These new ultrafast artificial intelligence algorithms (inspired by slow brain dynamics)have the potential to outperform all learning rates achieved by state-of-the-art learning algorithms to date. Through this technology, the scientists, from Bar-Ilan University, are aiming to rebuild a bridge between neuroscience and advanced artificial intelligence algorithms, one that was proposed some 70 years ago.
Discussing what is involved, lead researcher Professor Ido Kanter says: “The current scientific and technological viewpoint is that neurobiology and machine learning are two distinct disciplines that advanced independently.” The Israeli researchers have challenged this dichotomy.
The researchers contend there is merit in studying the slower human brain even in the era of super-fast computers, since the human brain is still capable of doing and perceiving many things that artificial intelligence is incapable of performing. Although the human brain computes at a much slower rate than modern computers, is extremely fast and efficient. Capturing this efficiency could help to advance AI in new ways.
An example is with processing asynchronous inputs and the being able to refine the information, such as assessing cars moving at different speeds on a road, moving along side pedestrians who are also walking at different speeds along a street. A conventional artificial intelligence algorithm is based on synchronous inputs, which means AI can struggle to interpret objects moving at different speeds.
Based on this, the researchers think that AI can be trained in different ways so that it can function more like a human brain and this different way of acting and perceiving the world will lead to faster ways of assessing moving objects, which will have advantages for progressing things like autonomous vehicles.
The researchers conclude that the brain has “to be once again at the center of future artificial intelligence. “The findings have been reported to the journal Scientific Reports. The research is titled “Biological learning curves outperform existing ones in artificial intelligence algorithms.”
