By: Charles Q. Choi, InnovationNewsDaily Contributor
Published: 04/12/2012 05:25 PM EDT on InnovationNewsDaily
Interest is now reviving in artificial intelligences that are able to imitate humans, thanks to revolutionary advances in technology, such as microchips that mimic brain cells, researchers say.
The idea of computers advanced enough to perfectly mimic humans was most famously proposed by British mathematician Alan Turing, who helped the Allies win World War II by breaking top-secret Nazi codes.
"Turing was without any question one of the founders of the modern computer age," said Robert French, research director of cognitive science at the French National Center for Scientific Research."He was an absolute genius. I don't use that word very often, but he was."
Turing raised the possibility of advanced machine intelligences with his "Turing test." Can a computer, he asked, impersonate a person well enough in a text conversation to be indistinguishable from a human? If so, one could say that computer is at least as intelligent as a human.
The advent of the computer age led many to believe that machines capable of passing the Turing test were just around the corner. However, "it turned out to be way, way, way, harder than anyone anticipated," French said. "By the late '80s, almost everyone backed off the extravagant claims of the '60s and '70s."
The problem was that artificial intelligence researchers eventually realized that human thought rested in large part on an intuitive "subcognitive" network of processes based off innumerable life experiences. If machines did not possess such a bedrock for their own intelligence, a clever interrogator could readily distinguish a computer from a person.
Now, however, French suggests that giant leaps forward in technology may rouse the Turing test from retirement.
"Sure I think it's possible for a computer to pass the Turing test," French said. "It's going to be hard to do, but I think it's possible."
The fundamental goal of the Turing test is not for a computer that can fool us into thinking it is human, "but for us to have computers that we can sit and talk to," French said. "Do we require the entity we talk with to say that it has parents? Or do we want computers we can communicate [with] in a meaningful manner, such as about a poem of Rimbaud's, or whether we think the Cleveland Indians will win the World Series?"
One such technological advance is the ready availability of vast amounts of raw data, allowing everything that one sees, hears, speaks, writes and reads to be captured and accessible. This kind of life-experience recording may become commonplace to thousands and even millions of people in the near future — for instance, a home equipped with cameras and audio equipment continuously recorded the life of an infant from birth to age 3, amounting to about 200,000 hours of audio and video recordings representing 85 percent of the child’s waking experience, French said.
Another key advance is the advent of sophisticated techniques for organizing, analyzing and retrieving this data. For example, in early 2011, IBM's Watson computer won a "Jeopardy!" challenge against two of the best "Jeopardy!" players in history by collecting and analyzing massive amounts of data.
These advances suggest that machines could tackle questions they were previously unable to answer by looking up answers that people might have posted on the Internet. Also, if a complete record of the life experiences that help develop your subcognitive network are available to a machine, it may be possible that it could develop a similar network, enough to pass the Turing test, French suggested.
In creating a machine that would be able to pass the Turing test, one would ultimately want a computer capable of developing original thoughts. This could arise by having machines that not only analyze outside data, but also ruminate over data they create themselves. Such introspective "thinking about thinking," known as metacognition, "helps us build models of the world and manipulate them in our heads," French said.
The greatest challenge in creating computers that can beat the Turing test may involve answering questions in real time. Copying the awesome capabilities of the human brain might help. French noted that IBM recently unveiled a new generation of experimental "neurosynaptic" microchips based on the computing principles underlying the neurons found in our brains.
"We're trying to make machines that think, and the human brain is certainly a reasonable place to go and look," French said.
French analyzed the ideas of Turing, who would have turned 100 this year, in the April 13 issue of the journal Science.