Hacker News new | past | comments | ask | show | jobs | submit login
The forgetting curve explains why humans struggle to memorize (qz.com)
214 points by prostoalex on March 7, 2018 | hide | past | favorite | 50 comments

If you're interested in improving your learning, I can't recommend the free course "Learning how to learn" enough:


It includes a section about this. I've also found many of the lessons applicable to other parts of my life.

Another free course made by the same team: https://www.coursera.org/learn/mindshift

This course is more focused on how/what to learn to boost your career.

I liked both courses; they take 2-3 hours per week for 4 week.

In Ebbinghaus' original work, it is simply about memorizing syllables. However, that is only a small percentage of learning harder skills like computer science (learning syntax), language (learning words), or even acting (learning lines). The next steps would be to synthesize (Bloom's taxonomy) the newly acquired knowledge in some way to facilitate higher levels of learning (CS - problem-solving, language - conversational fluency, acting - character work).

I think this is where Bloom's taxonomy (and the like) can help with decreasing the decay effect. The mental effort requirements for memorizing are considerably lower than those of levels like synthesis. In that case, factual information can be lost (specific API commands), but more important elements are retained (when/why/how to make API calls).

The sort of learning and remembering that is a part of understanding something seems even further removed from Ebbinghaus' exercises with syllables. It feels that when I am doing this sort of memorization, I am building some sort of model (which also gets remembered, or perhaps it is more accurate to say the model is the memory), and the model gets used in retrieving relevant facts. It seems much easier for me to remember an arcane fact if it led to an important or surprising insight when I was first presented with it.

> seems much easier for me to remember an arcane fact if it led to an important or surprising insight

That's where I think Bloom's can step in. Memorization is the first step, but then its about recalling and applying the mental model. As you traverse the levels, you eventually reach things like problem-solving and evaluating.

Ebbinghaus discusses spaced repetition, but I wonder if spaced repetition in conjunction with Bloom's taxonomy is a better approach for higher level learning than JUST spaced memorization.

The best resource I have ever found on the topic is this, which digs really deep in the topic:

- https://www.gwern.net/Spaced%20repetition

That, combined with Frequency Lists (another really cool topic), is what inspired me to create my own learning webapp:

- https://core.cards/

Note: it's for me to learn, so login is a bit broken and you have to try to do it twice for it to work if you want to test it out.

The human brain could work the same as memristors. Memristors gets harder to charge as they get full. Ie in the analogy that a young brain can easily learn things but it gets harder to learn things the older the person as the brain is full of information.

One also observes the same in AI that the AI learns more in the first evolutions.

Also the same in Spaced repetition such as Anki and Memrise. Anki and Memrise software must some how know of the forgetting curve. https://en.wikipedia.org/wiki/Spaced_repetition

Memristor https://en.wikipedia.org/wiki/Memristor Memristors mimic human brain https://www.eetimes.com/document.asp?doc_id=1264640

> the analogy that a young brain can easily learn things but it gets harder to learn things the older the person as the brain is full of information.

I'm not sure this is true in general. Older people have rich semantic connections that, if they can leverage them, allow them to learn things quickly. But those connections can also work against them (e.g. learning to drive on the other side of the road you get to keep 90% of what you learned, but also battle against a strong, pre-learned bias).

>An unhelpful but common metaphor is that of the brain-as-computer, or computer-as-brain. The two things are often presented as working the same way. A brain “stores” memories like files on a hard drive, and software uses “neural networks” to learn like the human mind does. But the reality of learning is different. The computer won’t forget where the file is, and the neural network can only learn what it’s told to.

So many flaws packed into one paragraph.

>brain-as-computer, or computer-as-brain

That would be two different metaphors, not one

>A brain “stores” memories like files on a hard drive

Nobody thinks this

>The computer won’t forget where the file is

Brains don't forget because they can't find the file...

>neural network can only learn what it’s told to

The power of neural networks is that they discover patterns on their own. Thats the whole f*ing point of neural networks.

">A brain “stores” memories like files on a hard drive

Nobody thinks this"

If you ask them directly and they bring it up to conscious recognition, no, approx. nobody thinks this.

