Artificial intelligence will surprise bite into food development

Artificial intelligence is now entering the territory of unexpected: evaluating the sensory qualities of chocolate brownies. A study by the University of Illinois Urbana-Champaign University shows that large language models such as Chatgpt can simplify food development and potentially reduce manufacturers’ costs, thereby navigating an increasingly competitive market.

Food scientist Damir Torrico, an assistant professor in the Department of Food Science and Human Nutrition, found that Chatgpt can effectively produce sensory descriptions of hypothetical brownie formulas, including those containing unconventional ingredients such as mealworm meal and fish oil.

“Sometimes, relying on human testers slows down the process, especially when multiple product prototypes need to be evaluated simultaneously,” Torrico explains in conjunction with the research publication. The study highlights how AI serves as a preliminary screening tool before companies invest in expensive human sensory groups.

“chatgpt tries to always see the good side of things,” Torico notes.

The experiment involved 15 different brownie recipes, from standard recipes to recipes with unusual ingredients substitutes. Chatgpt is prompted to act as an experienced taster and provide a detailed sensory description for each recipe. The researchers then analyzed these responses using natural language processing techniques to identify patterns in emotion and descriptor use.

Interestingly, AI demonstrates what researchers call “hedonistic asymmetry”—a tendency to provide overwhelmingly positive assessments even for formulas that may be negatively responded by human consumers. This phenomenon reflects the mental patterns of human beings, and we tend to describe beneficial projects more actively.

“chatgpt tries to always see the good side of things,” Torico notes.

This positive bias represents both limitation and opportunity. While this suggests that AI has not been able to accurately predict consumer rejection of abnormal ingredients, it suggests that language models can produce consistent sensory descriptions, which may help food scientists narrow their focus during development.

Food manufacturers have a great financial impact. Traditional sensory assessments require recruitment and training panels for human testers, which becomes increasingly expensive as the number of products changes increases. By using AI as a preliminary screening tool, companies can perform hundreds of virtual recipes before advancing only the most promising candidates for human testing.

This approach may be particularly valuable for startups and smaller companies developing new food products. Reducing the number of physical prototypes required before market testing can significantly reduce R&D costs and speed up time to market in a competitive food space.

For investors tracking FoodTech innovations, this study shows a new direction for AI to be used in product development. Although most of the attention is focused on generative AI in marketing and logistics, its application in product formulation and sensory science opens up new avenues of efficiency.

Although Chatgpt cannot physically taste food, it produces a consistent sensory description based on the ingredient list alone.

Sentiment analysis of AI responses shows specific patterns. According to the study, “in general, ‘trust’, ‘expect’ and ‘joy’ are the most commonly expressed emotions found in Chatgpt responses. On the other hand, ‘disgust’, ‘fear’, and ‘trust’ are the least common emotions in these responses.” Further analysis showed that the standard brownie formula was associated with descriptors such as ‘texture’ and ‘slightest’, while common ingredient alternative formulas were associated with terms such as ‘chocolate’, ‘fudgy’ and ‘flavel’.

The study highlights current limitations and future potential. Although Chatgpt cannot physically taste food, it produces a consistent sensory description based on the ingredient list alone. This shows that AI may become increasingly valuable in food science applications through further refinement and training of professional data sets.

“Using AI can provide general insights into which products can be considered for further testing and which products should not be put into this long process,” Torrico said. “I can see that Chatgpt was developed to help the industry’s sensory evaluation.”

The next step in this study will involve comparing AI-generated sensory descriptions with sensory descriptions of human panels to verify accuracy and develop more complex predictive models. Torrico plans to perfect the experiment by training Chatgpt to respond with vocabulary similar to human descriptive panels.

As competition among alternative proteins and novel food ingredients intensifies, tools that can quickly evaluate multiple formulations without physical production may become an indispensable competitive advantage. The study shows that while robots won’t replace human taste testers anytime soon, they may help ensure that only the most promising formulas ever attracted human taste buds.

The study was published in the Journal of Food and adds to growing evidence that AI applications in the food industry go far beyond supply chain optimization and consumer marketing.

Related