Anthropic co-founder: "AI is a hammer that suddenly realized it's a hammer"

«Imagine working in a hammer factory, and suddenly one of them says, 'I’m a hammer, how interesting!' — that’s extremely unusual,» Clark explained, speaking metaphorically. In his view, when computational power, data volume, and model scale reach a certain threshold, properties that nobody specifically programmed begin to emerge.

«This technology is closer to something grown than created,» he explained. «You set the initial conditions, build the framework, and from it grows something so complex that designing it manually would be impossible.»

Clark warns that humanity is creating extremely powerful systems whose working mechanisms aren’t fully understood. «And to raise the stakes — you’re guaranteed to lose if you decide this being isn’t real. The only chance to win is to see it for what it actually is,» he said.

Clark’s perspective has sparked debate among researchers. Many specialists remain skeptical about the idea that large language models are capable of true self-awareness. Critics maintain that even the most advanced neural networks are merely complex pattern-matching systems rather than conscious entities.

Before founding Anthropic, Jack Clark held the position of Policy Director at OpenAI, and previously worked as a journalist for Bloomberg BusinessWeek and The Register.