The human brain is the most intricate and energy-efficient computer ever made, and scientists are using it as inspiration to build a new generation of supercomputers. Researchers are examining several nonbiological materials whose properties could be customized to show signs of learning-like behaviors in some of their early attempts to create computers inspired by the human brain. These components might serve as the foundation for technology that, when combined with fresh software algorithms, could produce artificial intelligence (AI) that is more potent, practical, and resource-conserving. In a recent study, Purdue University researchers subjected oxygen-deficient nickel oxide to short electrical pulses and elicited two distinct electrical reactions that are comparable to learning. Professor Shriram Ramanathan of Rutgers University claims that the outcome is an entirely electric-powered device that exhibits these learning traits. At the time this work was completed, Ramanathan was a professor at Purdue University. The Advanced Photon Source (APS), a user facility for the Office of Science of the U.S. Department of Energy (DOE) is located at the DOE’s Argonne National Laboratory.
The material “gets used to” being mildly zapped, which leads to the first reaction, habituation. The researchers found that although the material initially exhibits higher resistance, it quickly adapts to the electric stimuli. Habituation is similar to what happens when you live close to an airport, according to physicist and APS beamline scientist Fanny Rodolakis. You may initially think, “What a racket,” but later you won’t even notice it. When a higher electrical dose is applied, the material exhibits the other response, known as sensitization. According to Rodolakis, “with a larger stimulus, the material’s response grows rather than dwindles over time.” It’s similar to watching a terrifying movie and then hearing someone shout, “Boo!” from around a corner you can practically see it jump. Almost all living things exhibit these two qualities, according to Ramanathan. They truly are a core component of intellect. The basis for a phase change in the material is provided by these two characteristics, which are governed by electron quantum interactions that are beyond the scope of conventional physics. “A liquid turning into a solid is an example of a phase transition,” Rodolakis remarked. The material we’re looking at is on the cusp, and the conflicting interactions occurring at the electrical level can readily be swung one way or the other by tiny stimuli.
Applications of brain-inspired computing require a system that can be entirely controlled by electrical signals, according to Ramanathan. Hardware will be able to assume part of the burden of intelligence if it is able to modify materials in this way, he said. A crucial step toward energy-efficient computing is the incorporation of intelligence into hardware via quantum characteristics. The stability-plasticity conundrum, a problem in the development of AI, can be solved by understanding the difference between habituation and sensitization. On the one hand, artificial intelligence systems can frequently be unwilling to adapt to new facts. However, they can lose part of what they’ve already learned when they do. By developing a material that has the ability to habituate, researchers can educate it to disregard or forget unnecessary information and therefore achieve more stability, while sensitization could enable plasticity by teaching the material to recall and incorporate new information.
According to Rodolakis, artificial intelligence frequently has trouble picking up new information and storing it without overwriting previously stored data. “Too much stability prevents AI from learning, while too much plasticity can cause catastrophic forgetting,” the study concluded. The current study’s focus on the nickel oxide device’s compact size had several significant benefits. “This type of learning had not previously been done in the current generation of electronics without a large number of transistors,” Rodolakis added. The possibility of developing neuromorphic circuitry is significantly impacted by the fact that this single junction system is the smallest system to date to exhibit these characteristics.
