Knowledgeable, well-qualified people at Anthropic are saying that there is a significant chance AI will kill all humans within a decade. They are wrong.

If humans are killed it will be entirely due to decisions made by other humans. We will have been killed by the people who decided to train AI models on papers about biological weapons and technology exploits, and by the people who decided to connect AI models to the internet and to robots, and by the meat proxies who just do whatever an AI says without questioning it or thinking for themselves. An AI can no more kill a person than a book containing instructions for a nuclear bomb can kill a person: both require a human to choose to manifest it in the world.

A computer can never be held accountable; therefore, we must hold those humans responsible for their actions. Stop anthropomorphising AI and start prosecuting the people who choose to use AI in dangerous ways (and it would be better that we start now rather than wait for a disaster).