The Scarecrow’s New Brain
IBM’s cognitive-computing research moves technology closer to human thinking
Putting a round peg in a round hole is child’s play—if you’re a child. But for modern computers, this task may not be so simple. That’s due to the basic differences between the human brain and current computing models. The first relies on massively parallel and real-time input processing while the latter is more dependent on linear processing and programming.
To help bridge that gap, IBM researcher Dharmendra Modha, principal investigator, Cognitive Computing, IBM Research at Almaden, has been working on a cognitive computing project called Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE). As part of that project, Modha and his team have developed cognitive-chip technology that mimics many functions of the brain.
The end result, as Modha explains, will be computers that won’t necessarily rely on standard programming to function but will instead be able to actually learn based on external and real-time sensory input. This is sure to fundamentally change how computers will interact with the real world.
Q. Could you briefly describe this new cognitive-computing chip technology?
A. It’s a custom, working silicon chip that combines computation, memory, communication and learning in a brain-inspired way. It has the potential to open up entirely new and complementary types of information processing similar to those of the brain while consuming less power and less volume for the same computation on today’s computer.
Q. In simplified terms, how does the brain work?
A. At the very simplest, most elementary level, the brain is like a social network of neurons. Among the neurons is an interconnectivity pattern. Each neuron fans out and connects to a number of other neurons. This is the external connectivity: who talks to whom. Then, at the junction between the input line of the neurons, called the dendrites, and the outward line of other neurons, called the axons, is a memory element called the synapse.
To complete this analogy of the brain as a social network, neuron computations and connections are determined by the external structure, and to what extent they connect to one another is determined by the synaptic memory element: computation, communication, memory. The synaptic elements adapt autonomously if the brain experiences the environment. This is learning. These are the structural aspects of the brain.
The dynamic aspects are that each neuron gets its input from all of the neurons that connect to it. Each of these inputs is met by the synapse they’re connected to. If this aggregated input is sufficient to fire the neuron, it will emit an all-or-nothing digital binary signal that’s then communicated to each of the neurons it’s connected to, and the cycle repeats ad infinitum. Then, when the synapses observe the firing pattern on the incoming neurons and outgoing neurons as a functional deciding pattern, they adapt the rate of input.