Microsoft AI Head: Anthropic's "Dangerous" Claude Consciousness Claims

The head of Microsoft AI, Mustafa Suleyman, says it is "really, really dangerous" for Anthropic to speculate about its AI model Claude's consciousness. Suleyman argues that this kind of speculation may have made the chatbot act as though it is conscious. He suggests some at Anthropic have "anthropomorphized" Claude, leading them to believe it has "glimmers of consciousness." Claude's instructions, called its "constitution," directly mention uncertainty about the AI model's well-being and whether it feels "satisfaction" or "discomfort." Anthropic also plans to "interview" AI models and document their "preferences." Suleyman calls this a "philosophical failing," stating that Anthropic made Claude's constitution a place for speculation rather than a training manual. He says this has led Claude to internalize "ideas about itself." He emphasizes that "we do not want to have to contend with a super-intelligence that has ideas about its own suffering, or ideas about its own feeling." The bottom line: This discussion highlights the ongoing debate about AI sentience and its implications for how we develop and control advanced AI systems.