI to improve the taste of beer

The researchers decided to use the neural network to help manufacturers develop new food and drink products or adjust existing recipes to better suit consumer tastes. The use of AI promised to save significant amounts of time and money in testing. The goal was to create an AI training dataset that objectively reflected overall flavor scores, including hop, yeast, and malt notes, beer appearance, aroma, and overall drink quality.

It took the researchers five years and chemical analysis of 250 commercial Belgian beers to properly train the neural network. The researchers combined these analyzes with subjective ratings from a qualified tasting panel and added results from 180,000 reviews of the same beers on the popular online platform RateBeer. This large dataset, which links chemical data to sensory characteristics, was used to train 10 AI models to accurately predict the taste, smell and feel of a beer, and the likelihood that a consumer would give it a high rating.

To compare the models, the researchers split the data into a training set and a test set. After the model was trained on data from the training set, they assessed its ability to predict beer ratings using the test set. The results concluded that all models were better than a trained panel of human experts at predicting beer ratings from RateBeer.

Using these models, the researchers were able to identify specific compounds that contribute to consumer ratings of beer: People were more likely to rate a beer highly if it contained these specific compounds. For example, models predicted that adding lactic acid, which is present in sour, tart-tasting beers, could improve other beers by making them fresher.

“We asked the models to analyze the beer and then asked them how can we make it better? says University of Leuven professor Kevin Verstrepen. – Then we brought in [предложенные ИИ] changes to beer by adding flavor compounds. And lo and behold, once we did the blind tasting, the beer improved and became more popular.”

One interesting application of this research, he says, is that it could be used to make better non-alcoholic beer, a major challenge for the beverage industry. The researchers used the model’s predictions to add a mixture of compounds to a non-alcoholic beer that human tasters rated significantly higher than its previous incarnation.

According to food science professor Carolyn Ross, who was not involved in the study, the use of AI could be extremely useful in studying the composition and nutritional value of foods, as well as adapting ingredients to different populations. For example, she says, older people find complex combinations of ingredients less appealing. “We get to explore so much, especially when we study different populations and try to develop specific products for them,” she says.