A project in Germany, involving participants from both the private and public sectors, has conducted a series of important studies on the use of Artificial Intelligence (AI) for evaluating wine. The initial findings indicate that the project holds great promise for the future.
In 2019, an organization called Genie contacted Oenology Professor Dominik Durner to develop an "Artificial Intelligence Sommelier." At the time, AI research was beginning to make waves, and Genie consulted Prof. Durner on how AI could be used to identify aromas. Finding the idea worth exploring, Durner collaborated with two academic institutions—Trier University and the Fraunhofer Institute—as well as three private companies. The German Ministry of Agriculture later supported the project as part of its "AI in Agriculture" initiative.
The initial goal was to determine which aromas could be detected by electromagnetic sensors and whether these sensors could achieve a level comparable to the human sense of smell. Early results were disappointing, as ethanol interfered with the measurement tools. As a result, the project's framework was adjusted to task the sensors with detecting wine faults. To achieve this, researchers selected 100 wines with a common fault: the sulfur effect commonly referred to as "rotten egg smell." After identifying the fault, the sensors were exposed to fault-free wines to determine differences. Prof. Durner likens this process to the myth of Sisyphus, in which the condemned figure endlessly rolls a boulder uphill, only for it to roll back down each time. Similarly, training AI involves repeated measurements over years. Despite this, Prof. Durner finds the progress of the three-year study promising, emphasizing that the application could extend beyond wine. "The human nose cannot always assess consistently, which is one of its major flaws. However, an AI system trained to interpret scent data from sensors can evaluate consistently with the same performance every time," he notes.
At its current stage, the AI system has achieved sensitivity equal to about one-tenth that of the human nose. However, the system can already detect sulfur faults, such as the "rotten egg smell," even before they are noticeable to human noses. The research team is confident that the system will achieve highly reliable results in the future. Once that level is reached, warnings provided by the system during the early stages of fermentation could enable much healthier production planning. They are optimistic that fingerprint-like evaluations of wine will provide effective insights when compared to its performance at later stages. However, it is clear that it will take much longer for AI to match the nuanced quality assessments of professional wine experts.
The most critical requirement for AI is data. Therefore, collecting not hundreds but hundreds of thousands of data points is essential for reliable results. To achieve this, the team partnered with UC Davis in California to involve American researchers in the project, thereby enriching the data pool.
The project aims to develop a portable device capable of providing instant results during production, eliminating the need to wait for laboratory analyses. The research team also envisions applications beyond wine, such as replacing trained dogs with AI systems for detecting drugs and explosives.
Meanwhile, Prof. Durner dismisses concerns that the tools developed from this research will standardize wine, emphasizing his belief that wines will always retain their tailored, bespoke identities.
