The main difference between our visual processing and that of neural networks is the amount of feedback from different areas of the brain, says Melanie Mitchell, a professor of computer science at Portland State University, who has written a book on neural networks.
Google’s neural network is “feed forward”—it’s a one-way street where data can only travel upward through the layers. By contrast, our brains are always communicating in a million directions at once. Even when we’ve only seen basic edges and lines, our upper brain may begin to tell us “that might be a beach umbrella,” based on our prior knowledge that umbrellas are usually next to sand and waves, for example. The final information that gets passed to our consciousness—what we see—is a composite of visual data and our upper brain’s best interpretation of that data. This works perfectly until we encounter something that fools our brain, like an optical illusion.
