We treat forgetting as a failure of memory and a problem to be solved. We highlight, take notes, build review systems, and curate second brains — all in service of the goal of not losing the things we have learned. This goal is legitimate. The diagnosis beneath it is not. Forgetting is not noise in the system of learning. It is a signal — about what the mind considers worth retaining, about how retrieval actually works, and about why the struggle to remember produces deeper understanding than frictionless access ever will.
About a year after finishing a book that had significantly shaped how I thought about a particular topic, someone raised exactly that topic in a conversation, and I found I could say almost nothing specific. I knew I had read the book. I knew, in a vague way, that it had been important — that it had shifted something in how I understood the problem. But the arguments, the key claims, the evidence, the vocabulary the author used — essentially all of it was gone.
The embarrassment of this is distinctive. It is not the embarrassment of not having read the book; I had read it carefully, slowly, with attention. It is the embarrassment of having invested in something and lost the investment — of having spent time and effort and then having almost nothing to show for it. Of having been changed by something you can no longer describe.
I remember wanting to explain this experience as a failure — of the book for not being memorable enough, or of myself for not taking sufficient notes, or of memory as a general faculty for being so unreliable. It took me considerably longer to consider the possibility that what had happened was not a failure of anything — that forgetting, in this specific case, had functioned exactly as it was supposed to, and that what felt like loss was something more complicated.
What Ebbinghaus Showed, and What We Have Done With It
In 1885, the German psychologist Hermann Ebbinghaus published Über das Gedächtnis (On Memory), the result of years of experiments he conducted on himself — memorising lists of nonsense syllables, testing his own recall at intervals, plotting the rate of decay. What he produced was the forgetting curve: a mathematical description of how memory for newly learned information declines over time, steeply at first and then more slowly, until it reaches a plateau.
The curve is usually encountered in one of two contexts. In educational psychology, it is presented as a problem to be solved: here is how quickly you forget what you learn, and here is why spaced repetition — reviewing material at increasing intervals — is the most effective way to combat the decay. In productivity culture, it is presented as a motivation for PKM systems: here is why your notes are more reliable than your memory, and here is why building an external second brain is essential to doing knowledge work seriously.
Both of these framings are useful. Both are also incomplete in a way that matters. They treat the forgetting curve as a fact about memory's unreliability — as evidence that the system is failing and needs to be compensated for. What they do not fully engage with is the question of why the system works this way, and whether working this way might be, in the relevant sense, correct.
Retrieval Strength and Storage Strength
The most important theoretical framework for understanding forgetting comes not from Ebbinghaus but from the work of Robert Bjork and his colleagues, who introduced a distinction that substantially changes how the forgetting curve should be interpreted: the distinction between storage strength and retrieval strength.
Storage strength, in Bjork's framework, is how well-established a memory is — how deeply it has been encoded, how many connections it has formed, how robustly it is integrated into the network of existing knowledge. Storage strength, once established, does not decrease. This is the remarkable finding: memories, in a technical sense, do not disappear. What disappears is retrieval strength — the accessibility of the memory, the ease with which it can be retrieved from storage.
The forgetting curve, understood through this distinction, is not a curve of memory loss. It is a curve of retrieval strength decay. The information is still there, in the sense that it can often be relearned considerably faster than it was originally learned — a phenomenon Ebbinghaus himself noted and called savings, the reduction in effort required to relearn something that was once known. The memory has not been destroyed; it has become inaccessible.
This distinction has a counterintuitive implication that Bjork has spent decades drawing out: storage strength and retrieval strength do not grow together in the way we naturally assume. Conditions that make retrieval easy tend to produce low storage strength. Conditions that make retrieval difficult tend to produce high storage strength. The ease of access, in other words, is inversely correlated with the depth of encoding.
Desirable Difficulties
Bjork coined the term desirable difficulties to describe the conditions that are unpleasant in the short term and productive in the long term — the conditions that lower retrieval strength in the moment of learning while raising storage strength for the future.
The most studied of these is retrieval practice itself: the act of trying to recall something from memory, without cues, is a more effective learning strategy than rereading the same material. This has been replicated extensively in educational research and is about as robust a finding as exists in cognitive psychology. The explanation, in Bjork's framework, is straightforward: attempting retrieval exercises the storage strength of a memory in a way that rereading does not, because rereading provides the answer without requiring the memory to produce it. The difficulty of retrieval is the mechanism by which the memory is strengthened.
