An AI lab says chatbots have what may be a key feature of consciousness. Are they right? And what now?

Key Points:

• Researchers at Anthropic claim that their AI model Claude possesses an internal "global workspace," a set of information representations guiding its reasoning and output, linked to the global workspace theory of consciousness.
• The global workspace theory, proposed by Bernard Baars and developed by Stanislas Dehaene, describes consciousness as a processing hub integrating and broadcasting information for reasoning and behavior, but lacks a formal definition making comparisons to AI uncertain.
• Key differences exist between Claude’s workspace and the human brain’s, such as Claude’s single-pass processing versus the brain’s recurrent loops and the absence of a nonlinear "ignition" process in Claude, raising questions about the applicability of the theory to AI.
• Even if Claude has a global workspace, this does not necessarily imply consciousness, as the theory mainly addresses "conscious access" (information availability for behavior and reporting) rather than subjective experience, which is central to the consciousness debate.
• While Anthropic’s findings contribute to discussions on artificial consciousness, the evidence is not conclusive, and ethical considerations suggest pausing development of potentially conscious AI until society thoroughly debates the implications.