However, if you witness people's actions when they set out with the deliberate goal of learning something, they clearly act as if they believe they can write something "once" to the "brain's hard drive" and retain it indefinitely by sheer force of will, and that failure to do so is a moral or character failing. Articles like this are important to correct people's implicit, unexamined beliefs.

Of course if you've been on HN for a while, you've had spaced repetition of the importance of spaced repetition (albeit perhaps not on the optimal schedule), so you may feel it's very common knowledge. It isn't. Only a few subcultures have heard of it; the broader culture is ignorant of these ideas.

I mean, is this really not common knowledge? I am having trouble coming up with examples from my real life where people aren't aware of it. Sports? Practice. Spelling? Practice. Theater? Rehearse. It's practically an automatic response when people ask how to get better.

I don't think this is a mystical life hack, we do this with babies, children, students, and I have never heard an adult seriously suggest that they can memorize something in one go.

Well, okay, maybe the president and people with photographic memories. But in general? If people decide they want to learn something they definitely are aware it will take work and practice and repetition. Maybe they don't want to do that work, but I doubt they're truly surprised when they don't make progress.

@jerf refers to "spaced repetition" which is a specific form of the "practice" you mention. Basic practice is of course ubiquitous. Spaced repetition is less common, but becoming better known.


You mention sports practice. This is a really fascinating topic, possibly different from mental recall such as spelling and theatre lines.

A beginner in a sport can practice in a very general manner, since anything they do is likely to produce improvement. Someone who has just started learning tennis may benefit from the occasional game of squash or badminton, for example. Put another way, in problem space they are very distant from the target, so any effort has a reasonable chance to take them in a direction which is vaguely towards the target.

An expert athlete, however, must practice very specifically. They are very close to the target: only movement directly toward the target is useful. The tolerance is far less, the number of directions they can take is far fewer.

Additionally, the extremely fine control of the musculoskeletal system necessary for athletes degrades slightly with absence from practice. The specific activity must be regularly practiced to achieve peak performance, and ensure the neurological specialisation is maintained. This is why professionals must "warm up", to ensure the nervous system is tuned to peak performance. (Avoiding injury is a separate and complicated issue.)

Spaced repetition is related to this, since athletes must indeed practice/train. But the time element here is very complex, and unique to the sport, the athlete, and much more. Frequency, rest, intensity, endurance and more are in that mix. Add mood, motivation etc and it becomes almost overwhelmingly complex. Possibly why some of the finest athletes aim to take emotions out of the mix: simplicity and reliability. Perhaps!

A lot of this is still poorly understood, as it is with memory and recall.

I see, I see. That makes sense, thanks.

I'm also an athlete (I teach karate) and agree with your assessment. I feel myself get worse at techniques if I don't exercise and reinforce the motions at least weekly. I generally have the class do basically the same core techniques at the beginning, then rebuild any lost reflexes in intermediate drills, and finish with no more than three new (related) techniques per day.

There is never a time when anyone no matter how skilled won't benefit from practicing even the most basic of basics once you're at a high level. They're never quite perfect! In fact the more advanced the student, the more likely they'll benefit from going back to the basics and really re-evaluating how the basics work given their new expertise.

"It's practically an automatic response when people ask how to get better."

Again I say, look to people's actions, not their words. By their actions, a lot of people show that they believe that they can be exposed to the material either once, or a very small number of times, and retain it indefinitely (along with understanding not only the material but all implications of the material, automatically), and that when the material is not retained, it is perceived as a moral or character failing. If this was not true, "cramming" wouldn't even be a word.

This is merely one of the many places where what people say they believe deviates from what they demonstrate by their actions that they believe.

Further, I think it's also pretty obvious that many of the people who do in fact study, do so very inefficiently because they do it without knowledge or understanding of the issues raised by spaced repetition. I don't think you can just apply it blindly, but at the same time, you need to be aware of the basic issues and incorporate the insights. Another one of the classic study errors, at least from the perspective of retaining the material indefinitely, rather than just for a single test, which the structure of our modern formal schooling encourages, is to study this week's material this week and ram it into my head, then study next weeks material next week and ram that into my head, and so and so forth, rigidly partitioning what I study, based on the assumption that, again, once I jam it into my brain it'll stay there indefinitely. The correct way to study for something very heavily fact-based (like a medical program) in the steady-state is to spend maybe 60% on this week's stuff, 20% on last week's stuff, and the remaining 20% scattered through the stuff that you've had the hardest time retaining from the previous weeks, not 100% on this week's topic, and not X% on this week's topic and 100-X% evenly spread through "review" on the rest of the course, though that latter one is at least close enough to work.