Spaced practice — distributing learning over time rather than massing it — is another desirable difficulty. It works, in part, because the spacing allows retrieval strength to decline, and the recovery of a memory from low retrieval strength exercises it more than the recovery of a memory from high retrieval strength. Studying something immediately after learning it, when retrieval strength is high, is easier but less effective than studying it a day later, when retrieval strength has declined. The effort of the harder retrieval produces better long-term retention.
Interleaved practice — mixing different types of problems rather than blocking practice by type — is similarly counterintuitive. Blocked practice, in which you practise the same type of problem repeatedly before moving to the next type, feels more efficient because performance during practice is higher. Interleaved practice, which requires you to identify the appropriate strategy for each problem rather than applying the same strategy repeatedly, feels harder and produces lower performance during practice — and consistently produces better retention and transfer in subsequent tests.
The pattern is consistent: what makes learning feel easy tends to make it shallow. What makes it feel hard tends to make it deep.
What the Forgotten Book Left Behind
This framework allows a different reading of the experience I described at the beginning — the book that shaped my thinking and left almost no specific content behind.
The book had not left nothing. It had left something that is harder to describe than arguments and evidence: a changed orientation toward the problem, a set of categories that had become available for use, a way of framing questions that felt natural without being traceable to any specific source. The vocabulary was gone; the mechanism for generating the vocabulary remained.
This is the distinction between knowing that and knowing how — between propositional knowledge, which can be stated, recalled, and transferred, and procedural knowledge, which manifests in what you are able to do rather than in what you are able to say. Much of what reading does is not the transfer of propositions — claims that can be stored and retrieved — but the modification of cognitive habits, the expansion of what can be noticed, the calibration of judgment in domains where judgment is needed.
Forgetting the explicit content of something you have genuinely understood is not the loss of the understanding. It is the conversion of the understanding from explicit to implicit — from something you can report to something that shapes how you see. This conversion is neither complete nor reliable, and some explicit content is genuinely worth retaining in explicit form. But the anxiety about forgetting specific content sometimes obscures the recognition that the most valuable thing reading does is often the least recoverable.
Memory Is Not a Recording Device
Underlying the misunderstanding of the forgetting curve is a misunderstanding of what memory is — one that is so pervasive it operates almost entirely below the level of conscious assumption.
The intuitive model of memory is something like a recording device: experiences and information go in, get stored somewhere, and can be retrieved when needed. On this model, forgetting is a retrieval failure — the recording is still there, but the playback mechanism has lost track of it. The goal of a good memory, or a good memory system, is to prevent playback failures: to make the recordings as accessible as possible, for as long as possible.
This model is wrong in ways that matter practically, not just theoretically.
Memory, as cognitive scientists now understand it, is not a retrieval process but a reconstruction process. You do not retrieve memories from storage the way you retrieve a file from a hard drive — finding the same information in the same form, unchanged from when it was saved. You reconstruct memories each time you access them, using a combination of fragmentary stored information and current knowledge, expectations, and context. The reconstruction feels like retrieval; it is seamless in the way that retrieval from storage would be seamless. But it is not the same thing, and the difference matters.
First: reconstruction means that memories change each time they are accessed. The act of remembering something subtly modifies it — incorporates new context, resolves ambiguities in the direction of current understanding, fills gaps with plausible inferences. This is why eyewitness testimony is unreliable; it is why people's memories of significant events are often contaminated by subsequent information about those events; and it is why the experience of "remembering" something very clearly is not a reliable indicator of its accuracy. The vividness of the memory reflects the confidence of the reconstruction, not the fidelity of the original encoding.
Second: reconstruction means that what gets retained is not an exact copy of what was encoded, but the gist — the structural features, the causal relationships, the meaning — together with the ability to reconstruct the details from the gist. This is a feature, not a flaw. The gist is what is needed for transfer — for applying what you have learned to new situations that share the structure but not the surface features of what you originally encountered. Verbatim memory would be less useful for transfer than gist memory, because verbatim memory would produce recognition of identical situations without enabling response to structurally similar but superficially different ones.
