What memory is not, and what this course is for
95 min
Two hosts talk the lesson through. The voices are synthetic; the script was written from this lesson and checked against it, and asserts nothing the lesson does not.
- State the difference between a recording model of memory and a reconstructive one, and say what each predicts
- Say what this course promises and what it refuses to promise, and identify a claim about memory that the course could not support
- Name the population and the sample size behind each of this course's main sources, and say what that licenses
The popular literature on memory is unusually confident, and most of it assumes memory is a recording that has got worse. This course is about what it actually is, which is stranger and more useful.
The recording and the reconstruction
Ask most people how memory works and you get some version of a recording. Something happened, it went in, and remembering is playing it back. On that model a memory is either intact or degraded, and forgetting is loss.
You have already been given the better account, and it is worth saying where. How to Learn Anything used reconstruction to explain why retrieval practice works: recall rebuilds a memory rather than reading it off a shelf, which is why the rebuilding strengthens it. That course used reconstruction to explain what builds a memory. This one is about what reconstruction does to one afterwards, which is the whole boundary between them.
The account the research uses is different, and the difference is the whole course. Remembering is an act of assembly. What you have stored is partial, and what arrives in your head is built from it each time, using what else you know, what you have been told since, and what the question you were asked suggests.
Those two models predict different things, which is what makes this a claim rather than a metaphor. The three contrasts below are this course's own way of setting them out, drawn from what lessons 3, 4 and 5 report rather than from a source that lays them out this way.6
On a recording model, how confident you feel about a memory should track how accurate it is, because confidence would be reading off the state of the recording. On a reconstructive account it need not, because the confidence is generated along with the memory rather than before it. Lesson 5 is that case, and the answer there turns out to be conditional.
On a recording model, asking somebody about an event doesn't change the recording. On a reconstructive account it can, because the question supplies material to the assembly. Lesson 4 is that case.
On a recording model, recalling one thing has no effect on the recordings next to it. On a reconstructive account it might, because retrieval is an act rather than a read. Lesson 3 is that case.
Before you read on. If remembering were playing back a recording, what would you expect a study of forgetting to look like? And if it were assembly, what would change? Two sentences each, before the lesson answers.
Show the answer
Most people get the first and the second is where the work is.
On a recording model, forgetting is a story about the recording: degradation, or loss, and the sensible thing to measure is how much of the original is still there. That's where "you forget half of what you learn within an hour" comes from, and it is a very natural way to think about it.
On a reconstructive account, forgetting is a story about access, which is the distinction How to Learn Anything lesson 1 taught as storage strength against retrieval strength. This course assumes it rather than re-teaching it: material can still be there and not be reachable from where you are standing. That's a testable difference and the next lesson's measure is built on it: the method of savings asks not what you can produce, but how much faster you relearn something you once knew.
And the second prediction, which is the one to sit with, is that on a reconstructive account forgetting has causes other than time. Something can become hard to reach because something else was reached instead. That is lesson 3, and interference of that kind is part of why forgetting is not simply a function of elapsed time.
One term from the next lesson, because the scope table below uses it. The 2015 study's measure is savings: not how much somebody can produce, but how much of the original learning time they are spared when they learn the same material again. Lesson 2 is about why that distinction matters more than it sounds.
What this course is and isn't
How to Learn Anything already taught the constructive half of this subject, and taught it well. Retrieval practice, spacing, interleaving, chunking, and the distinction between storage strength and retrieval strength are all there. This course uses every one of them and re-teaches none.
What it adds is the other half. Not how to get something in, but what happens to it afterwards: what forgetting looks like when somebody plots it, what retrieving one thing does to its neighbours, what a question does to a memory, what a confident recollection is worth, and what the techniques people sell actually do.
It won't give you a better memory, and that is a statement about what the research is rather than modesty. Lesson 6 is the closest thing to a technique in the course, and its honest verdict is narrower than the advertising.
And one refusal is worth stating now, because it is this course's sharpest. This course can't tell you whether a particular thing you remember actually happened. It measures what memory does across people, in conditions somebody set up, and that is a different question from adjudicating one recollection. A reader who takes lesson 4 and uses it on a relative has misunderstood what the lesson is. Lesson 7 comes back to it with the whole course behind you.
The gap in the evidence, named now rather than discovered later
This course reads five things properly. Two of them are open access: the forgetting-curve replication and the retrieval-forgetting meta-analysis. Here is what each of the five was measured on.12345
| The source | What it measured | On whom | When |
|---|---|---|---|
| The forgetting curve replication | Savings on relearning lists of nonsense syllables, at six intervals | One subject, over 70 hours | 2015 |
| The retrieval-forgetting meta-analysis | Recall of unpractised items, pooled | 472 samples from the published literature; this course did not read who was in them | 2014 |
| The leading-question experiments | Speed estimates, then a question about broken glass | 150 participants in the second experiment | 1974 |
| The eyewitness synthesis | Confidence against identification accuracy, reviewed | Adults, on initial identifications, under lineup conditions many places do not meet | 2017 |
| The mnemonic training study | Recall after six weeks of training, three arms | About 50 people, plus memory athletes | No date recorded: a preprint, and this course did not locate the published version |
Read the third column. One subject. A hundred and fifty undergraduates. About fifty people. These are the smallest samples this course teaches from, and that isn't a failure of these particular papers; it is what this subject's best evidence looks like.
