中国科研团队研究发现：人工智能可以自发形成人类级认知

Core Insights - The research confirms that multimodal large language models (MLLMs) can spontaneously form object concept representation systems similar to humans, marking a significant advancement in artificial intelligence cognitive science [1][2] Group 1: Research Findings - The study was conducted by the NeuBCI research group at the Chinese Academy of Sciences, combining behavioral experiments and neuroimaging analysis [1] - The research results were published in "Nature Machine Intelligence" [1] - The study analyzed 4.7 million behavioral judgment data points to construct a "concept map" for AI large models [2] Group 2: Cognitive Understanding - Traditional AI research has focused on object recognition accuracy, but this study explores whether models truly "understand" the meaning of objects [2] - The research identified 66 "mental dimensions" from large model behavioral data, which were given semantic labels and found to be highly interpretable [2] - The study revealed that humans tend to combine visual features and semantic information in decision-making, while large models rely more on semantic labels and abstract concepts [2]