Matrix Learning10 Mar 2013, Posted by Article Sampler in
Discover, March 2013
In a well-known scene from The Matrix, Neo (played by Keanu Reeves) lies down in a high-tech dentist’s chair, straps on a wild array of electrodes, and begins downloading a series of martial arts training programs into his brain. Apparently—if the mechanics can be parsed—the information is transferred via the visual cortex. Afterward, he blinks his eyes open and speaks the words geeks have been quoting ever since: “I know kung fu.”
Automatic learning, the technical term for this idea, has been a longtime dream of the cyberpunk set. Most people thought it would remain in this aspirational realm for a while longer, but thanks to recent research by Brown University neuroscientist Takeo Watanabe, what has long been science fiction may soon become science fact.
To understand Watanabe’s breakthrough, it helps to know a little bit about the visual system’s plasticity—its ability to change. Twenty years ago, neuroscientists held that after a certain critical period, usually no more than the first 12 months of life, the entire visual system has become far too rigid for any real learning to take place. In other words, it has lost its plasticity.
This view of the brain started to change about 15 years ago, when Israeli neurobiologist Dov Sagi discovered that with intensive training in specific visual tasks, such as target orientation (the ability to look at a dot on the wall, look away, then look back at the dot’s exact spot), people much older than 12 months could improve their performance in those tasks. Sagi’s study of this “perceptual learning” in 1994 upended the concept of the rigid vision system.
Subjects in Sagi’s research still had to consciously train with visual cues before they saw any improvement. The learning did not manifest suddenly, as it did for Neo. But in 2011, Watanabe designed an experiment to see if something like automatic learning might be possible. He wondered if he could train the vision system without a subject’s knowledge, and without the use of a stimulus like a dot.
Watanabe’s experiment had two parts. First, subjects had their brains scanned by a functional magnetic resonance imaging (fMRI) machine as they stared at a computer screen. On the screen was a simple image: a series of diagonal lines. Merely studying those lines produced a very specific activation pattern in the visual cortex, which the fMRI decoded and stored.
The second part of the study took place the following day. Again, subjects looked at a computer screen while being scanned by an fMRI machine, but this time instead of the lines, a small disk appeared on the screen. Whereas the previous day’s work just required the subject to stare at the image, the goal today was to mentally make the disk bigger. But there was a catch: No one told the subjects how to enlarge the disk.
The solution was far from obvious. The only way to increase the disk size was for subjects to get their brain to reproduce the pattern that it produced when they stared at the previous day’s diagonal lines. The equipment had been set up to recognize that initial pattern and make the disk on the screen bigger as soon as the pattern was replayed. “The more similar the brain activation pattern was,” Watanabe says, “the bigger the disk became.”
In fact, the task might seem impossible to the uninitiated, but it was not. In attempting to solve an apparently unsolvable problem (how does one mentally make a circle grow?), the brain automatically replays recently learned perception patterns, which in the subjects’ case included the pattern produced by those diagonal lines. When they hit upon the pattern, the disk started to expand—automatically, no training required.
From Dots to Data
But here is where things get really interesting: That first activation pattern—the baseline sequence produced by staring at those diagonal lines—was just meaningless information. Hypothetically, that doesn’t have to be the case. In theory, if the target acquisition sequence produced by staring at those first lines had actually contained meaningful information (like a series of kung fu training programs, for example), then the subject would automatically repeat that pattern, essentially practicing it every time the brain tried to enlarge the disk. Genuine skill acquisition would take place.
It is still a long way from that kind of advance to automatic learning. Matrix-style knowledge downloads will require much more than just recording and replaying visual cortex activation patterns. Nobody knows if this kind of phenomenon also arises in places like the motor cortex or auditory cortex, which would be useful in mastery of physical or linguistic skills.
But Watanabe thinks this method can be used to cure depression. “I think we could easily train people to be happy,” he says. “Just show subjects pictures of babies and kittens and other images known to elevate mood, and record and use this pattern as the trigger for disk enlargement. Then, when subjects perform this task, they’ll be making themselves happy as well.”
Equally fantastic, though much less Matrix than Manchurian Candidate, is another possibility: “I think we could use the technique to erase memories, like removing 12 months from a person’s life,” Watanabe says. “If a memory is associated with some cue, we could make a subject induce a pattern that has nothing to do with the memory while the cue is being presented. In that way, when the cue is given, the subject would remember the implanted memory rather then the real memory.” It might come in handy for people with post-traumatic stress disorder, though it is easy to see how the power to erase a person’s memory could also be abused.
As Reeves’s Neo might say, “Whoa.”