For artificial intelligence to become smarter, it must first be as intelligent as one of the simplest creatures in the animal kingdom: seaweed.
New research has found that the material can mimic the most important intelligence characteristics of a snail. The discovery is a step toward building hardware that can help boost artificial intelligence more efficiently and reliably in technology ranging from self-driving cars and surgical robots to social media algorithms.
The study, which will be published this week Publications of the National Academy of Sciences, conducted by a team of researchers from Purdue University, Rutgers University, the University of Georgia, and Argonne National Laboratory.
“By studying sea locks, neuroscientists discovered the hallmarks of intelligence that are essential to the survival of any organism,” said Shriram Ramanathan, a professor of materials engineering at Purdue. “We want to leverage the mature intelligence of animals to accelerate the development of AI.”
The two most important signs of intelligence that neuroscientists have learned from marine ethos are habituation and sensitization. Getting used to getting used to a stimulus over time, like tuning sounds as you drive the same route to work every day. Sensitization is the opposite – it responds strongly to a new stimulus, such as avoiding poor food in a restaurant.
It is really difficult for artificial intelligence to learn and store new information without replacing the information they have already learned and stored, and this problem is called the “stability-plasticity dilemma” by researchers studying brain-influenced computing. Getting used would allow artificial intelligence to “forget” unnecessary information (achieve more stability), while sensitization could help preserve new and important information (allow for plasticity).
In this study, researchers found a way to demonstrate habituation and sensitization to quantum material in nickel oxide. The material is called “quantum” because its properties cannot be explained by classical physics.
If quantum material could reliably mimic these forms of learning, it may be possible to build artificial intelligence directly into hardware. And if artificial intelligence could work through both hardware and software, it could be able to perform more complex tasks using less energy.
“We are basically emulating experiments on marine locks made in quantum materials to understand how these materials can be of interest to artificial intelligence,” Ramanathan said.
Neuroscience studies have shown that a snail shows habit when it stops pulling its gills equally in response to a siphon knocking. But an electric shock to his tail makes that gill retract much more dramatically and shows sensitization.
For nickel oxide, “torture removal” is an increased change in electrical resistance. The researchers found that repeated exposure of the material to hydrogen gas causes a change in the electrical resistance of nickel oxide over time, but the introduction of a new ozone-like stimulus greatly increases the change in electrical resistance.
Inspired by these findings, a research team led by Kaushik Roy, Purduen’s Edward G.Tiedemann Jr., a respected professor of electrical and information technology, modeled the behavior of nickel oxide and built an algorithm that successfully used these habituation and sensitization strategies to classify data points into clusters.
“The stability-plasticity dilemma has not been solved at all. But we have shown a way to deal with it based on the behavior we observe in quantum material,” Roy said. “If we could turn material that learns this way into hardware in the future, artificial intelligence could perform tasks much more efficiently.”
In practice, the use of quantum materials as artificial intelligence equipment requires researchers to find out how habit and sensitization can be applied to large-scale systems. They also need to determine how the material can respond to stimuli when integrated into a computer chip.
This study is the starting point for guiding the next steps, the researchers said. In addition to the experiments performed on the rupture, the Rutgers University team made detailed theoretical calculations to understand what happened in nickel oxide at the microscopic level to mimic the intelligence properties of the sea block. The properties of the nickel oxide sample were characterized by the Argonne National Laboratory and the University of Georgia measured the conductivity to further analyze the behavior of the material.