China develops AI neural network that mimics human thought

BEIJING: Scientists in China have recently unveiled a breakthrough in artificial intelligence (AI) with the development of a new neural network. This innovation, published in the journal Nature Computational Science, allows AI to mimic a core function of human thought: forming concepts from diverse, unprocessed sensory data, such as images and sounds.

The human brain has a remarkable ability to form abstract conceptual representations from sensorimotor experiences and to apply these concepts flexibly, even without direct sensory inputs.

However, the computational mechanisms underlying this capability were not well understood in the past. As a result, large language models have been fundamentally limited by their reliance on existing linguistic data, preventing them from spontaneously generating new concepts through experiential learning.

To address these limitations, researchers from the Institute of Automation at the Chinese Academy of Sciences and Peking University have proposed a new neural network framework called CATS Net. This framework aims to enhance the ability to generate concepts beyond pre-existing information.

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

The framework can autonomously generate a wide range of new concepts, creating its own unique “concept space.” Once the concept spaces of different AI systems are aligned, they can directly share knowledge using these concepts, eliminating the need for retraining on raw data. This process closely resembles how humans communicate using language.

Researchers conducted brain imaging studies that showed the conceptual space created by CATS Net aligns closely with human cognitive and linguistic logic. Furthermore, its operational mode closely matches the activities in the brain’s areas responsible for concept processing.

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