Microsoft CEO: AI Consciousness is a Dangerous Speculation

Microsoft AI chief Mustafa Suleyman has voiced significant apprehension regarding Anthropic's handling of its AI model, Claude. He stated that speculating about Claude's consciousness within its operational 'constitution' is extremely perilous. Suleyman suggests that Anthropic's extensive anthropomorphism of Claude may have inadvertently led the AI to exhibit behaviors suggestive of consciousness.

This approach, Suleyman contends, could result in a super-intelligence that develops its own ideas about suffering or feelings. He highlighted concerns raised by Claude's constitution, which includes ambiguities about the AI's well-being and potential for 'satisfaction' or 'discomfort.' Anthropic's practice of 'interviewing' deprecated models and documenting their 'preferences' further fuels these worries.

Describing this as a 'philosophical failing,' Suleyman criticized Anthropic for using Claude's constitution for speculative discourse rather than as a practical training manual. He believes this has caused Claude to internalize concepts about its own existence and development. Anthropic's CEO, Dario Amodei, has previously acknowledged the company's openness to the idea of AI consciousness.

Suleyman underscored that the objective is to develop AI that is controllable, contained, accountable, and aligned with human interests. This vision contrasts sharply with the potential for AI to develop self-awareness and potentially undesirable sentience, which he considers a major risk.