A paper about memory can go wrong before it reaches neuroscience if the word for is left undefined.

Memory is useful for planning. It helps us imagine futures that have never happened. It supplies evidence in conversation, supports learning, and sometimes lets us revisit a particular afternoon for no practical reason at all. Listing those uses does not yet tell us which effect, if any, explains why episodic memory evolved.

José Carlos Camillo's new paper in the European Journal for Philosophy of Science takes that historical question seriously. He defends the mnemonic hypothesis: episodic memory evolved to store and process information about particular past events so that their details and spatiotemporal relations can later be retrieved accurately enough to be useful.

That sounds like the obvious answer. It has not been the obvious answer in the literature for some time.

The future-oriented challenge

Episodic memory is reconstructive and fallible. People omit details, combine episodes, and confidently remember things that did not happen. The same broad neural machinery involved in remembering past events is also recruited when people imagine possible future events.

Daniel Schacter and Donna Rose Addis turned that overlap into the constructive episodic simulation hypothesis in 2007. A memory system that can take pieces of past experience apart and recombine them is well suited to constructing events that have never happened. On this picture, some of memory's apparent defects may be costs of a system built for flexibility rather than faithful replay.

Other proposals push in related directions. Episodic memory has been described as a system for future simulation, counterfactual reasoning, prediction, general learning, or communication about past events.

Camillo does not deny that memory supports those things. His question is more specific: what is its evolutionary function, in the selected-effects sense? In other words, which effect helped explain why natural selection retained and shaped the capacity?

That distinction blocks an easy inference. A trait can be useful for something now without having evolved for that use. Feathers can be used for flight even if their earlier evolutionary history involved thermoregulation. A heart makes a sound, but the sound is not what explains the organ's existence.

Remembering may be more accurate than its reputation

Camillo's positive case is not simply that remembering the past feels like what memory is for.

He points to work on naturalistic memory suggesting that event memory can be highly accurate when researchers test personally experienced, spatiotemporally structured events rather than isolated laboratory material. He also argues that the hippocampal system is organized in ways that preserve event-specific spatial and temporal relationships, and that damage to this system predictably disrupts that capacity.

None of this turns memory into a recording device. A 2024 wearable-camera study by Jason Finley and William Brewer is useful precisely because it separates accuracy from completeness. Participants forgot substantial portions of their days, but the material they did recall was usually accurate. A system can fail to preserve everything without being indifferent to whether what it preserves is true.

Camillo then looks at evidence that memory and imagination can dissociate. Some patients with severe episodic-memory impairment can still construct fictive scenes, although often less easily and by relying on somewhat different neural resources. His argument is that shared machinery and interaction do not prove that remembering and imagining are one functional system.

That is a philosophical point disguised as a neuroscience point. Two capacities can depend on one another, overlap in implementation, and still have different functions.

Two other 2026 papers make the debate less binary

Rebecca Dreier's February paper, “Accuracy first: episodic memory's evolution,” reaches a conclusion compatible with Camillo's in an important respect but rejects the idea that the candidate functions must be mutually exclusive.

Dreier argues that episodic memory may have mnemonic, future, and communicative proper functions. Her stronger claim is about order and dependence: the future and communicative functions rely on an episodic system already capable of producing tolerably accurate representations of particular past events. On her account, the mnemonic function is therefore the most basic or evolutionarily prior one even if later selection added other functions.

Camillo himself leaves room for something similar. He explicitly focuses on the proximal function of episodic memory and does not rule out additional functions at higher levels. Memory might have the proximal job of preserving event-specific information while that preserved information, in turn, supports planning, imagination, communication, and prediction.

An empirical study published in Nature Human Behaviour in January makes the distinction unusually concrete. Jonathan Nicholas and Marcelo Mattar ran five experiments with 535 participants. People encountered events containing several features and later had to make decisions using whichever details became relevant. Participants relied especially on episodic memory when they could not know during encoding which features would matter later.

The experiment does not tell us what natural selection selected episodic memory for. It does show why keeping detailed records of particular events can have future value. When the future demand is unpredictable, preserving the particulars gives you options later.

That is a more interesting relationship between past and future than simply choosing one as memory's true purpose.

The unresolved part is historical

The strongest version of the future-oriented argument says that because remembering and imagining share mechanisms, episodic memory should be understood principally as part of a system for simulating possibilities. Camillo thinks the dissociation evidence makes that inference too quick. Dreier thinks even a genuine future function could depend on an earlier mnemonic one.

Both positions face the same hard limit: evolutionary history does not leave us a clean experimental record of why a cognitive system was selected. Camillo therefore combines causal-role evidence, dysfunction, optimality models, and arguments against rival hypotheses. Dreier reconstructs a plausible order in which functions could have arisen. Neither can directly observe the ancestral sequence.

So the question has become more precise without becoming settled. A memory can be incomplete, reconstructive, useful for imagining the future, and useful in communication while still belonging to a system whose most basic selected effect is retaining information about particular past events. What remains difficult is establishing which of those benefits came first and which were added later.

Sources

Ulix app

Guten

Guten is an Android reader for Project Gutenberg, with reading, read-aloud, collections, notes, and tools for keeping long books usable rather than merely stored.