Work one row, because the table is easier to read than to use. Take the first. What was measured is savings on relearning, which isn't the same as how much somebody could recall. Who was measured is one person. When was 2015, replicating a study from the 1880s. So a sentence beginning "people forget" is already wrong at the second word, and the sentence the row supports is: "in one subject, relearning lists of nonsense syllables after intervals from twenty minutes to thirty-one days, the savings curve reproduced the shape reported in the 1880s." That is duller, and it is the one lesson 2 can defend.
And it cuts both ways. A study of one person is a real limit and it isn't nothing: the whole point of that study is that an older result reproduces, which is a claim about the result rather than about people. What the small samples rule out is the sentence "people do". Every figure in this course arrives with its sample attached, in the body, where you meet it.
Worked: one event, remembered twice
This case is constructed, because no source read for this course reports an individual's two accounts side by side.6
Ravi describes a car breaking down on a family holiday. In the first account, written this year: the car stopped on a hill, his father walked to a farmhouse, they waited about two hours, and it was raining.
Eight months later he writes it again, without looking at the first. The car stopped on a hill. His father walked to a garage. They waited most of the afternoon. It was raining hard and his sister was crying.
Four differences, and the four readings below are this course's own reasoning from a reconstructive account rather than findings anybody has reported.6
- Farmhouse to garage. A detail replaced by a more typical one. If assembly draws on what usually happens, a garage is what usually happens when a car breaks down.
- Two hours to most of an afternoon. If duration is rebuilt from how the event felt rather than read off a clock, it has nothing holding it in place.
- Raining to raining hard. Intensification, which is the direction this course would expect and has not measured.
- The sister crying is new. The most interesting one, because it could be a recovered detail that was always stored, or an inference, or something from a different day.
What the case can't show you is which. That is the honest position and it is also the point: a reconstructive account predicts that differences will appear and doesn't tell you which direction is the true one. Ravi is made up, and so is every detail.6
Somebody says this all adds up to memory being unreliable, so nothing anybody remembers should be trusted. Is that what the course is saying?
Show the answer
No, and getting this right is worth more than any finding in the course.
Reconstruction is a claim about the mechanism, not a verdict on the output. A process that assembles can assemble accurately. The most striking result in this whole course is in lesson 5: in adults, on an initial identification, under five named lineup conditions, a confident identification turns out to be right the great majority of the time. Those identifications are reconstructions like every other memory. Every one of those qualifications is load-bearing, lesson 5 gives the figures with them attached, and the word "initial" in particular is what the whole claim rests on.
What reconstruction does buy you is a prediction about where the errors will be. Not randomly distributed, but in the places assembly has to do work: gaps filled with the typical, durations drifting, details supplied by whatever was said afterwards. That is far more useful than a general warning, because it tells you which parts of an account to check.
And the over-correction has a cost. A person who concludes that no memory can be trusted has no way to treat one account as better than another, which is exactly the judgement this subject can help with. Lesson 5 is about a field that held the over-corrected view for thirty years, for reasons its own synthesis calls entirely reasonable, and then measured its way to something more useful.
So the sentence to carry is narrower and more useful: memory is reconstructed, reconstruction fails in specific ways, and knowing the ways is what lets you tell a strong account from a weak one.
Five things people believe about memory
"Memory is a recording." Everything in this course is downstream of this being false, and it is what most people mean by the word.
"Reconstruction means memory is unreliable." The over-correction above. It confuses how memory works with how well it works.
"How sure you feel tells you how accurate you are." Lesson 5 is the one place in this course where a feeling about a memory does predict accuracy, and it does so only under five lineup conditions that many police forces have not adopted.
"Small samples tell you nothing." This would throw away the founding measurement of the subject. A small sample makes a finding narrower, not empty.
"A famous study is a strong study." Lesson 4 is about the most famous experiment in this subject, which found its key result in sixteen people against seven and six.
What this course leaves out on purpose
Memory and ageing. The thing readers most want to ask about and the literature this course hasn't searched. A course that improvised it would be doing harm, given what people are frightened of.
Clinical memory disorders. A different literature and a different question.
Recovered and repressed memory. The most contested question in the whole subject, and this course has read none of it. It's named here so that its absence is a decision rather than an oversight.
The neuroscience. Real accounts exist and this course hasn't read them.
Practice
Take 30 minutes. This is the first section of the course project, and it can't be done later.
Pick one remembered event, at least two years old, that you can describe in detail. Not the most important thing that ever happened to you; something ordinary enough that you haven't retold it many times.
Write it out in as much detail as you can give. Aim for specifics rather than significance: what was said, what things looked like, how long things took, who was where, what the weather was.
Then three lines at the end.
- Today's date.
- When you think the event happened, as precisely as you can.
- How confident you are, out of ten, that the account is accurate.
Then save it somewhere you won't read it again. The last lesson asks you to write it a second time without looking, and the comparison is the project. If you read it in between, the project is spoilt, and there is no way to unspoil it.
Take 20 minutes.
Find three claims about memory. An article, an app, a book jacket, something somebody told you.
For each, write down three things.
- The claim, word for word, and where you found it.
- What would have had to be measured, and by what method: recall, recognition, savings on relearning, an identification, a self-report about one's own memory, something else.
- On how many people, and who, if the source says.
Then one line: how many of the three told you either of the last two?
In this subject the usual answer is none, and the claims are usually about everybody.
Connections
Back. How to Learn Anything is the course this one continues, and its lessons 1 to 5 are assumed throughout rather than repeated. Focus and Deep Work lesson 1 taught the instrument question, which this course uses in lesson 2 and lesson 5. Habits and Self-Discipline lesson 2 is the nearest relative of the scope table here.
Forward. Lesson 2 is the founding measurement of the subject and its sample of one. Lesson 3 is what retrieval costs the things you did not retrieve, and is the lesson that talks to How to Learn Anything most directly. Lesson 4 is what a question does. Lesson 5 is the best-evidenced thing in the course and the hardest to write. Lesson 6 is the techniques. Lesson 7 is what stays useful once every figure here has dated.
Go deeper
- Replication and Analysis of Ebbinghaus' Forgetting Curve (PLOS ONE, 2015). Open access, read in part by this course. Lesson 2 is about it. Read the abstract before that lesson, because the sample is in its second sentence and meeting it cold is instructive.
- The Relationship Between Eyewitness Confidence and Identification Accuracy: A New Synthesis (Psychological Science in the Public Interest, 2017). Read in part by this course. Fifty-six pages, and its one-page Summary is the best short statement of how a field changed its mind that this institute's reading lists contain.
Sources
- Jaap M. J. Murre and Joeri Dros, "Replication and Analysis of Ebbinghaus' Forgetting Curve", PLOS ONE 10(7), 2015, e0120644. Read in part: the abstract verbatim and the summary statements of method and result. The curve fits were not read. Supports this lesson's scope row only: one subject, 70 hours, six relearning intervals, and the method of savings. Lesson 2 uses it properly.
- Kou Murayama, Toshiya Miyatsu, Dorothy Buchli and Benjamin C. Storm, "Forgetting as a Consequence of Retrieval: A Meta-Analytic Review of Retrieval-Induced Forgetting", Psychological Bulletin 140(5), 2014, pages 1383 to 1409. Read in part: the abstract verbatim, the overall effect sizes and the output-interference analysis. Supports this lesson's scope row only. Lesson 3 uses it properly.
- Elizabeth F. Loftus and John C. Palmer, "Reconstruction of automobile destruction: An example of the interaction between language and memory", Journal of Verbal Learning and Verbal Behavior 13, 1974, pages 585 to 589. The original was not opened. What this course read is a published teaching summary of it, and lesson 4 says so where the figures appear. Supports this lesson's scope row only.
- John T. Wixted and Gary L. Wells, "The Relationship Between Eyewitness Confidence and Identification Accuracy: A New Synthesis", Psychological Science in the Public Interest 18(1), 2017, pages 10 to 65. Read in part: the Summary, Box 3, the section introducing the five recommendations, and one field-study passage. The evidence review was not read. Supports this lesson's scope row, and the checkpoint's statement that under five named lineup conditions an initial confident identification in adults is right the great majority of the time. The figures behind that sentence are in lesson 5, with their conditions, and this lesson gives no number.
- Isabella C. Wagner and colleagues, "Durable memories and efficient neural coding through mnemonic training using the method of loci". Read in part: the abstract and the participant counts, from the preprint. The published version was not located and the body was not read. Supports this lesson's scope row only. Lesson 6 uses it properly.
- Ravi is constructed, and so is every detail attached to him; the lesson says so where he appears and again where it says which difference is which. The four readings of his four differences are this course's own reasoning from a reconstructive account as described, not findings anybody has reported, and the body marks them as such. The three contrasts that separate a recording model from a reconstructive one are also the course's own framing, marked inline where they appear, and drawn from what lessons 3, 4 and 5 report rather than from a source that sets them out this way.
Check your understanding
This lesson has a 6-question quiz. Pass it and the questions come back on a schedule in Review, so what you learned stays learned. Your progress is saved in your browser; no account needed.