Anthropic announced Monday that it has discovered a small internal workspace within its Claude AI model used to store and process thoughts before expressing them in words. The company notes that this structure bears striking similarities to how humans consciously access thoughts.
While Anthropic does not claim that Claude feels or experiences anything, it found a surprisingly human-like division of information processing: between information used for deliberate reasoning and the much larger volume of automatic computations occurring below it. This adds new fuel to the debate surrounding machine consciousness.
The company explains in a video that Claude uses a special area to plan strategies that may not be related to its immediate task and are separate from the "chain of thought" it shares with users. "We can see Claude silently perform logical steps in its head – detecting errors in code, identifying images, and more," Anthropic stated in an X post accompanying its video.
Anthropic has named this space "J-Space," after the Jacobian, a mathematical technique used to detect what is happening. "Similar to how humans can think about one thing while doing another, Claude can activate concepts and computations in its J-space that are unrelated to its outputs," Anthropic added.
