Should we build computer prisons to trap dangerous AIs?

What do you do when you've got a renegade artificial intelligence on your hands? What should we do with Skynets and Hal 9000s of the world when they decide to turn their deadly power on human beings? One computer scientist wants to start building virtual prisons to house our most dangerous AIs.

Roman Yampolskiy, a computer scientist at the University of Louisville in Kentucky, wants to build these prisons to contain powerful artificial intelligences before they become self-aware. He fears a Skynet-like scenario where an intelligence is allowed to roam free through computer systems, triggering some sort of technological apocalypse. Yampolskiy proposes developing a new field of cybersecurity that could hold even the most clever intelligences.

Yampolskiy warns that such a security system would have to be technological in nature, so that human guards wouldn't be able to free the AI. He suspects that a sophisticate intelligence that really wanted to break free, it could blackmail or befriend a human guard into letting it go. Of course, he acknowledges that the real hazard is that we can't fully test how such a system would perform against a dangerous and ingenious AI until we got our hands on one — at which point we'd be closing the gate after the deadly cyber horse had already escaped the barn.

Humanity Must 'Jail' Dangerous AI to Avoid Doom, Expert Says [LiveScience via Presurfer]