Every year, computers surpass human abilities in new ways. In 1997, IBM’s “Deep Blue” computer beat the World Champion of chess, and earlier this year IBM’s “Watson” computer beat the best human players at Jeopardy!. Recently, a robot named “Adam” was programmed with our scientific knowledge about yeast, then posed its own hypotheses, tested them, assessed the results, and made original scientific discoveries. This year, a machine surpassed human skill in recognizing traffic signs — an advance with implications for the debate over the safety of self-driving cars.

What happens when our machines surpass human ability in all things? The Pixar film Wall-E imagines an outer-space city staffed by robots who maintain the city, keep things clean, and hand over fresh food to increasingly obese humans relaxing in flying armchairs.

But the true potential of benevolent smarter-than-human machines — so called “Friendly AI” — is much greater than this. Friendly AI could mean the end of cancer. The end of starvation. The end of war. Even the end of death.

But building Friendly AI will not be easy. Isaac Asimov used science fiction stories — some of them combined for the Will Smith movie I, Robot — to show why a simple set of rules like “Don’t harm humans” won’t be sufficient to make AI friendly.

Eliezer Yudkowsky has been studying the challenges of Friendly AI for more than a decade. His talk for the 2011 Singularity Summit, “Open Problems in Friendly Artificial Intelligence,” will show how the research program of Friendly AI breaks down into technical sub-problems that can be addressed by today’s researchers.

This entry was posted on Wednesday, September 28th, 2011 at 5:51 pm by Luke Muehlhauser and is filed under speakers. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.