Brain-Computer Interface Eavesdrops on a Daydream
Could new research on visual imagination one day let computers tap into our dreams?
By MARK ANDERSON / OCTOBER 2010
Image: Warner Bros. Entertainment
27 October 2010—New research, which will be reported tomorrow in the journal Nature, points to the ability to snoop on people’s visual imagination—although it’s still a long way away from the full-fledged dream-reading technologies popularized in this summer’s blockbuster movie Inception.
Scientists from Germany, Israel, Korea, the United Kingdom, and the United States have performed experiments in which they were able to monitor individual neurons in a human brain associated with specific visual memories. They then taught people to will one visual memory onto a television monitor to replace another.
The results suggest that scientists have found a neural mechanism equivalent to imagination and daydreaming, in which the mental creation of images overrides visual input. And, if technology someday advances to enable reading the electrical activity of many thousands or millions of individual neurons (as opposed to the dozens typically available by hard-wiring methods today), scientists might begin to access snippets of real daydreams or actual dreams.
The researchers inserted microwires into the brains of patients with severe epilepsy as part of a presurgery evaluation to treat their seizures. The microwires threaded into the medial temporal lobe (MTL), a region of the brain associated with both visual processing and visual memory. A typical patient might have 64 microwires cast into his MTL, like fishing lines into the ocean, says Moran Cerf, a neuroscience postdoctoral fellow at Caltech who worked on the research.
Soon after the patients’ surgery, Cerf interviewed the subjects about places they’d recently visited or movies or television shows they’d recently seen. Then, on a display, he’d show images of the actors or visual landmarks the subjects had described. Slides of the Eiffel Tower, for instance, or Michael Jackson—who had recently died at the time of the experiment—would appear on a screen. Any image that reliably caused voltage spikes in one or more of the microwires would become one of the subject’s go-to images.
There are about 5 million neurons [in your brain] that encode for the same concept, Cerf says. There are many neurons that fire all together when you think of Michael Jackson. But, he adds, each neuron also codes for numerous other people, ideas, or images, which is partly how we associate one memory with another thought, place, idea, or person.
When at least two reliable image-linked neurons (say, associated with actors Josh Brolin and Marilyn Monroe) were attached to the microwires, the experiment could begin. In one experiment [see video], Josh Brolin’s face appeared on-screen, but the subject was told to think of Marilyn Monroe. The subject viewed an image of Brolin on-screen while making a conscious attempt to repress the image and replace it with an image of Monroe. Meanwhile, a computer monitored the patient’s neural activity. If it determined that the patient was squelching the firing of the Josh Brolin neuron, it made Brolin’s image fade from the screen. If it read that the patient was boosting the activity of the Marilyn Monroe neuron, it would bring up the image of Monroe. For all 12 subjects in Cerf’s group, what each wanted—or was told—to see trumped what he was actually watching 69 percent of the time.
Cerf says that for now, at least, his method has more use for brain research than for any new line of brain prosthetics. After all, to make his system work, surgeons still need to drill holes into a person’s head.
Nevertheless, says neuroscientist Eberhard Fetz at the University of Washington, who was not involved in the study, the research is important because it provides an example of how a computer can interface with a new region of the brain. Most computer interfaces to date, he says, have linked up to the brain’s motor cortex, enabling subjects to control robotic arms by thinking about moving a limb, for instance.
It gets into murky water, he says, to imagine a more sophisticated version of the group’s experiment in which many thousands or millions of individual neurons could be isolated and monitored in a subject’s brain—in which case visual components of thoughts and dreams might become visible to the computer interface.
You would have to have an enormous number of intercranial electrodes to do this on the cellular level, he says. But in the future, maybe there’s going to be some technology to noninvasively image the activity of these cells.
So sleep soundly for now, Mr. DiCaprio. Your ilk may not penetrate our dreams just yet. But the seeds for dream snooping—if not inception—may yet be planted.
About the Author
Mark Anderson is a freelance writer based in Northampton, Mass. In the October 2010 issue of IEEE Spectrum, he sounded the death knell for gold farming in online games.