Researchers have begun developing new kinds of beers designed to taste even better thanks to the use of artificial intelligence in the development of recipes.
Until now, consumer taste tests have been the only way to find out which alcoholic and non-alcoholic beers are well received, the team from Belgium explains in the scientific journal Nature Communications.
However the researchers, who fed data on dozens of different Belgian beers into AI, showed that machine learning models could soon help brewers and other manufacturers to concoct new recipes and cater to specific consumer wishes more efficiently and cost-effectively.
Predicting which new food flavours consumers will like is a complex science, the researchers led by Kevin Verstrepen from the Catholic University of Leuven said.
This is mainly due to the fact that there are an immense number of flavour-active chemicals in food. There are also interactions and complex enhancing or diminishing effects on flavour perception. Sweetness and bitterness, for example, mask each other.
The use of trained tasters is common in the development of new food and beverage products, but this incurs high costs. Online evaluation databases, on the other hand, are prone to errors because factors such as the price or reputation of a product can sway results.
The scientists have now recorded over 200 chemical properties of 250 Belgian beers belonging to 22 different beer styles such as blonde, tripel and lager.
These were combined with descriptive sensory profile data from a trained tasting panel of 16 people on hop, malt and yeast aromas, off-flavours and spices, for example, as well as data from more than 180,000 consumer reviews from an online beer review database.
The resulting dataset was used to train and test 10 machine learning models, which then predicted consumer taste and appreciation.
The effectiveness of the most powerful AI approach was tested by implementing predictions to modify an alcoholic and a non-alcoholic commercial beer.
In tastings, the AI beers were said to have received a better overall rating from tasters.
The study confirms that the concentration of flavouring substances does not always correlate with perception - which points to complex interactions that are often overlooked by conventional approaches.
However the research team cautions that the models they used are still immature overall. More comprehensive datasets are crucial for further improvements.
Nevertheless, the team is convinced that AI could provide a basis for the development of new, customised foods with flavours that consumers are more likely to enjoy.
AI-enhanced development processes may then supplement or replace the usual evaluation of new recipes by trained tasters, which is expensive and time-consuming and can deliver varying results.
And what if AI develops a beer that is too hard to stop drinking? Verstrepen's team is already sounding the alarm: We should take care that AI is never used to increase the addictive potential of alcoholic drinks.
