Will computers ever be able to think independently of humans?
It's a question that's become more science fact than science fiction in recent years, and according to the results of a recently announced Google experiment, the answer to it could be yes -- at least when it comes to cats and the internet.
Andrew Ng, director of Stanford's Artificial Intelligence Lab, worked with Google X-Lab engineers for several years to create one of the largest artificial neural networks in the world.
The network is comprised of 16,000 computer processors with more than one billion connections -- a "Google Brain," as the New York Times is calling it.
Using state-of-the art machine learning techniques, this artificial brain was able to teach itself what a cat looks like using 10 million images extracted from YouTube videos. Researchers were surprised by the result of the simulation.
We never told it during the training, 'This is a cat,'" said Google fellow Jeff Dean in the New York Times article. "It basically invented the concept of a cat. We probably have other ones that are side views of cats."
Because the videos were selected completely randomly, researchers note that the results of this simulation also make for an interesting comment on what interests humans in the internet age.
Scientists at Google's ominously-named X Laboratory, the same lab credited with innovating technologies for augmented reality glasses and self-driving cars, has been experimenting with artificial intelligence and neural networks for years, said Ng.
"This research is representative of a new generation of computer science that is exploiting the falling cost of computing and the availability of huge clusters of computers in giant data centers," writes John Markoff. "It is leading to significant advances in areas as diverse as machine vision and perception, speech recognition and language translation."
While programmers note that the recognition of cats on YouTube by a computer isn't earth-shattering, they're making significant progress in the field of artificial intelligence.
This simulation performed far better than any previous effort, roughly doubling its accuracy in recognizing difficult objects.