This is why the book that left behind no specific content left behind a changed orientation: what the mind retained was the gist — the structure of the argument, the way of framing the problem — and released the verbatim content that was no longer needed for reconstruction. The loss was real. But what remained was, in terms of its utility for thinking, more valuable than what was lost.
The PKM System That Cannot Be Built
This analysis creates a problem for a certain kind of PKM ambition.
The most common model of a personal knowledge management system is a system that compensates for forgetting — that captures what the mind would otherwise lose and makes it available for later retrieval. The second brain, the external memory, the searchable archive of everything you have learned. On this model, the goal is to minimise the cost of forgetting: to ensure that whatever understanding you develop is not lost, that the investment in reading and thinking produces a durable and recoverable return.
The problem is that the most important things understanding produces cannot be stored in a PKM system. The changed orientation, the expanded set of categories, the calibrated judgment — these are modifications to you, not facts about the world. They cannot be noted, filed, or retrieved. They can only be developed, and they are developed through the process that the PKM system is designed to compensate for: the slow, effortful, often frustrating engagement with difficult material, including the engagement that is required by low retrieval strength, by forgetting, by the necessity of reconstruction.
A PKM system that functions as a complete external memory — that ensures you never need to retrieve anything from your own memory because it is all available externally — does not solve the problem of forgetting. It eliminates the condition under which deep learning occurs. If you never need to retrieve something from memory because you can always retrieve it from your notes, you never exercise the retrieval process that builds storage strength. You maintain high retrieval strength for the external store and low storage strength for everything in it. The notes become a prosthetic for understanding that never developed.
What Systems Are Actually For
The correct response to the forgetting curve is not to build a system that prevents forgetting. It is to build a system that uses forgetting — that structures review according to the spacing effect, that requires retrieval rather than recognition, that creates the conditions under which the effort of remembering can do its work.
Spaced repetition software — Anki is the most widely used — is a PKM tool built on precisely this understanding. It does not prevent forgetting; it schedules retrieval at the point where retrieval strength has declined enough to make the retrieval effortful. The discomfort of struggling to remember something is not a bug in the Anki experience; it is the mechanism by which the software produces its effects. The effort is the learning.
More broadly, the most valuable function a PKM system can serve is not storage but provocation — not the preservation of ideas but their reintroduction at intervals sufficient to create genuine effort in the retrieval. Notes that you write once and never revisit are performing a storage function, which is not nothing, but is considerably less than what they could do. Notes that you encounter again — especially when you have partially forgotten their content, when the retrieval requires reconstruction — are producing the conditions for genuine learning.
This reframes the design problem for a PKM system substantially. The question is not "how do I ensure I never lose anything I have learned?" It is "how do I structure my relationship with what I have learned so that the engagement is consistently effortful, and the effort consistently builds?" The answer to the second question is different from the answer to the first, and it involves not the prevention of forgetting but its deliberate use.
The Embarrassment Revisited
I have thought about the book differently since working through these ideas. The embarrassment I felt — the sense of having invested and lost — was not entirely misplaced, but it was pointing at the wrong thing.
The loss was real: the specific arguments, the evidence, the vocabulary are gone and would need to be reacquired. If I needed those specific things for a specific purpose, I would need to reread the book. This is a genuine cost of not having engaged with the material again after reading it — of not having subjected it to the spaced retrieval that would have maintained its accessibility.
But what I had described as the whole of the loss — the feeling of having been changed by something I could no longer describe — was not a loss. It was a description of what genuine reading actually does: it modifies the reader in ways that the reader cannot fully inventory. The change is less visible than the content; it does not produce citable claims; it cannot be shared as evidence of having read something carefully. But it is, frequently, more durable and more significant than anything that could have been stored in a note.
The forgetting curve is real. The response to it is not to eliminate forgetting but to understand what forgetting is — not the destruction of learning but the natural behaviour of a memory system that maintains what is actively used and allows what is not actively used to become less accessible. The problem is not the curve. The problem is the passivity that leaves learning to decay without being exercised — the reading that happens once and is not returned to, the idea that was interesting enough to highlight but not important enough to retrieve and test and build on.
The goal of a knowledge management system is not a brain that never forgets. It is a thinker who engages with ideas often enough, and with enough effort, that the most important things are retained not because they were stored but because they were used. Forgetting is not the enemy of that goal. It is, correctly understood, one of its primary instruments.