Israeli scientists study secrets of human brain to bring AI to next level

Researchers at Bar-Ilan University are exploring the bridge between neurosciences and machine learning again.

Artificial intelligence (photo credit: INGIMAGE)
Artificial intelligence
(photo credit: INGIMAGE)
The concept of artificial intelligence started developing decades ago from the idea of giving machines a “human brain.” The connection between them was crucial, but only on a philosophical level as computers became more advanced and very little of their way of functioning could be seen as mirroring a human mind.
But researchers at Bar-Ilan University are exploring the bridge between neurosciences and machine learning again. They have demonstrated a new accelerated brain-inspired learning mechanism.
“The main idea behind artificial intelligence was to imitate brain functions using artificial machines and computers,” Prof. Ido Kanter, lead author of the study published in the journal Scientific Reports on Tuesday, told The Jerusalem Post. “The field was initiated in 1949, and the main idea was to adopt the model of the connection between neurons for the purpose of machine learning. However, this connection disappeared a few years after because no relations or ideas for advanced machine learning came from experimental neurosciences.”
Team of Bar Ilan University works on connection between neuroscience and artificial intelligence (Credit: Courtesy Bar Ilan)Team of Bar Ilan University works on connection between neuroscience and artificial intelligence (Credit: Courtesy Bar Ilan)
One of the challenges in this perspective was that the brain adaptation was considered very slow and complicated compared to computers’ adaptation, with every learning step believed to last tens of minutes or even more, while it took a nanosecond with a computer, he said.
The question of how on a practical level in many situations, for example, driving a car, the performance of the human brain is comparable if not superior to the state of the art of artificial-intelligence technology remained.
All, or at least the vast majority, of the current achievements of computers are attributed to their speed, Kanter said.
“For example, if I was to give a company building autonomous vehicles a slower computer, they would not be interested,” he said.
Kanter’s team focused on experimental neurosciences and theoretical work, combining them.
Their experimental setup allowed them to stimulate neurons intracellularly and extracellularly and precisely control the timing of the reactions.
“We found out that, for example, the brain learning process is more effective by showing an image 10 times in a minute than 1,000 times in a month,” Kanter said, adding that accelerated learning is more efficient than slow learning, contrary to what was assumed.
Understanding better the deep-learning mechanism of the brain and then combining it with the speed of a computer could be the key to reach unparalleled results, he said.
“If we were able to implement the slow biological deep-learning mechanism in our brain on a very fast computer, the sky would be the limit,” Kanter said. “Today we are not really imitating the brain, but rather using some general ideas inspired by it. Many of its mechanisms are still to be revealed. If we managed to do so, we would be able to implement them on a computer.”
The scientists are now working on bringing the research to the next level.
Among the questions they are exploring is how much can computers achieve given a limited amount of information.
“If given a specific problem with limited information, current machine algorithms can achieve 50%, and we as humans can achieve 70%,” he said. “Maybe if we were to understand more about the brain’s secrets, computers could reach 80%.”
“With this research, we have revealed one hidden mechanism of the brain,” Kanter said. “But I’m confident that there are more secrets to discover, as long as we ask the right questions, different from what people did before.”