JERUSALEM, March 29 (Xinhua) -- A new Israeli study suggests that artificial intelligence is powerful not just because it grows larger in scale, but because its parts learn to work together, Bar Ilan University said in a statement on Sunday.
The study, published as a preprint on SSRN (Social Science Research Network), draws on a concept from physics known as "More is Different," first proposed by physicist Philip W. Anderson in 1972. The idea holds that when many parts combine, they can produce new and unexpected behaviors.
Researchers found that AI systems appear to follow this principle. As they are trained, different components within the systems begin to specialize, each taking on distinct roles that contribute to solving tasks more effectively.
The findings suggest that coordination among these components plays a key role in improving performance, with the combined system achieving results beyond what individual parts can do alone.
The study indicates that advances in artificial intelligence may depend not only on increasing model size but also on improving internal organization, which could lead to more efficient systems.
Researchers said the results may also have implications for neuroscience, pointing to similarities with how the brain processes information through networks of specialized and cooperating neurons. ■
