Your brain doesn’t run out of room

You know the feeling. You’re watching for your exit, your phone buzzes, someone asks you a question, and for a second you’re worse at all three. Or you sit down to work without knowing whether the next hour’s for writing or for answering email, and somehow you manage to do both badly. We treat this as ordinary, and we treat the explanation as settled. The brain’s got a fixed amount of processing room, the story goes, and when you ask it to hold too much at once, it runs out.

A study published in Nature Neuroscience on September 10 says that story’s wrong, or at least badly incomplete. The cost of not knowing which task’s coming isn’t your brain running out of space. It’s your brain getting in its own way.

A rule that keeps moving

The work, led by Cheng Xue and Marlene Cohen, ran the same problem three ways and lined the results up against each other. More than 200 people did the task online. Two rhesus monkeys did it while the team recorded from neurons in their primary visual cortex and their parietal cortex. And an artificial neural network was trained to do it too.

The task was slippery on purpose. You watch a striped circle shift from one image to the next and report which of two things about it changed, how wide its stripes are or how it’s rotated, and on some trials how sure you are. The catch is that nobody tells you which of the two features is the one that counts. You have to work that out from whether your recent answers came back right, and every so often the feature that matters switches with no warning. Not knowing which rule’s live at any given moment is the uncertainty, and once you count which way the feature moved it’s a four-way call, so a blind guess only gets you to 25 percent.

The behavior came out the way you’d expect for the living brains. Both the monkeys and the people were reliably less accurate when they weren’t sure which feature mattered. The artificial network, trained on the exact same work, showed no such cost. It handled the uncertainty for free.

If the loss were a basic limit of the computation, the network should pay it too. It didn’t.

That gap’s the whole clue. A machine doing the identical task doesn’t get worse under uncertainty, so the cost can’t be baked into the problem. It’s got to be something about how a biological brain goes about solving it.

What the AI gave away

To find that something, the researchers trained a second network. This one they didn’t reward for getting the answer right. They trained it to copy the animals’ actual choices, mistakes and all. That copy reproduced the uncertainty cost, and because it’s a model they could open it up and look inside. When they did, and then looked back at the real neurons, they found the same signature in both.

Here’s what it is, in plain terms. When you know which feature matters, your brain holds a clean, strong representation of that feature and lets the other one stay quiet. When you don’t know, it can’t afford to ignore either one, so it builds up a stronger representation of the very feature it should be tuning out. Xue puts it simply. “When we aren’t sure which rule to follow, our brains hold onto visual details we’re supposed to ignore. That extra information bleeds into our decisions, directly contaminating how well we perceive the things that actually matter.”

And it gets worse than just holding on. The neural codes for the two features stop being cleanly separable. They entangle. Reading out the one you care about now drags along the one you don’t, and your answer picks up the noise. Cohen’s description is blunt. “Information gets mixed together in your brain even if you know it very well.” The team calls this feature interference, and they didn’t just watch it happen. On half the trials they nudged neurons in the monkeys’ visual cortex with a small electrical current, and the interference grew exactly when uncertainty was highest, which is the effect the account predicts.

A ceiling, or a tangle

The reframing’s the reason this matters beyond one clever paradigm. For decades the limits on attention, working memory and multitasking have been described as capacity, a fixed ceiling on how much the brain can handle at once. It’s why you can hold only a handful of things in mind, why a second task drags down the first, why you can’t really follow two conversations. This study argues those limits might not be a ceiling at all. They might be interference, the crosstalk between the brain’s representations of the different things it’s trying to hold.

That’s a different kind of limit, and a more hopeful one. A ceiling’s a ceiling. You can’t train your way through a wall. Interference is a property of how information’s arranged, and arrangements can, in principle, change. It suggests the reason you fumble under uncertainty isn’t that your head’s full. It’s that two signals are fighting over the same space.

None of this makes the feeling go away. But next time you catch yourself getting worse because you’re not sure what you’re supposed to be doing, it’s worth knowing what’s going on. Your brain’s not out of room. It’s holding two things that won’t stay out of each other’s way, and for a moment, neither one comes through clean.