Chinese scientists have developed a novel neural network that enables artificial intelligence (AI) to form concepts from raw sensory data like sight and sound, simulating a fundamental aspect of human cognition, according to a study published recently in the journal Nature Computational Science.

A remarkable capability of the human brain is to form more abstract conceptual representations from sensorimotor experiences and flexibly apply them independent of direct sensory inputs.

However, the computational mechanism underlying this ability was previously poorly understood. This meant that large language models were fundamentally limited by their dependence on pre-existing linguistic data, making them incapable of spontaneously generating new concepts from experiential learning.

The researchers from the Institute of Automation of the Chinese Academy of Sciences and Peking University proposed their new neural network framework, named CATS Net, as a means to overcome these limitations.

The framework consists of a concept-abstraction module and a task-solving module that can precisely instruct the framework to perform tasks like recognition and judgment when processing visual information, such as images.

The framework can also autonomously generate a vast array of new concepts, building its own unique "concept space." Once the concept space of different AI systems are aligned, they can directly transmit knowledge using these concepts, bypassing the need for retraining on raw data. Notably, this process simulates how humans communicate using language.

Via brain imaging studies, the researchers revealed that the conceptual space constructed by CATS Net aligns closely with human cognitive and linguistic logic, and its operational mode closely matches activities in the human brain's concept-processing areas.

This suggests that the model does more than just imitate brain function, shedding light on the computational mechanisms by which humans form and use concepts in the brain.