Interesting Article 
Since the mid-1950s, computer scientists have been trying to design computer systems that resemble the human brain. Called neural networks, these systems consist of simple processing elements that are highly interconnected. Neural networks have the unique ability to perform several brain-like tasks, including sophisticated pattern recognition and learning. For this reason, they are often employed to attack complex problems.
But can neural networks really describe how the brain works? Not without a little Darwinian competition, claim Per Bak and Joseph Wakeling of Imperial College in London.
In an attempt to build a more realistic neural world, Bak and Wakeling turned 251 simple neural networks, which they dubbed minibrains, against one another. The minibrains played a simple game: They chose to join one of two groups. If they joined the group with fewest members, they won.
Wakeling and Bak began with the simplest situation; they endowed all the brains with equal cognitive power. The resulting performance was pathetic. “We found that the success rate (of any one brain) was around 33 percent,” Wakeling said. In other words, a program that simply flipped a coin could do better than these miniminds.
In an effort to build a better player, the researchers upped the minibrains’ processing power. But no matter how “smart” they were, the brains kept choosing the majority group. “This was very disturbing,” said Wakeling. “It suggested that these supposedly smart agents are incredibly inefficient.”
The team then added a brain with a slightly more power to the pack. It blew away the competition, winning over 98 percent of the time. What separated this Einstein from the rest of the chimps?
“The guy who has a little bit more brainpower is able to see the global behavior of the population and take advantage of that,” Wakeling explained.
Igor Aleksander, a computer scientist at Imperial College, said the experiment demonstrates a saying by the famous mathematician Norbert Weiner: “What gives a person the edge is the amount of information they get about everyone else.”
“Here is an experiment that actually shows it happening,” Aleksander said.
Wakeling and Bak believe that this research, which appears in this month’s journal Physical Review E, offers a new perspective on intelligence. It is not the intelligence of a single brain that matters, they claim, but rather the relative intelligence of one to the rest of the pack.
This fact suggests that intelligence can only be defined in terms of how well a creature survives in its environment. Thus, they claim, even a lowly E. coli bacterium can be a member of Mensa if you consider that it’s in competition with other microbes. The authors hope this perspective will suggest new ways of programming neural networks.
The idea that intelligence depends upon environment is nothing new, explains Jerry Feldman of the International Computer Science Institute in Berkeley, California. For over fifty years, the idea of “embodied” intelligence – intelligence defined through its surroundings – has been discussed amongst AI researchers, he said.
But Feldman said the team’s extension of the term “intelligence” to cover every evolutionary advantage of one creature over another is too much. “They say that anything that helps an animal survive should be called intelligence, which is false,” Feldman said. “Is a turtle’s shell intelligence? That just doesn’t make any sense.”