Vansh Kumar | How can we better teach metacognition?

A potentially better framing:

How might we help people acquire tacit metacognitive know-how to let them generate better internal experience from the same external input?

Requires some definitions:

Internal experience: What goes on inside you when you do something. When you see a chessboard, what do you notice? How do you group pieces? When you read some text, do you ask yourself questions about the content & try to answer them? When a see a landscape, what do you feel in your body? The majority of this is tacit (not communicable in words) & may be personal/idiosyncratic. It is often what distinguishes experts from novices.

Metacognitive know-how: Strategies you use to steer that internal experience. Typically acquired through experience (solving many hard technical problems, solving many chess puzzles, reading & writing a lot, etc). This is also often tacit knowledge.

The best article I have read so far on transmitting tacit knowledge: https://nintil.com/scaling-tacit-knowledge/

I am increasingly starting to believe that Metacognition may be the most important skill.

In my own experience, despite attending a top school, I did not think explicitly about my metacognitive habits until just a year and a half ago, when I was 30! I had never even tried to improve them. Whatever practices I had, I had learned implicitly while growing up through school & peers.

From this article:

There is ample evidence that key aspects of cognitive inequality have already crystallized by the time children acquire the language capability to make sense of metacognitive guidance. Whether it’s the genetic lottery, the socioeconomic lottery, or the idiosyncratic neurodevelopment lottery, the conclusion is basically the same: you cannot train your kid to become the next Einstein.

There is, however, a crucial nuance. Mental habits and metacognitive approaches cannot be directly transferred from one brain to another, but it is possible to discuss them—or at least try to—and everyone has a meaningful margin for improvement.

I suspect my practices were fine, as I did well enough, but it makes me a little sad that I wasn’t even exposed to this as an axis of improvement. It’s funny that people say “oh school is where you learn how to learn.” While that is clearly true for good schools, it’s annoying that there is not even an attempt to explicitly teach it! Instead, it’s done implicitly and often through the process of selection (“weeder” classes) rather than a more inclusive process.

Just yesterday (June 2026) I met someone who told me they were almost kicked out of their first year of college because they didn’t take it seriously. That summer, they happened to read How to Become a Straight-A Student by Cal Newport, which apparently explains strategies like time-blocking, active recall, etc. They didn’t become some insanely exceptional student, but they graduated & even passed a difficult “weeder” biology class. Perhaps more importantly, it was obvious from talking to them that now even as an adult they knew they could learn.

How many people are there that would see themselves as competent adult learners if they were exposed to metacognition as an axis they could improve on? How can we teach children this skill, ideally nonverbally but also through explicit instruction?

Any instruction on this skill, just like for any supposedly transferrable skill like “critical thinking,” needs to happen in context – while students are working in/learning about some domain. This is a difficult problem as it is about the transfer of tacit knowledge. Brainstorming follows.

Implicit/nonverbal instruction

As with many things education, implicit nonverbal transfer is the most effective but harder to do. Given how personal metacognitive processes seem to be.. the only way to teach them may be to provide a scaffolded environment that allows individuals to develop their idiosyncratic methods. Explicit instruction may only be useful in so far that it demonstrates that metacognition is something you can do, something that is possible. That has largely been my experience – I didn’t even realize asking questions and trying to answer them while reading was a move I should be making until exposure to Andy Matuschak’s work.

An effective example: Montessori teaches proto-metacognition.

Apprenticeships are tailor made for this. This is probably the best approach for older children & adults, especially as metacognitive skills are both domain-specific & can only be taught in context – a description of cognitive apprenticeships.

Explicit instruction

It’s extremely edifying to hear an expert speak aloud what’s going through their mind as they do something

A great example by Andy Matuschak, especially of the main idea of How To Read A Book – asking questions as you read and trying to answer them yourself
Another great example from Andy Matuschak writing his Evergreen Notes

Frankly just being told about the benefits of spaced repetition & that Anki exists would have been very helpful.

