Scientists who tricked monkeys by swapping images of sailboats for teacups have figured out how the brain learns to recognize objects, a finding that could lead to robots that “see.”
“One of the central questions of how the brain recognizes objects and faces is that you never essentially see the same image twice,” said James DiCarlo, an associate professor of neuroscience at Massachusetts Institute of Technology.
He said humans have no trouble recognizing a dog, regardless of whether it is running, lying down, wagging its tail or begging for food.
“The pattern of light in your eyes is never the same when you view your wife or your dog, yet you can still recognize that as the person or creature that you love,” said DiCarlo, whose research appears on Thursday in the journal Science.
Scientists think people do it by gathering a host of different snapshots of the same object over a short period of time.
“Even though we don’t see the same images twice, nearby images in time tend to be images of the same object,” DiCarlo said in a telephone interview.
To test this idea, DiCarlo set up an experiment on two monkeys in which the scientists tried to trick them into unlearning their assumptions about an object.
The researchers attached electrodes to the parts of the brain responsible for recognizing objects. They specifically were testing changes in neurons that recognize images of a sailboat.
The monkeys were given treats if they looked at a video screen that contained several different pictures of a sailboat. Occasionally, when they looked away, the researchers switched one of the sailboat pictures for an image of a teacup, but only in one spot.
Eventually, some of the neurons in the monkeys’ brains responsible for sailboat images responded to a teacup instead.
DiCarlo said the study in monkeys follows a similar experiment in humans and suggests this is likely how people learn to categorize and recognize objects they see.
He said the finding opens a window into the visual learning system and will help researchers as they attempt to build computers with vision-like systems.
“There’s a lot of tasks that are essentially mindless for humans, but that only humans can do,” DiCarlo said, such as inspecting things on assembly lines, searching for explosive devices and looking at radiology images.