Discussion about this post

User's avatar
Amanda Pustilnik's avatar

Love your work, Adam, but have to take issue with 2 things here - one big & one small (I’m not an academic psychologist; I’m an academic in another field, so I’m not defending my turf).

1. Solving psychiatric illness: Real, important, replicable progress is being made every year. Tens, maybe hundreds, of thousands of people just in the US can live independent and meaningful lives because they take psychiatric medication for serious psychiatric illnesses. This includes major and treatment-resistant depression, treated with TMS or neurostimulators. I recognize that you are writing about psychology and not psychiatry, but: (a) you said that nothing in the DMS has been “cured”; yet people with serious DMS diagnoses *can* be fully treated; and (b) making the argument that psychology is no better now than it was 50 years ago leads to treatment nihilism, which can seriously harm people who then miss out on effective treatments.

2. “Big 5.” It may be that the big five are no more predictive than hocus pocus, but you can’t tell that by looking at whether somebody buys a house in their lifetime. It’s the wrong endpoint. I could predict, on an average basis, who will or won’t buy a house in their lifetime and it doesn’t have anything to do with psychology. I would ask: 1. Are they White? 2. Have they received an inheritance or other chunk of change from a family member? Being white and receiving intergenerational wealth (two variables that are also correlated) are the strongest predictors of home ownership. We should not expect psychology to predict unlike types of social facts.

Thanks again for the great article. Please keep up the great work!

Expand full comment
Soren Bjornstad's avatar

I was interested to see you list Anki as a success at “cramming it into a box,” because I’ve spent about ten years developing, supporting, and writing about spaced-repetition systems in some capacity (including Anki), and I’ve often written about how using memory tools well is actually very complicated, noting that if only we could find a prescriptive set of simple rules you could follow (like you could encode into software) that would consistently work, it could finally take off!

Anki does do a good job at packaging up the insight of spaced repetition itself. Unfortunately, that is only a tiny part of the problem of learning/remembering things, and solving this restricted problem leaves most of the real problem unsolved. Just a few common and important problems it doesn't solve:

* Lack of motivation / habit – people forget to study, don't want to study, don't study for a few days and then come back and have a huge pile of cards and give up, etc. (This is moderately tractable – I think RemNote, the startup I’m currently involved in, does much better than Anki does here – but habits are just hard, and there’s only so much you can do to help people remember things if they don’t review them!)

* Learning something that's useful / you care about knowing – it's super common for people to use Anki to memorize, say, all the symbols on the periodic table, and then if you ask them why they wanted to know that, they have no idea.

* Writing cards that are unclear – oftentimes when you write a card (or write one for someone else to import), you think you know what it’s asking, but when you actually see it out of context 3 months later it is inscrutable, or makes you think you were supposed to give some other answer instead. In the best case, this is frustrating and messes up the spacing algorithm because you can’t accurately evaluate how well you remembered the content once you fix it. More commonly, people who don’t have a lot of experience don’t even notice why they didn’t know the answer, and keep forgetting the card over and over again without ever doing anything to fix it.

* Learning the thing the way you need to remember it – it’s really easy to memorize a card based on, say, some unusual wording on the front of it, and find yourself unable to recall the information when given a real-life cue. Or you might just ask the question in a way that doesn’t match how you’ll be asked to recall it, and find you just can’t quite make the connection when you are. (Memory is much more sensitive to tiny shifts in context than most people realize.)

* Keeping cards relevant / understanding why you’re struggling – it’s common for cards to become stale because you no longer care about the information on them, or you don’t understand what they were supposed to mean, so you struggle to remember them, and you just keep repeating them over and over again until you get frustrated.

Maybe the core issue is that the apparently simple problem of "remembering question/answer pairs" is deeply tangled up with all sorts of other complicated human stuff that prevents you from actually effectively using the tool unless you also solve those other problems (or at least partially address them). I appreciate the claim that the “people are just complex” argument is unimaginative, but I suspect this may be where psychology really does diverge from, say, physics. Sure you're ignoring all kinds of factors when you, say, model the trajectory of a ball you throw. But with just a couple of equations, you still get answers that are accurate enough to do amazing things to a pretty decent degree of accuracy – you can really forget about a lot of the adjacent questions (e.g., relativity, or often even air resistance). Whereas if you just hand someone Anki, it’s still better than them not having it, but the vast majority of people get resoundingly mediocre results, because of all these adjacent issues, and there's not an obvious way to cram solutions to those issues into the box, because they deeply depend on parts of your internal mental state that can't be reliably introspected. This is frustrating, so most people give up, even though the spaced-repetition tool itself is extremely powerful.

Now I guess you could argue something similar about the ICE. Like of course having a great ICE doesn’t mean people will become good drivers, that's a separate problem to solve, and that shouldn’t diminish the beauty of putting combustion in a box. But there’s some way in which dismissing it like this feels unsatisfying to me – with the ICE that seems so obvious as to be absurd, while with spaced repetition it really feels to me like the fundamental problem is basically unsolved, and we’ve only handled some tiny restricted subproblem. Is it just that we think about psychology differently? Or that the path to making cars safer is clearer than the path to understanding what cards will change someone's mental state into being a person they prefer to be? I’m not sure.

(I have a bunch of blog posts about addressing some of these issues with memory tools – by developing personal skills, not by improving the tools – for anyone curious. You can start at https://controlaltbackspace.org/precise/.)

Expand full comment
81 more comments...

No posts