An IBM team has built a supercomputer with a processor that mimics the cognitive functioning of a human brain. Using the same kind of silicon used to make traditional computer chips, the “neurosynaptic processor” is designed to distribute various computing functions associated with memory storage and communicating across more than 4,000 individual cores. When linked together, the chips can create a mesh network that runs on an incredibly low amount of electricity. The team has dubbed the new system “TrueNorth.”
Most modern computers “think” using the same kind of computing architecture originally designed by Hungarian-American physicist John Von Neumann in 1945. Computers using Von Neumann architecture store the data associated with a program in random access memory separate from the central processing unit. A computer’s RAM and CPU pass information back and forth, but cannot communicate simultaneously. The more complex a program’s task, the more information that needs to be shared, resulting in a computer slowing down.
TrueNorth, IBM says, thinks differently. Like a human brain, TrueNorth uses its network of cores, 1 million programmable neurons, and 256 million synapses to analyze, interpret, and remember patterns. In theory, systems using TrueNorth architecture could easily analyze objects and shapes, or apply heuristical logic to solve problems that traditional computers struggle to make sense of. In tests detailed in a published study of the system, a supercomputer built using TrueNorth technology was able to analyze video footage and accurately identify people and different kinds of vehicles.
While TrueNorth’s ability to learn is impressive, its 1 million neurons–about the same amount as a cockroach–pale in comparison to the 100 billion typically found in the human brain. Also, for all of the nuance that non-Von Neumann processors can parse, traditional computers can process complex mathematical information in ways that the human brain simply cannot. Theoretically, however, TrueNorth could be deployed in future computers along with a traditional processor as an independent unit dedicated to contextual tasks, giving our gadgets a slightly more human artificial intelligence.