Researchers utilize AI to enhance beer flavor
Researchers use AI to improve the taste of beer, analyzing data from 250 Belgian beers to predict and enhance flavors for both alcoholic and non-alcoholic varieties.
Whether you enjoy a fruity lambic or a rich Trappist ale, Belgian beers have long been renowned for their diversity, quality, and heritage. Now, researchers claim they've utilized artificial intelligence to further enhance these brews.
According to Prof. Kevin Verstrepen from KU Leuven university, who led the research, AI could assist in unraveling the intricate relationships involved in human aroma perception.
"Beer, much like other food products, contains numerous aroma molecules that are detected by our taste buds and olfactory system. Our brain then synthesizes these inputs into a cohesive perception. However, these compounds interact with each other, meaning that our perception of one can be influenced by the concentrations of others," he explained.
In their article published in Nature Communications, Verstrepen and his team detailed their analysis of the chemical composition of 250 different Belgian beers representing 22 distinct styles, including lagers, fruit beers, blond ales, West Flanders ales, and non-alcoholic beers.
The study examined various characteristics of these beers, such as alcohol content, pH levels, sugar concentrations, and the presence and quantity of over 200 different flavor-related compounds. These compounds included esters, produced by yeast, and terpenoids from hops, both of which contribute to fruity flavors.
To assess the beers' flavors, a panel of 16 tasters sampled and rated each of the 250 beers based on 50 different attributes, such as hoppy notes, sweetness, and acidity. This evaluation process spanned three years.
Additionally, the researchers analyzed 180,000 consumer reviews of different beers from the online platform RateBeer. They found that while factors like price influenced overall consumer ratings, other aspects such as bitterness, sweetness, alcohol content, and malt aroma closely mirrored the ratings and comments provided by the tasting panel.
Verstrepen noted that even minor changes in chemical concentrations can significantly impact flavor, especially when multiple components are involved. He also mentioned a surprising finding: certain compounds, traditionally considered undesirable in higher concentrations, can actually enhance the flavor when present in smaller amounts and in combination with other aroma compounds.
The team utilized machine learning, a form of artificial intelligence, to construct models that could predict the taste and appreciation of a beer based on its composition. These models were developed using various datasets.
They applied the results of these models to improve a commercial beer by adding substances identified as important predictors of overall appreciation, such as lactic acid and glycerol.
The tasting panel's feedback indicated that these additions enhanced ratings for both alcoholic and non-alcoholic beers, improving attributes like sweetness, body, and overall appreciation.
Despite their potential, the models have limitations, as they were only trained using datasets from high-quality, commercial beers. Verstrepen suggested that the most significant application of these models could be in refining non-alcoholic beers.
However, he emphasized that the skill of brewers remains crucial, as the AI models only predict chemical changes that could optimize a beer. It is still the brewer's responsibility to implement these changes starting from the recipe and brewing methods.