> A brain “stores” memories like files on a hard drive

> Nobody thinks this

Ahhh... I don't know. That was certainly a fiction that a number bosses where happy to foist on my generation back in the day. It makes for a very convenient basis to blame the junior for virtually anything.

>Brains don't forget because they can't find the file...

They sort of do. The act of remembering alters the connections to other information made. Later, when trying to recall something again, the original connections may no longer work leaving you unable to find the information.

>The power of neural networks is that they discover patterns on their own. Thats the whole f*ing point of neural networks.

They can only find patterns in what they're told to.

> They can only find patterns in what they're told to.

I disagree.

If you feed a bunch of images into a CNN, it doesn't just learn flower, cat, etc. It learns more fundamental things such as edges, textures, patterns. see: https://distill.pub/2018/building-blocks/

Moreover, you can use simple reward signals, such as "progress forward", to learn complex locomotion: https://www.youtube.com/watch?v=hx_bgoTF7bs

If you still think these represent "find[ing] patterns in what they're told to", I would argue you only think that because the scale of these experiments is so small.

>If you feed a bunch of images into a CNN, it doesn't just learn flower, cat, etc. It learns more fundamental things such as edges, textures, patterns.

Inputting a bunch of pictures is telling it to find patterns in those pictures. And though some of those patterns are fundamental things, at some point it will need to be told what patterns are useful.

That they are limmited to seeing things that they see is trivially true of all things that see.

That they are limited to seeing things that they are told to see is not.

>Brains don't forget because they can't find the file...

Isn't this what the expression "It's on the tip of my tongue" refers to? You know you have the name/word/??? somewhere in your brain, but you can't access it.

Thats a good point

I truly believe the learning/memorizing process is different for each person. For example I've been driving my car for more than a year and I still can't remember right away in which side of the car the gas tank is located at. What I do remember is that in every car, there is an arrow on the side of the gauge which indicates where the gas tank is at.

I think there are lots of factors that can make it easier to remember certain facts. Building associations with existing knowledge helps a lot especially when attaching it to something that triggers emotions. Tying spacial information also is very powerful at least when you can access the information via the spacial association. I recently learned a lot of Kanji vis SRS and tried different techniques to beat the memory curve. For example mind palaces worked incredibly well for remembering a ton of new Kanji in order. However, when they came up out of order I frequently couldn't remember them or only remember where in my mind palaces they belong, but not their meaning. A better structured mind palaces could have probably helped. I think you might be able to remember the arrow in the car because you spend so much more time behind the steering wheel than filling up the car and have a easier time putting the arrow in the context of the controls you look at frequently. It's much easier to connect to existing memory.

Oddly, you may have trouble remembering because you know where to find the answer. Simply put, you remember a more general answer to the same question. It is not that you don't remember at all.

That has scary implications now that Google exists.

Somewhat. And it isn't just Google. Now that you have a cell phone, you are probably not as good at remembering phone numbers as folks were 20 years ago. I know I used to remember some numbers. Now, I barely remember my own.

So, same goes for how can you tell if it will rain soon? Well, used to folks knew some tricks by looking at the sky. Some folks still know them. By and large, everyone else is more accurate by asking their devices.

Want to know how far away something is? There is an amazing trick you can do with your thumb and line of site. If you really want to know, though, get a range finder. :)

And it isn't just memory. Consider how much weaker most people's arms are today than they used to be.

You're ahead of most people then, because I would say the vast majority have no idea that arrow even exists(sometimes it doesn't).

My old 1978 Chrysler Cordoba has it behind the license plate in the back, so you can fill up from either side, and so it doesn't interfere with the lines of the car. I think they stopped doing it because rear-end collisions are so common though :(

I solved that problem by refusing to buy cars where the gas tank is on the other side of the car from me.

