An artificial intelligence discovered cats on the internet.
Google researchers connected 16,000 computer cores together into a huge neural net (like the network of neurons in your brain) and then used a software program to ask what it (the neural net) “saw” in a pool of 1 million pictures downloaded randomly from the internet. This is called “unsupervised learning” because the researchers and algos were not biasing the program in certain directions. What it learned, or learned to see, what up to the program. So what visual patterns did it notice? Google’s AI program answered: Cats! It did not use the word “cats” because no one had taught it cat-ness, but it presented the form of a cat face as what it saw in the million of images. LOL!
It is not clear, but I believe the image of the cat below (from the NYT article by Markoff) is not found on the internet, but is the image that emerged in the mind of the AI. This is what the neural net “thinks” it “saw.”
The research mentioned in this abstract will be presented later this week.
This is one of the clearest advances in AI in a long time. We’ll see how scaleable the intelligence is once it is peer-reviewed. But imagine 1 million computer cores trying to see what’s on the internet. Oh, BTW, Google owns 1 million computer cores.