I suspect I would benefit from watching someone I intellectually respect do their weekly/monthly/quarterly/annual review process. I do these myself, but only inconsistently & get mixed benefit from them.

so my main questions:

for any particular domain, is there a faster way to get to high quality secondary stimuli/tacit knowledge? is it generalizable?
- yes to the first – domain specific, but things like videos, apprenticeships, etc can clearly help
- kind of to the second – metacognition to the extent it’s a “transferrable” skill would be one way. i mean it clearly can help to have high quality secondary stimuli in one domain, especially if domains are similar (chess<>math)
  - are there domains that are most upstream? probably.. reading, writing, math, science
is there a faster way to learn metacognition?
- as an adult, think so, with a lot of it being conceptually fairly clear & requiring mostly discipline if you know to even ask the q
- as a child… maybe?
  - for early childhood, imo montessori teaches this
  - for older kids, i’m not aware of a specific approach. i would think a similar approach as w adults would work? prolly more of a focus on apprenticeships & generally

I wonder if/how tech & LLMs can help? One interesting proposal, though it’s only focused on adult learners. LLMs could control 3D printers to quickly create a material that isolates & provides immediate feedback for some concept. Can definitely create apps that kids can use to isolate some concept (young kids using messaging on ipad comes to mind, mentava, mathacademy)

An interesting other angle: using tech to provide feedback for caregivers. Videos of young children that point out when the child initiates, withdraws, gets overloaded, asks for help, or tries again. I suspect high overlap between this and what Montessori trained teachers learn in their observation practice of children. Maybe a first version of this looks like a Montessori observation training app? Could you train a model to detect these signs in children?

Some examples:

From here:

Piaget was not an educator and never enunciated rules about how to intervene in such situations. But his work strongly suggests that the automatic reaction of putting the child right may well be abusive. Practicing the art of making theories may be more valuable for children than achieving meteorological orthodoxy. And if their theories are always greeted by “nice try, but this is how it really is…” they might give up after a while on making theories. As Piaget put it: “Children have real understanding only of that which they invent themselves, and each time that we try to teach them something too quickly, we keep them from reinventing it themselves.”

A key thing with all of this is that you really do not want to suppress the theory-building & knowledge-constructing instinct that kids have. Rather you want to allow them to construct their own knowledge, build their own theories, and provide ways to test them against the real world.

Why are videos so good at helping learn tacit knowledge?

Tend to have running commentary, which is at least an attempt at making tacit knowledge explicit
It’s more input data – you can see more clearly what experts are paying attention to as they do some task they’re experts at
- A decent chunk of expertise seems to be knowing the right questions to ask, noticing important things, etc. What kinds of things to pay attention to.
Input/memorization of a large number of diverse examples naturally allows you to learn the underlying patterns (a la DNNs)
The extreme case is fully simulating the experience of being an expert doing a task, and doing that across many experts. Your brain will naturally generalize across them.

Could serious close observation of a coach/expert, eg in tennis, followed by high levels of awareness of yourself, be enough to learn technique etc? Zero words required. This is the Inner Game approach.

From here:

Read from the source material of master practitioners, copy their actions, climb their skill trees, and reflect through trial and error. Don’t read third-party accounts of technê.

To read: https://commoncog.com/the-tacit-knowledge-series/, https://commoncog.com/accelerated-expertise/, https://www.leiderschapsdomeinen.nl/wp-content/uploads/2016/12/Zimmerman-B.-2002-Becoming-Self-Regulated-Learner.pdf, https://www.aft.org/ae/winter1991/collins_brown_holum, https://notion.moontowermeta.com/principles-of-learning-fast, https://nintil.com/bloom-sigma, https://andymatuschak.org/hmwl/, https://x.com/ejames_c/status/2076105059855970725, https://jtpeterson.substack.com/p/faq-on-expertise