MIT Study Finds That A Computer System Can Identify Objects As Well As Primate Brains

A computer vision simulation. Image provided to bostonmagazine.com

A team of MIT researchers found that an existing computer vision system can achieve object recognition as well as humans and other primates, opening the door to possible brain research and vision aid improvements.

The study was led by Professor James DiCarlo, head of MIT’s Department of Brain and Cognitive Sciences. DiCarlo and his team compared primates’ visual recognition abilities—in other words, the capability to see something or someone and register that it is there and what it is—with one of the most advanced computer vision systems ever developed, Super Vision, and found that the two work at the same level.

“This is one of the things that humans and even non-human animals do remarkably well relative to machines, and that speaks to the power of the brain to process information,” DiCarlo says of visual recognition. “We had some ability, some computational skills in our heads, that we as engineers had not yet been able to emulate.”

This study suggests, however, that the engineers behind Super Vision achieved just that—which DiCarlo says is exciting on a number of levels, first of which that it could help neurologists better understand the complex inner workings of the human brain. “[Visual recognition] is thought to be accomplished by processes in cortical networks that are probably not only restricted to vision,” he explains. “These same kind of algorithms are probably at work in other sensory systems, so if we think we could understand what’s going on in one of these systems, it will help us to understand all of them.”

What that means, he says, is that at some point in the future—and DiCarlo cautions that it’s not immediately foreseeable—scientists may be able to take what they know about the way the brain’s neurons work and use it to fix ailments such as autism or dyslexia. Another possible application is helping the blind or visually impaired population. “For instance, if you’re blind you can’t tell me there’s a face out there, but a computer vision system now is going to be much better able to do that,” DiCarlo says. “Essentially, better computer vision will help blind people because it now will act as your vision system.”

Putting the research into practice may be a ways off, but DiCarlo says that the study is an important step in merging the worlds of technology and neuroscience, since the computer vision system examined seems to be not only as capable as primate brains, but similar in structure and function.

“That’s what’s most exciting from a science point of view, that we’ve not just matched performance but the models are now offering us hypotheses of what could be going on,” DiCarlo says. “We have machines that play chess as well as us or better than us, but we think those machines work differently than us. This machine is built in a way that looks a lot like our own brain.”