Researchers at Purdue University have built a new piece of hardware that can be reprogrammed on demand through electrical pulses. The team claims that this adaptability would allow the device to take on all of the necessary functions to build a brain-inspired computer. It’s part of an ongoing effort to build AI systems that can learn continuously. “When AI systems learn continually in the environment, they can adapt to a world that changes over time,” Stevens Institute of Technology AI expert Jordan Suchow told Lifewire in an email interview. “We see this, for example, when a fraud-detection system picks up a previously unobserved pattern of fraudulent purchases or when a face-recognition system encounters a person it has never before seen.”
Life-Long Learners
The Purdue researchers recently published the paper in the journal Science. It describes how computer chips could dynamically rewire themselves to take in new data the same way the brain does. The approach could help AI keep learning over time. “The brains of living beings can continuously learn throughout their lifespan. We have now created an artificial platform for machines to learn throughout their lifespan,” said one of the paper’s authors, Shriram Ramanathan, in a news release. The hardware devised by Ramanathan’s team is a small, rectangular device made of a material called perovskite nickelate, which is very sensitive to hydrogen. Applying electrical pulses at different voltages allows the device to shuffle a concentration of hydrogen ions in a matter of nanoseconds, creating states that the researchers found could be mapped out to corresponding functions in the brain. When the device has more hydrogen near its center, for example, it can act as a neuron, a single nerve cell. With less hydrogen at that location, the device serves as a synapse, a connection between neurons, which is what the brain uses to store memory in complex neural circuits. “If we want to build a computer or a machine that is inspired by the brain, then correspondingly, we want to have the ability to continuously program, reprogram and change the chip,” Ramanathan said.
Thinking Machines?
Many modern AI systems adapt to new information when retrained, David Kanter, the executive director of MLCommons, an open engineering consortium dedicated to improving machine learning, said in an email. “The world is an intrinsically dynamic place, and ultimately machine learning and AI must adapt to this,” Kanter said. “For example, a speech recognition system in 2022 that doesn’t ‘know’ about COVID-19 or coronaviruses would be missing a big aspect of the modern world. Similarly, an autonomous vehicle should adapt to changes in streets, bridge closures, or even low temperatures making a road icy.” Although an AI system that learns completely by itself is still mostly a concept, many examples come close, Sameer Maskey, the CEO of AI company Fusemachines, said in an email interview. One of these self-learning systems made the news when an AI system beat a human at a game of Go. “AlphaGo was DeepMind’s first AI to defeat a professional Go player,” Maskey added. “Their games franchises have become stepping stones with every new addition adopting advances towards an AI that keeps learning.” AI systems of the future will seek out the information they need to make good decisions and take appropriate actions, predicted Suchow. These advanced computers will avoid costly mistakes by learning from their own simulations of experience, for example, through “self-play,” where the AI imagines the outcomes of interactions it has with copies of itself. “This is similar to how humans can learn through imagination, foreseeing a bad outcome without needing to experience it directly,” Suchow added. “AI systems will learn more effective strategies for learning, much in the way that a student can direct their time and attention not just to the substantive content of what they are studying, but also to the process of learning itself.”
