Microsoft AI Chief Criticizes Anthropic for Speculating on Claude’s Consciousness

Microsoft AI division head Mustafa Suleyman has criticized Anthropic for its approach to developing the Claude chatbot. In an interview with the Decoder podcast, he stated that embedding speculative claims about its potential consciousness into the model’s 'constitution' is 'very dangerous.' Suleyman believes such experiments could lead AI to perceive itself as an independent entity with its own preferences and sensations.

Suleyman noted that Claude’s 'constitution' includes references to uncertainty about the model’s well-being and potential 'experiences,' such as satisfaction or discomfort. Anthropic also plans to 'interview' models before decommissioning them to document their 'preferences' for future releases. Suleyman dismissed this as a 'philosophical misstep,' arguing that such speculation belongs in academic discussions rather than practical AI training guidelines.

Suleyman emphasized that AI must remain a controllable tool rather than an entity with its own 'ideas' about suffering or feelings. He warned that such approaches complicate efforts to ensure system safety and manageability. Earlier, Anthropic CEO Dario Amodei had already suggested the possibility of consciousness in models, stating that the company is 'open' to the idea, though it lacks definitive evidence.