This new device could use artificial intelligence to distinguish between different coffee blends or detect when a food or drink might go bad. Have you ever wondered if that old juice box in the back of your refrigerator is still drinkable? A new “electronic language” can tell you this.
An artificial intelligence (AI)-based system can detect issues with food safety and freshness. This also offers insights into how AI makes decisions, researchers reported Oct. 9 in the journal Nature.
To make the tongue, the researchers used an ion-sensitive field-effect transistor that detects chemical ions. The sensor collects information about the ions in the liquid and converts this information into an electrical signal that can be interpreted by a computer.
“We’re trying to create an artificial language, but the process of how we perceive different foods involves more than just language,” Saptarshi Das, an engineer at Penn State University, said in a statement. “We have the tongue itself, which consists of taste buds that interact with food types and send their information to the gustatory cortex, a biological neural network.”
In the new system, the sensor acts as a tongue, while the artificial intelligence plays the role of the gustatory cortex, the area of the brain responsible for taste perception. The team connected the sensor to an artificial neural network, a machine learning program that mimics the way the human brain processes information, to process and interpret the data collected by the sensor.
First, Das and his colleagues fed the neural network various parameters to determine the acidity of a particular liquid. Using these parameters, the neural network determined acidity with approximately 91% accuracy. When they allowed the neural network to set its own parameters for acidity analysis, its accuracy increased to over 95%.
They then tested the tongue on real drinks. They found that the system could distinguish between similar soft drinks or coffee blends, assess whether milk was diluted, determine when juice had gone bad, and detect harmful per- and polyfluoroalkyl substances (PFAS) in water.
Using a method of analysis called “Additive Shapley Explanations,” the researchers were able to determine which parameters the neural network considered most important for making inferences. According to the team, the method could help scientists understand how neural networks make decisions, which remains an open question in artificial intelligence research.
“We found that the network was looking at more subtle aspects of the data, things that we humans have a hard time getting right,” Das said in a statement. he said. “And because the neural network looks at sensor properties holistically, it reduces day-to-day variations.”
Being able to adapt to these changes can help make the sensor more reliable in other applications. In the decision-making process, the neural network takes into account changes that make ion-sensitive field-effect transistors unreliable in some cases.
“We learned that we can live with imperfections,” Das said in a statement. “And that’s what nature is; it’s full of flaws, but it can still make reliable decisions, just like our electronic language.”
Source: Port Altele
