Chinese scientists confirm AI capable of spontaneously forming human-level cognition

Chinese researchers at the Chinese Academy of Sciences teamed up with neuroscientists to show that today’s multimodal large language models (LLMs) don’t just label images—they actually form object‐concept representations almost identical to ours. By crunching 4.7 million triplet judgments on 1,854 real‐world items, they distilled 66 interpretable dimensions that mirror human semantic clusters and outperformed people in consistency tests.

The study, published June 9 in Nature Machine Intelligence, found that while humans mix visual cues and semantics, LLMs lean on abstract labels—yet both arrive at the same “mental destinations.” In short, these AI models aren’t just stochastic parrots but seem to develop genuine, human‐level understanding of objects.