The AI-box experiment

The AI-box experiment was conceived by AI researcher Eliezer Yudkowsky: It revolves around the hypothetical idea of a superintelligent artificial intelligence, which is locked in a piece of isolated computer hardware, kept constrained in such a way that it cannot communicate with the world except for a simple text chat, where it can talk to its programmer. The question is whether the AI could convince its guard to let it out of the box, by connecting it to the Internet, for example.

In 2002, CS major Nathan Russell from the University at Buffalo doubted any persuasion could convince him to let an AI out of the box in such a situation, so Yudkowsky offered to run a simulation, where he would take the role of the AI, they would talk for 2 hours and see what happens. For incentive, Yudkowsky offered Russell $10 if he would manage to keep him in the box.

Details of the conversation are not known to the public, but after the conclusion of the experiment, Russell posted a signed message to the mailing list where the discussion had begun:

I decided to let Eliezer out.

Nathan Russell

Yudkowsky later repeated the experiment with another person with the same result. Most people think Yudkowsky used social engineering strategies like: "If you tell people that you let me out, they will be worried about the future, and invest more time and money into AI safety research."

This concludes this year's edition of the Advent Calendar of Curiosities. Did you enjoy it? Know any curiosities of your own? Let me know! Thanks for reading, and have a great 2018! <3