A new discovery by Massachusetts Institute of Technology researchers suggests that the brain may be able to work much like the instant replay on sports shows. But here is the surprising part - it runs backward and super-fast: instant replay on rewind.
Scientists at MIT's Picower Institute for Learning and Memory have managed to record such reverse replay signals from rats that ran laps on a small track, according to an online report in the journal Nature last week.
And that reverse replay appears to occur only during pauses for reflection, helping the rodents learn from what they have just experienced, the researchers say - as if it were the rapid review at the end of a college course.
"What this suggests is that, while there certainly is some record of your experience as it's occurring, that the actual learning - when you try to figure out what was important, what should I keep and throw away - that could happen after the fact, during periods of quiet, wakeful introspection," said Matthew A. Wilson, the paper's senior author. It is not clear whether the replay is conscious or unconscious, he said.
As the four rats ran, dozens of electrodes in their brains picked up patterns of electrical activity corresponding to different points on the track as the rodents moved from start to middle to end and a food reward.
Then, after the rats ate, they tended to rest a bit, and perhaps groom themselves. During that rest period, their brain patterns ran rapid-fire "end-middle-start, end-middle-start, end-middle-start" patterns.
Wilson's lab had already found that when rats sleep, their brains often repeat - in forward chronological order - the patterns corresponding to the routes they have recently traveled, as if cementing the geography they had learned.
But if replay was an important learning mechanism, it never made sense that it would have to wait hours until an animal slept, said David Foster, first author on the Nature paper.
"Now, we actually see it happening immediately," he said. "It's possible that replay could be more fundamentally involved" in learning.
In fact, reverse replay could help answer a central question about learning that applies to animals and machines, said P. Read Montague, a Baylor College of Medicine neuroscience professor spending this year at Princeton University who was not involved in the Nature paper.
Say that an animal is poking around, turns over a rock, runs around a bush, and then suddenly finds a grubworm, Montague said. The animal needs to learn what it did right to find the worm, he said, but so many things have happened that it does not know whether the most important step was the last thing it did or some action taken 50 events ago.
To learn efficiently, he said, an animal - or a machine - needs to "rehearse its experience to be able to assign credit at the right point in time." And why begin at the beginning when there may have been 10 or 20 false starts that led to nothing? Rather, he said, better to begin with "Bang! That really worked out! Now let's look back in time and see if we can assign proper credit to what really got me here."