Kan AI smage ting som kaffe eller chokolade med sensorer og forbedre deres smag til menneskelig konsum ?

AI kan assistere med at analysere og forbedre smagen af fødevarer som kaffe eller chokolade ved at anvende sensor-data kombineret med maskinlæringsmodeller trænet på kemisk sammensætning og menneskelig sensorisk feedback. Elektroniske tunger og gassensorer registrerer smagsstoffer, og AI korrelerer disse signaler med opfattede smagsprofiler, hvilket muliggør justeringer i formuleringen for at forbedre smag, aroma og den samlede acceptabilitet. Selvom AI ikke kan "smage" som mennesker, accelererer det produktudviklingen ved at forudsige, hvordan ændringer i ingredienser eller forarbejdning påvirker sensoriske kvaliteter. Denne tilgang anvendes i stigende grad inden for fødevidenskab til at optimere smag og kvalitet.

— Opdateret 15. maj 2026 · Kilde: Nature Food, 2023

Background

AI-assisted taste engineering relies on electronic tongues and gas chromatography sensors that detect volatile organic compounds and non-volatile taste-active molecules. These instruments quantify compounds such as furfuryl acetate (ethyl-maltol-like aroma), 2-ethylphenol (smoky/phenolic notes), theobromine, and trigonelline in coffee; and theobromine, phenylethylamine, and various Maillard reaction products in chocolate. Machine learning models trained on both GC-MS or LC-MS chemical fingerprints and human sensory panels (e.g., trained assessors scoring attributes such as bitterness, acidity, sweetness, astringency, and aroma intensity) learn to predict perceived flavor profiles from raw chemical data. Partial least squares regression and deep neural networks are commonly used to map sensor outputs to human ratings, enabling rapid “virtual tasting.” Industry workflows iteratively adjust roast profiles or conching times, then re-measure, reducing sensory evaluation cycles from weeks to days. Similar approaches are reported for modulating bitterness in cocoa by varying fermentation time or adding natural bitter-masking peptides derived from enzymatic hydrolysis of milk proteins (Cao et al., Sci. Rep. 2022). In coffee, controlling 3-methylbutanal and guaiacol concentrations via controlled roasting can shift profiles from “green/grassy” to “caramel/smoky,” aligning with consumer preference clusters identified by preference mapping in supermarket panels (Nature Food, 2023). Sensor arrays and electronic noses have demonstrated classification accuracy above 90% for roast level and origin in Arabica lots (Romano et al., Food Chem., 2021).