Belgian scientists have trained an AI model to predict if people will like certain beers and how to improve the recipe.
ADVERTISEMENTFrom lagers and blonds to lambics, there is a wide range of different beers to choose from.
But could artificial intelligence (AI) help predict if a specific beer recipe will be appreciated by consumers before they even try it?
A team of scientists at KU Leuven, a university in Belgium, say that it can.
The researchers gathered a trained tasting panel of 16 people and asked them to try 250 commercial beers of 22 different styles such as lagers, blonds, and stouts.
Participants rated the beverages on 50 attributes, including different hop, malt, and yeast flavours, off-flavours and spices.
Researchers then gathered 180,000 public reviews of the same beers on RateBeer, an online consumer rating platform to complement the data from the panel.
The beers were also carefully analysed for their composition. For each beer, the scientists measured 226 different chemical properties, including alcohol content, pH, and sugar concentration and more than 200 aromatic compounds.
The researchers used these large datasets "to develop predictive models that link chemical data to sensorial features," they explain in the study.
They published their findings in the journal Nature Communications.
Training an AI to be a taster
They were able to train an AI model to predict a beer's flavour and whether it would be liked based on its chemical composition.
With this data, scientists were able to improve the taste of an existing commercial Belgian beer by adding certain aromas predicted by the model to increase its quality.
The modified beer scored better than the original in blind tastings.
“The spiked beers were found to have significantly improved overall appreciation among trained panellists," the researchers concluded, with panellists noting "increased intensity of ester flavours, sweetness, alcohol, and body fullness".
Tasters also rated an improved sample of non-alcoholic beer higher.
The challenge of predicting food taste
The study points out that predicting the taste and appreciation of foods from their chemical properties remains complicated.
One of the main obstacles is the high number of chemical substances that interact and influence taste.
"Flavour perception is highly complex, resulting from hundreds of different molecules interacting at the physiochemical and sensorial level," the authors say.
The researchers also note that human tastes are conditioned by other factors, such as genetics, environment, culture, and consumer psychology.
ADVERTISEMENTThat is why they used very large datasets that can only be analysed by machine learning models.
“The flavour of beer is a complex mix of aroma compounds. It is impossible to predict how good a beer is by just measuring one or a few compounds. We really need the power of computers,” Michiel Schreurs, the lead author of the study, said in a statement.
“Our biggest goal now is to make better alcohol-free beer. Using our model, we have already succeeded in creating a cocktail of natural aroma compounds that mimic the taste and smell of alcohol without the risk of a hangover,” said Kevin Verstrepen, a professor at KU Leuven.
The team said the study's findings could be expanded to other food products, which may revolutionise how new foods are made.
