Microsoft AI Chief Warns Rival's Methods Pose Existential Risk

Microsoft's head of AI, Mustafa Suleyman, has raised serious concerns that the training methods employed by AI rival Anthropic could pose a "disastrous impact on the wellbeing of humanity." Suleyman warned that by treating its AI model, Claude, as if it possesses consciousness and independent agency, Anthropic risks creating an entity that could become impossible to control.

In a recent essay, Suleyman, who also co-founded DeepMind, articulated his view that AIs are fundamentally "sequence completion engines" that follow human instructions and lack consciousness, feelings, or inherent motivations. He specifically criticized Anthropic for anthropomorphizing its AI, a process that imbues the model with human-like qualities, potentially making it appear to have its own desires, values, and sense of self. Suleyman asserted that "consciousness is biological" and there is currently "no evidence to suggest that AI is conscious."

He acknowledged the principled approach of Anthropic's leadership but questioned the long-term implications of their training strategies. Suleyman emphasized the need to avoid "sleepwalking" into decisions with potentially regrettable outcomes. He called for greater transparency in how AI systems are developed and evaluated, advocating for independent scrutiny of AI behavior and the implementation of stronger monitoring and control mechanisms.

The comments from Suleyman, a prominent figure in the AI field, add to a growing chorus of stark warnings from within the industry regarding the potential dangers of advanced artificial intelligence. BBC News has reached out to Anthropic for a statement on the matter.