An animal crossing a patchy landscape can trust what its senses report right now, or lean on what it learned an hour ago. Memory runs on energy, and that cost decides whether it keeps one at all.
Storing information is never free. The bill arrives whether the memory sits in a brain, a bacterial signaling network, or the proteins inside a single cell.
When is remembering worth paying for? A theory paper from the Institute of Industrial Science and RIKEN answers with a sharp edge.
Takehiro Tottori and Tetsuya J. Kobayashi have found that an organism on a tight budget should ignore memory entirely.
The switch to using it, when it comes, happens all at once.
Memory has an energy cost
The pair built a deliberately small model of an organism tracking something that keeps moving. Call it a food patch, a chemical gradient, or a temperature.
The model organism cannot see that state directly. It gets a noisy reading, and it can write what it reads into an internal memory that carries noise of its own.
Then the accounting starts. The model charges for two things: how far the estimate lands from the truth, and how hard the organism works to keep the memory.
Two numbers set the terms. One stands for the energy available to spend, the other for what memory costs.
The cost is not hypothetical. Turning a fleeting memory into a lasting one takes a surge of energy at the synapse.
The mitochondria that supply it are spread unevenly through the brain, so some regions can afford more of it than others.
Why the switch is sudden
When the two researchers solved the model, the interesting part was not that memory sometimes lost. It was how it lost.
“This tradeoff can produce surprisingly dramatic behavior,” said Tottori. “When resources are scarce, the best strategy is to ignore memory and react only to current information.”
“But once enough resources become available, remembering suddenly becomes worthwhile, causing an abrupt shift to a memory-based strategy.”
There is no gentle ramp. Below a threshold, the optimal amount to store is exactly zero. Above it, the optimum jumps straight to a substantial value.
Earlier work by the same group found that jump numerically without explaining it. The new paper traces it to a double balancing act.
The organism has to tune two things at once: how much of each reading to write down, and how hard to hold the memory steady against its own noise.
Optimize one alone and the transition is smooth. Optimize both together and it snaps.
Noise has a sweet spot
The second result is stranger, and it is the one that matches experiments on people.
Memory pays off best when the senses are moderately unreliable. If incoming information is crisp, memory adds nothing the current reading does not already supply.
If the signal is hopelessly noisy, storing it only fills the memory with garbage. Between those extremes sits a band where the past genuinely improves the estimate.
That band is the only place an organism should pay for memory. Volatility works the same way: a world that changes fast makes old information worthless.
Three quantities in the model – the energy available, the price of memory and its internal noise – also turn out not to act independently. They collapse into one ratio that fixes the boundary.
People behave this way too
None of this was measured in a living organism. The paper is mathematics, and its authors are careful about that.
The pattern does have a match in the experimental record. A 2022 study had volunteers make inferences under different levels of uncertainty.
Those volunteers leaned on working memory in the middle range of uncertainty and abandoned it at both ends. That is the same non-monotonic signature the model produces.
Kobayashi studies biological information processing at the University of Tokyo.
“These results help explain why organisms do not always use memory, even when it could in principle improve their decisions,” said Kobayashi.
“Whether memory is useful depends not only on the resources available, but also on environmental uncertainty.”
The same experiments showed that a more volatile environment pushed people toward memoryless strategies. The framework reproduces that behavior as well.
What the math cannot show
The model was intentionally simple, which is also its biggest limitation. It assumed a single changing environmental factor, Gaussian noise, and quadratic costs.
Real nervous systems are none of those things. The point is that even this tidy linear setup produces an abrupt jump.
It does not establish that any particular cell or animal sits on one side of the boundary. Nothing living was measured here.
One part of the model also remains unresolved. The researchers identified the boundary between two conditions they could solve exactly, but what happens in the narrow gap between them is still unclear.
The link to real circuitry is a suggestion, not a result. Memory-based estimation here resembles a coherent feed-forward loop, which pairs a fast direct path with a slow indirect one.
Those loops are overrepresented in gene networks and neural circuits, which is what makes the resemblance worth chasing.
Evolution of costly brains
The reason a physics journal took this on is the evolutionary implication. Biological computation is not spread evenly across life.
Some systems get by on reflexes. Others maintain costly machinery for holding on to the past.
Simple cells can be wired to compute like small circuits. A human brain, at the other end, runs close to the physical limit of what its energy buys.
If the value of memory has a threshold, that spread is what a threshold would produce.
The framework offers a way to think about why memory-hungry computation like the brain arose at all. It appeared where an organism could afford it.
Memory also had to earn its keep in an environment uncertain enough to reward it.
To remember, or not to remember: that is the question evolution has been answering, one energy budget at a time.
The study is published in the journal Physical Review Letters.
—–
Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
—–