Chinese scientists have developed a neural network capable of simulating human-like concept formation, marking a significant leap in artificial intelligence (AI) research. Published this year in Nature Computational Science, the study introduces CATS Net—a framework designed to overcome limitations of traditional language models by enabling AI to autonomously generate abstract concepts from sensory data.
The system, created by researchers from the Institute of Automation of the Chinese Academy of Sciences and Peking University, features dual modules: one for concept abstraction and another for task execution. This allows AI to perform recognition and judgment tasks while building its own "concept space"—a digital representation of cognitive logic comparable to human reasoning patterns.
Brain imaging comparisons show CATS Net's operational patterns align closely with neural activities in human concept-processing regions. This breakthrough not only advances machine learning but provides new insights into the computational mechanisms behind human cognition. Unlike conventional models reliant on pre-existing data, CATS Net enables knowledge transfer between AI systems through shared conceptual frameworks, mirroring how humans communicate complex ideas.
Experts suggest this innovation could reshape industries ranging from education technology to cross-cultural AI applications, offering tools for more intuitive human-machine collaboration. The development underscores the Chinese mainland's growing role in foundational AI research while opening new pathways for understanding the brain's conceptual architecture.
