Is there finally a computer that solves problems the way we do?

Bringing us one step closer to giant robot overlords is CLARION, a program that performs the same way human subjects do in some impressive cognitive tests. It mimics not what we think, but how we think.

When presented with a complex problem, no person knows the solution instantly. There is a problem solving process, that in the early twentieth century was officially broken down into named stages by researchers; preparation, incubation, insight, and verification. This was termed 'stage decomposition,' and as most theories do, it formed the basis for more theories. The stage termed 'incubation' was particularly hotly contested. What exactly happened during that period? Was it a time of rest for the mind? Was it a time to pare away bad ideas? Perhaps it was just a time to search one's memory for outside information that might help with the problem.

Ron Sun and Sèbastien Hèlie didn't want to limit themselves. They incorporated all the possible stages into the CLARION program. They then gave the model problems to solve. Specifically, they gave the model problems that, in the past, many humans had solved in closely controlled trials. The tests with the human subjects had measured how well humans solve problems under certain conditions. When humans were asked to discuss their thoughts about the problem prior to giving a solution, they answered correctly 35.6 percent of the time. When they were assigned another task, interrupting their work on the problem before going back to thinking about it, they answered correctly 45.8 percent of the time.

When CLARION solved problems under similar conditions, the results uncannily mirrored those of human subjects. Five thousand runs of the program returned a correct result 35.3 percent of the time when the scientists imposed conditions similar to a human 'discussing' their work, and 45.3 percent of a time when it was interrupted to work on another task.

The Turing Test is a famous, if controversial, measure of artificial intelligence. It holds that a machine is as 'intelligent' as a human if a human can have a conversation with a machine and not know they are talking to a machine. This may be a psychological Turing Test, in which both human and machine subjects turn in results so similar that scientist can't tell, by the numbers, if they are testing humans or computers.

If this is the case, have Sun and Hèlie succeeded in recreating at least one aspect of the human brain?

Via RPI.