Computer technology still lags far behind the abilities of the human brain, which has billions of neurons that help us simultaneously process a plethora of stimuli from our many senses. But Qualcomm hopes to shrink that bridge with a new type of computer architecture modeled after the brain, which would be able to learn new skills and react to inputs without needing a human to manually write any code. It's calling its new chips Qualcomm Zeroth Processors, categorized as Neural Processing Units (NPUs), and already has a suite of software tools that can teach computers good and bad behavior without explicit programming.
Learning without code
Qualcomm demoed its technology by creating a robot that learns to visit only white tiles on a gridded floor. The robot first explores the environment, then is given positive reinforcement while on a white tile, and proceeds to only seek out other white tiles. The robot learns to like the white tile due to a simple "good robot" command, rather than any unique algorithm or code.
The computer architecture is modeled after biological neurons, which respond to the environment through a series of electrical pulses. This allows the NPU to passively respond to stimuli, waiting for neural spikes to return relevant information for a more effective communication structure. According to MIT Technology Review, Qualcomm is hoping to have a software platform ready for researchers and startups by next year.
Qualcomm isn't the only company working on building a brain-like computer system. IBM has a project known as SyNAPSE that relates to objects and ideas, rather than the typical if-this-then-that computer processing model. This new architecture would someday allow a computer to efficiently recognize a friendly face in a crowd, something that takes significant computing power with today's current technology. Modeling new technology after the human brain may be the next big evolutionary step in creating more powerful computers.