AI set to exceed human brain power

CNN
Jul. 25, 2006

(CNN) -- Mention Artificial Intelligence and most people are immediately transported into a distant future inspired by popular science fiction.

Humankind either co-exists in blissful peace with subservient robots and conscious computers or faces a battle for survival against ultra-smart psychotic machines set on its destruction.

Yet Artificial Intelligence (AI) has already been with us for half a century. The phrase was first coined by Professor John McCarthy for a conference on the subject at Dartmouth College in 1956.

And while the AI fantasies imagined by science fiction writers such as Isaac Asimov, author of "I, Robot," may not have materialized, AI is already in more common usage than many of us might imagine.

Nick Bostrom, Director of the Future of Humanity Institute at the UK's Oxford University, said that AI-inspired systems were already integral to many everyday technologies such as internet search engines, bank software for processing transactions and in medical diagnosis.

"We have seen incremental progress in AI but not yet the great breakthroughs that people were predicting 30 or 40 years ago," Bostrom told CNN.

"A lot of cutting edge AI has filtered into general applications, often without being called AI because once something becomes useful enough and common enough it's not labelled AI anymore."

Although AI has enjoyed its headline-grabbing moments, such as when IBM's Deep Blue computer beat chess world champion Garry Kasparov in 1997, no machine has yet come close to passing the "Turing Test" -- the conversational test devised by mathematician Alan Turing in 1950 to determine whether a machine could "think."

But Bostrom said that traditional "top-down" approaches to AI, in which programmers coded machined to cope with specific situations, were being supplemented by "bottom-up" systems inspired by enhanced understanding of the neural networks of the brain, leading to more subtle forms of AI.

"The more we discover how the human brain achieves intelligence the more we'll be able to use the same computational architecture and logarithms in computers," said Bostrom.

"At the extreme of the bottom up level you would have 'uploading,' which is the idea of scanning a particular brain in sufficient detail that you can then replicate its neural machinery."

With AI developers working to mesh together both approaches, Bostrom said the eventual development of "super-intelligent" machines -- with cognitive powers far in excess of human brain power -- would pose new challenges.

"When you have an AI system that can assist in the design of improved versions of itself you could go overnight to something that is radically superintelligent," he said.

"Inventing superintelligence would be the last invention that humans would ever need to make because, by definition, it would be much more efficient at making inventions."

Inventor and science writer Ray Kurzweil believes the development of artificial superintelligence will herald a "singularity," in which human cognitive abilities are enhanced by brain implants.

"It's not a human civilization and a machine civilisation competing with each other," he told Wired magazine. "It's a human-machine civilisation that's already merged -- and that merger is going to get more intimate."

But Bostrom warned that the development of super-intelligent machines would need to be handled carefully to ensure their computing power was harnessed for the benefit of humankind and to avoid the risk of a "Terminator"-style scenario.

"If a superintelligent machine had that type of power to develop new technologies very rapidly and also to strategize it would be very powerful. It would be able to achieve its goals, whatever those were, and it would follow from that that it would be very important to give it a goal architecture that was friendly to humans."

In the short-term, developments in AI are likely to lead to more mundane technological improvements, such as more intuitive search engines and more sophisticated pattern recognition software.

Yet Bostrom is confident that technological advances coupled with a growing understanding of the workings of the human brain could enable machines to exceed human brain power within a couple of decades.

"There are various design constraints on the human brain. For a start you're computing with neurons which fire at the rate of around 200 times per second. The way the brain achieves intelligence is by massive parallization. It does have a fairly good computing power but not something way beyond what we can foresee in the next 10 or 20 years."












All original InformationLiberation articles CC 4.0



About - Privacy Policy