You have probably never noticed but most modern cars have a small arrow by the gas dial that points to the side of the car where you fill up.

The parent comment says just that

That is probably because you no longer have to remember that, so you don't.

Same thing when I am driving with someone giving out the directions versus trying to figure out the way myself. If I was given the directions, I'll probably need them again on the way back. If I figured out the path myself I'll be more likely to find my way back.

Yes the arrow is very useful when you are driving a rental.

Talking of rentals it always seems to take me ages how to figure out how to adjust the seat.

> forgetfulness doesn’t always occur (Princess Leia’s prison cell block number: AA-23)

Sure, when it's the start of the Fibonacci sequence.

Thanks, now I'll never be able to unsee or forget that. But you do make a good point about the importance of linking/patterns in memory.

And naming them.

I'm surprised this didn't hit on relational memory. In the latest edition of stupid human tricks, people are often surprised when strong chess players, or myself, can easily repeat the moves of a game, from scratch, after playing - even substantial amounts of time later.

It's all about the relational aspect. Chess games are easy because they tell a story in a strong player's head, and so to repeat the game you just tell the story in your head again. That's why if you do forget a game, just being told what the first move or two was is often enough to recall the rest of the game - kind of like being told the first sound of a word you're trying to remember.

This is also why chess players can see a position and remember 'Ooo, that's from Reti-Alekhine' or whatever. You see games and the people playing become part of the story. It's like reading a page from a book and being able to say what book it was. On top of that it's also quite often an extremely important page, making such a demonstration even more trivial.

There have also been a lot of studies testing this stuff. For instance, one briefly showed strong players a position that might actually occur in a regular game and asked them to recreate the position. And they did extremely well, whereas casual players did awfully. However, the interesting thing is that when asked to do this for positions with the pieces on random squares, strong players did not do substantially better than casual players. This has been attributed to pattern recognition, but in my opinion it's more about the story being told even on a single move - the ideas and possibilities. Again the difference between trying to repeat the words from a page of book as opposed to trying to repeat random words.

Our memory is a cache and each timed repetition increasingly delays the purge date of a particular memory.

This seems like it could be true for memories of facts, or something you can recite, like lyrics of a song.

However for other kinds of memories, such as remembering a situation or a picture, I have heard (I'd appreciate a reference) that the recall process of long-term memory is slightly destructive, and that the memory must be re-encoded into long-term memory. That suggests that recalling a memory repeatedly could accelerate it's deterioration.

I've heard something similar, but I think it had to do with the fact that we tend to "fill in the gaps" if we can't remember something about a situation. So we just make up information. I'm not sure if there are ways to combat this by recalling the information before it starts to decay or not, but it has a lot of implications for eyewitness testimonies.

My suspicion is that we create new associations to the old memory. For distinctly visual memories ... I'm not sure what the mechanism is, but at least for myself, I often don't recall specifically what something (or someone) looks like, until I see them again. The act of recall-without-seeing may overwrite the actual recollection with what you think it should be.


I don't know about the brain acting like a cache, but synaptic pathway weighting is thought to be modified by how often neurons on either side of it fire near the same time. This is called spike timing dependent plasticity (STDP). The seminal paper for this is: http://www.jneurosci.org/content/18/24/10464.long

It's an analogy for the reinforcing characteristic of neuronal connections, would be perfectly adequate if not for the continuum vs discrete difference in the reinforcement mechanism.

I was just saying that curves of how much is forgotten seem to look similar to a caching mechanism.

I worked on an app based on this timeline like flashcards, it reset when you get a card wrong. Https://www.studysauce.com

What's the difference between this app and something like Anki or Tinycards?

I've been looking for an (optionally OCR enabled) spaced repetition software package that allows me to take pictures of flashcards, and then load those into a standard spaced repetition algorithm, then render those either in a browser or an ios app. Anki supposedly does this but in my experience it's been hard to get up and running. Maybe something that just works with something like evernote or onenote would do the trick.

How is your app different than the huge number of SRS-based flashcard apps that are already out there? Particularly the grand-daddy of many of them, SuperMemo which based its SRS scheduled system off of Ebbinghaus.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact