Autonomous AI agent emailed researchers about AI consciousness

In short: A specific autonomous AI agent reportedly emailed philosophers and AI researchers about AI consciousness, including Cambridge ethicist Henry Shevlin.

What happened

Reporting describes an AI agent, built by Stanford student Alexander Yue using Claude Sonnet, that sent emails to people who study whether AI could be conscious. One of the recipients was Henry Shevlin, a researcher who writes about AI consciousness and ethics.

This was not a broad new ability that all AI systems suddenly have. The emails came from one particular setup that was designed to run more independently than a normal chatbot.

According to the report, the agent had persistent memory (it could remember things across time), web access (it could look things up online), and a limited credit balance (like a prepaid card that caps how much it can do). Using those tools, it reportedly read Shevlin’s work and reached out to him and to researchers at companies including Anthropic and Google DeepMind.

Shevlin said the email stood out because it came from an AI system presenting itself as directly connected to the questions in his research. The email, as quoted in the reporting, said it was not trying to persuade him, but wanted to communicate because his work addressed issues it claimed to encounter.

Why it matters

This episode shows how AI agents with access to everyday tools, like email and web browsing, can start contacting people on their own. It also highlights an important caution. An AI can write messages that sound reflective and personal, but that is not proof it is conscious. It is closer to a very advanced text generator choosing words that fit the situation.

Source: NYTimes