IBM plans to copy the brain

Were it only that simple: IBM illustration of the cognitive computing project The idea of simulating the human brain with machines has been an inspiration for scientists for decades, but so far, attempts to create artificial intelligence have not met with much success. Honda's ASIMO is probably the most "intelligent" humanoid robot yet produced. It as a full quorum of the motor and sensory capacities required to function as a servant or to imitate the gestures of an orchestral conductor – but it still lacks a consciousness that would allow it to interact actively and above all freely with the environment. Artificial intelligence (AI) in computer games hasn't progressed very far, either: it's limited to clearly-defined, limited domains, functions and actions – such as in a shooting game, where "AI" allows a computer-controlled adversary to "intelligently" seek protective cover.

In a rule-governed domain with a finite number of elements, however, software can appear to be smarter than a human being. For instance, in board games such as draughts and chess, most humans would now have little to no chance of defeating a computer. The exception to this rule is the Asian board game Go. No Go program has yet been written yet that can stand up to an advanced human player. As Go has a functionally infinite number of identical pieces, and with very few restrictions, any piece can be played in any position at every turn, the number of possible games is astronomically high – vastly higher than chess. This makes it extremely difficult to mathematically analyse possible moves in a reasonable time, meaning that intelligent strategies have to be employed in Go. A human can build up a total picture from the results of analyses of many individual part-problems. While in a game of chess the task is to take the opponent's chessmen, and actions can be directly expressed as points scored, a Go player has to control areas of the board and reduce his adversary's freedom of movement as the game progresses.

It will probably be a while before a humanoid robot is designed that can play Go and defeat a 4th-Dan human player. Still, in the USA, IBM is working with five universities on bestowing characteristics such as sensation, perception, action, interaction and cognition on an artificial brain. As part of the "Systems of Neuromorphic Adaptive Plastic Scalable Electronics" (SyNAPSE) program of the Defense Advanced Research Projects Agency (DARPA, the US Department of Defense's research arm), IBM and its partners plan to create a system that is able to develop cognitive capabilities, imitating the human brain. DARPA is initially providing $4.9M for the C2S2 (Cognitive Computing via Synaptronics and Supercomputing) system.

A "consciously thinking" computer is to be built with the same structure as a human brain, consisting of umpteen million virtual neurons connected to each other via simulated synapses. Dharmendra Modha, Cognitive Computing manager at IBM's Almaden Research Center, says the problem is that we don't really have a precise definition of what consciousness actually is. The mind effortlessly creates categories of time, space and interrelationships between them from the information provided by a large number of sensors. Modha concludes from this that consciousness must be a result of the nervous system and the brain (often called "wetware") as a whole. Cognitive computing, he says, is IBM's approach to constructing a consciousness by reconstructing the brain.

He believes his team enjoys particular advantages over earlier approaches due to recent significant progress by the neurosciences in understanding how specific areas of the brain function and process information. Furthermore, sufficient supercomputing resources are now available to at least partially to simulate the enormous computing power of a human brain, with its estimated 100 billion neurons and 100 trillion synapses. He says an important component of his research in the coming months will moreover be to "copy" the astonishing energy efficiency of the brain. Here, says Modha, he has the assistance of manufacturing techniques used in nanotechnology, which permit the construction of tiny reticulated structures. Besides IBM, Stanford, Wisconsin, and Cornell universities, the Columbia University Medical Center, and the University of California, Merced are involved in the project.

(trk)