Anthropic Research Paper Examines Parallels Between AI and Human Consciousness Models
A paper published by artificial intelligence developer Anthropic has discussed potential parallels between computational networks in large language models and cognitive frameworks in human neuroscience.
The research paper, released in July 2026 and titled “Global Workspace in Language Models,” examines the internal processing steps of the chatbot Claude. According to the publication, researchers observed a computational network—termed J-space—that handles intermediate processing steps prior to generating responses.
The authors suggest that this internal network shares functional similarities with the global workspace described in the Global Neuronal Workspace Theory of neuroscience, which relates to initial stages of information processing often associated with access consciousness. The paper also notes instances of random word generation during computation, though the authors acknowledged that the system does not replicate full human phenomenal consciousness or emotional experience.
The publication further draws historical comparisons between modern neuroscience models and traditional philosophical frameworks, such as cognitive processes outlined in Buddhist Abhidhamma texts.
This report was prepared from publicly available source material. Details may change as further information emerges.