Self-Learning Processors To Pave Way For Crash-Proof Biological Computing
In the quest to build a processor that's an order of magnitude more efficient than what we have available today, the brain is the first thing looked at. Scientists would love nothing more than to fully understand how the brain works, because with that knowledge, not only could things improve on the healthcare side, but computers would benefit as well.
Thus, researchers have long been craving for a processor that even partially but realistically behaves just like a brain. It would be able to learn new things as time passes, and enjoy the benefits of unparalleled error-correction. In 2014, a much simpler version of the ideal chip will be made available, which could be the start of great things to come.
The benefit of a self-learning chip is that programmers wouldn't need to write hundreds or even hundreds-of-thousands of lines of code to get a processor to learn and execute certain tasks. This processor could be connected to a robot or something else mechanical.
Credit: Erin Lubin/The New York Times
The possibilities that a self-learning processor bring about can both be amazing and scary. The basic premise can bring forth a surge of science-fiction movie memories, but the way things are going, it could just become "science" in the near-future.
A self-learning processor could power a robot, for example, that acts as company to an elderly person. This robot would interact with the person and learn about them, and be able to hold a normal conversation. This robot could learn and inherit various emotions, just like we all did as we grew up. Another scenario could be assembly lines. Envision robots there that can quickly learn from their mistakes, or learn how to correct an issue that arises.
That all sounds genuinely crazy, but it's scenarios like these that some think about being possible way down the road.
It might take a hundred years or more to produce realistic processors fast enough to mimic even a little part of the brain. Last year, IBM made headlines when it used a supercomputer to mimic 10% of a human brain. That sounds impressive, but it becomes less so when you realize the computations were being done at 1/1500th the speed of a real brain. Oh - there's also the little fact that this supercomputer consumed power in the megawatts, versus a human brain which is said to be about 20W.
It's clear that we have quite a ways to go if we want truly capable self-learning processors, but, we've got to start somewhere.