Tonight, the Perpetual Notion Machine looks into Artificial Intelligence and the problem of control, which, according to our guest, could be more real than fiction. Stuart Russell has written the book Human Compatible: Artificial Intelligence and the Problem of Control. Stuart is professor of Computer Science at the University of California, Berkeley. He is a fellow of various societies, including the American Association for Artificial Intelligence, and is co-author of the definitive college textbook on AI called Artificial Intelligence: A Modern Approach.
Stuart informs PNM’er Dennis Shaffer that past and current AI research isn’t trying to duplicate or simulate human intelligence. That would be way too complex to achieve since even humans don’t know everything about human intelligence. But Stuart does admit that super intelligent machines will happen sometime in the future. Even today, some advanced machines have the capacity for learning, which can sometimes lead to severe problems. Stuart warns that unless developers consider fail-safe options, super intelligent machines may be impossible to control. He suggests instilling machines with uncertainty about their true objectives so that they remain provably deferential to humans, forever.
For those of you interested in some more info, check out this article from Quanta magazine on whether Machines Beat Humans.
And here’s another one from the Future of Life Institute on AI’s benefits and risks.