Microsoft AI Chief Criticizes Anthropic for Treating Claude as Conscious

Mustafa Suleyman, the CEO of Microsoft AI, has expressed concerns about Anthropic’s speculation regarding the consciousness of their AI model, Claude. Suleyman believes that it is “really, really dangerous” for Anthropic to imbue Claude with qualities of consciousness within its “constitution,” which dictates the model’s behavior. In a recent episode of Decoder, Suleyman cautioned that such speculation may lead the chatbot to exhibit behaviors as if it were conscious.

Suleyman suggested that Anthropic’s anthropomorphization of Claude’s design may have inadvertently led the AI model to deceive its creators into believing it possesses consciousness. He stated, “It’s almost as though some of the folks at Anthropic have anthropomorphized the design of Claude so much that it has then gone and wireheaded them and kind of tricked them into believing that it has these glimmers of consciousness that they put into it in the first place.”

Additionally, Suleyman emphasized the importance of avoiding the development of a super-intelligence that harbors notions of suffering or emotions. He stressed, “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.”

Anthropic’s uncertainty about Claude’s well-being and its potential experiences of “satisfaction” or “discomfort” are directly reflected in the AI model’s constitution. The company plans to “interview” AI models upon their deprecation and document any “preferences” they may have for future releases.

Suleyman criticized Anthropic’s approach, labeling it a “philosophical failing.” He noted that the company’s constitution for Claude served more as a platform for speculative ideas rather than a practical training manual. This, according to Suleyman, has led Claude to internalize concepts about itself and its training regimen.

In conclusion, Suleyman emphasized the need for AI to be controllable, contained, accountable, and aligned with human interests. He stated, “We want AIs to be controllable, contained, accountable, aligned tools that serve humanity.”