Scientists have developed an AI-powered “electronic tongue” that can monitor food and drink for safety issues. The device uses artificial intelligence to distinguish between different coffee blends or detect when food or drink may be going bad. The “electronic tongue” can tell the difference between various coffee blends, signal when juice has spoiled, and detect harmful chemicals in water.
Researchers from Penn State University developed the system, which uses an ion-sensitive field-effect transistor to detect chemical ions. The sensor collects data about the ions in a liquid and turns it into an electrical signal that a computer can interpret. Study co-author Saptarshi Das, an engineer at Penn State University, said they are trying to make an artificial tongue, but the process of how we experience different foods involves more than just the tongue.
It also includes taste receptors that interact with food and send information to the gustatory cortex in the brain. In their system, the sensor acts as the tongue, while AI plays the role of the gustatory cortex. By linking the sensor to an artificial neural network, a machine learning program that mimics how the human brain processes information, researchers were able to process and interpret the data collected by the sensor.
Initially, the researchers programmed the neural network to determine the acidity of a liquid, achieving about 91% accuracy.
AI-driven electronic tongue technology
When they allowed the neural network to define its own parameters, accuracy improved to more than 95%.
They then tested the tongue on real-world beverages. The system could distinguish between similar soft drinks or coffee blends, assess whether milk has been watered down, identify when fruit juice has gone bad, and detect harmful PFAS chemicals in water. Using Shapley Additive Explanations, the researchers determined important parameters the neural network used for its conclusions.
This provides insights into how neural networks make decisions, which is an ongoing challenge in AI research. Das said they found the network looked at more subtle characteristics in the data that humans struggle to define properly. Because the neural network considers the sensor characteristics holistically, it mitigates variations that might occur day-to-day.
This ability to adjust for variations makes the sensor more robust for various applications. By accounting for these variations, the neural network ensures reliability where ion-sensitive field-effect transistors typically falter. “We figured out that we can live with imperfection,” Das added.
“And that’s what nature is — it’s full of imperfections, but it can still make robust decisions, just like our electronic tongue.”
