It’s almost a good thing that we’ve never been entirely able to figure out how human memory works, because if we did, we’d probably just forget. Memory has always been that kind of meta-mystery, and one of its greatest puzzles is the question of what’s known as working memory: information we hold in short-term storage, like a phone number we’ll need to call or a face we’ll need to recognize at a meeting, and can then forget.
Unlike long-term memories, which are thought to be preserved in synaptic connections among nets of neurons that are effectively permanent, the neurons involved in short-term memories have to be able to decouple easily. Those temporary memories are forged at all, researchers have believed, thanks to a low level of electrical activation that keeps the particular pattern of brain cells linked for only as long as they have to be before they power down and the memory can be erased.
Now, however, in a paper published in Science, a team of investigators at the University of Wisconsin, Madison, have discovered an entirely different mechanism. Working memories, it seems, are preserved in a latent or hidden state, existing without any evident activation at all until the moment they’re needed.
The study, led by psychologist Nathan Rose, involved a sample group of subjects who participated in three different memory tasks. In one, they were asked to remember a face flashed on-screen and then select a match from a group of faces that were displayed later. In some cases, a precisely matching face would be among the later samples; in other cases it would be merely a very similar one. In either case the goal was to pick the best face from among the ones offered.
“We used computer generated, novel faces rather than, for example, famous faces,” Rose explained in an email to TIME, “so that people couldn’t simply hold onto associated names or details.”
The other tests were a bit more straightforward: remembering the direction in which a group of dots was moving and picking the closest match from a series of later groups; and remembering a word and selecting the closest rhyme for it from a group of other words. All of the tasks were made a bit more difficult by the fact that the original images flashed on the screen for only one second, followed by a 7.5 second pause, followed by a one-second flash for the later matching choices. What’s more, the subjects would have to keep all three original images in mind—the face, the moving dots and the rhyming words—before being tested on any one of them. In some cases they were told which one to expect to have to match up first. In other cases they weren’t.
During the testing, the subjects’ brains were scanned with functional magnetic resonance imaging and electroencephalograms which—with the help of pattern-analysis software—were able to spot peaks in the readings that indicated synaptic activation for specific memories. Under the old model of working memory, there would be detectable peaks in the synaptic connections that represented all three original images—since that would be the only way for the memories to exist—even if there was perhaps a slightly higher peak for the one that would have to be put to use first.
Instead, however, while there was indeed detectable neural activity for the so-called attended memory item (AMI)—the one that the subjects knew they would need right away—there was none at all for the unattended memory items (UMI), which the subjects might also need, but not until later. “The neural evidence drops back to baseline levels of activation, as if the item has been forgotten,” wrote Rose. All the same, when subjects were asked about a UMI, a peak appeared for it just as it did for an AMI. In both cases, working memory worked just fine, but in one case it did so without the benefit of any visible storage system.
To confirm the findings, Rose and his team used a pulse of transcranial magnetic stimulation—a low, harmless charge of magnetism applied to the scalp—to try to stimulate the dormant UMI sites artificially. The magnetism did cause the UMIs to register activity, but only until a particular round of the test was done and the subjects knew they with certainty they would not need any of the memories. At that point the magnetic stimulation didn’t work. The conclusion: unattended memories are maintained in what the researchers called “a privileged state” only as long as they had to be.
The study does not explain what does maintain working memories if low-level activation doesn’t, but changes in synaptic weights—or the potential one neuron has to affect the behavior of another across a synapse—might be the answer. Whatever the explanation, the work has implications for understanding not just memory but other cognitive functions like perception, attention and goal maintenance.
What’s more, Rose writes, “the results have exciting implications if noninvasive brain stimulation techniques can be used to reactivate and potentially strengthen latent memories”—in other words, recovering information that had been forever lost.
This article originally appeared on Time.com.