Robin Hanson has repeatedly pointed out the difference between near and far modes of thinking, and Alicorn has repeatedly pointed out the difference between procedural and propositional knowledge. Just occurred to me that it's pretty much the same difference. I will now proceed to play some obvious riffs on the theme.

Do you ever feel yourself to be an expert on some topic just from reading about it a lot? Every young enthusiastic programmer just entering the workforce feels like that. (I certainly did.) Every wannabe entrepreneur who hangs out on Hacker News and has read enough Paul Graham essays feels they can take on the world by just applying those valuable insights. (I certainly did.) It's soooo fun to feel knowledgeable, maybe even project it outward by giving Internet advice to newbies. The public relations department of my brain doesn't seem to care that I have no actual experience: reading stuff is quite enough to change my self-image.

The problem is, excessively liking propositional knowledge over procedural is a bias that harms us every day. Though some information is directly useful, most of it is worthless. A couple days ago I had to give advice to a classmate of mine who wants to start his own "thing" but isn't sure. See, he has this theory that one should accumulate propositional knowledge until one reaches critical mass, at which point the successful venture happens by itself. Being the wise and experienced mentor that I am (hah... on my second "thing", without much success), I told him outright that his theory was bullshit. Propositional knowledge doesn't spontaneously turn into procedural.

(Digression: come to think, I'm not even sure why we need the kinds of propositional knowledge that we tend to accumulate. It reeks of a superstimulus. I know more about programming that I'll ever need for work or play, but don't remember the birthdays of all my acquaintances, which would obviously be more useful. Memorizing birthdays just isn't as exciting as reading about comonads or whatever.)

At this point I sincerely wish I had a recipe for overcoming this bias. Like pjeby, he always has a recipe. Well, I don't. Maybe perceiving the bias will turn out to be enough; maybe some kind of social software thing can help cure it on a mass scale, like is trying to cure Bowling Alone; or maybe each of us will have to apply force, the oldskool way. It's too early to say.

Thank me for never mentioning the example of riding a bicycle.

New Comment
34 comments, sorted by Click to highlight new comments since:

Propositional knowledge does not make anything happen by itself, but it can help when actually doing something. For example, the venture won't happen by itself, but once it is started, it might be that certain things happen, and you go "hmm, this looks like something I read about". This allows you to make predictions and act upon them, potentially avoiding a pitfall others have already trod in.

To take programming, once you start programming something, you might go "wait, this looks exactly like that theory, and there was a solution", or "wait, here I also need to guard against this problem" --- both things you have not experienced yet. Learning by doing can be quite expensive.

An overlooked factor here is risk.

We enjoy learning, yes, but a lot of opportunities to learn "procedural" skills are quite risky. Starting a company has big financial risks, reading about starting a company doesn't. Video games are mostly about procedural skills, but aren't risky, so they're very attractive too.

A way to counter this would be to recognize that quite a few of those risks are overblown. The risk of quitting your job to start a company is real, how about the fear that girls will reject you if you approach them, or that people will laugh at you if you don't speak their language correctly ? Are those really such a big deal ?

Good point about risk, definitely something to think about. And thanks for mentioning the "language barrier" - it's a very good example. I've broken it twice - with English and Italian - and both times weren't too easy, and with French I haven't broken it yet, even though I can parse French written text easier than Italian. (My native language is Russian.)

I'm not sure this is really a bias. Maybe it's just that the cost of acquiring propositional knowledge has decreased more relative to the cost of acquiring procedural knowledge, in which case it's quite rational to accumulate more propositional knowledge. Emile made a similar point.

Also, I have an anecdotal example of propositional knowledge clearly beating procedural knowledge. I used to have shoulder pain from using computers too much. I visited a doctor and a physical therapist, and neither of them were able to help me. I ended up solving the problem myself after reading books about anatomy and physiology, and then doing Internet searches, which found a crucial piece of propositional knowledge that the doctor and therapist apparently weren't aware of.

About your digression, personally I've found that most of the propositional knowledge I've learned for fun has ended up being useful. Maybe that's just a coincidence, but there are also things I think would be useful for me to learn that aren't fun for me, so I think it's probably just that the value of propositional knowledge is pretty high in general.

BTW, if you have a lot more knowledge about programming than you need for your current work, why not look for a more challenging job or project?

BTW, if you have a lot more knowledge about programming than you need for your current work, why not look for a more challenging job or project?

My projects are challenging in other ways, programming just isn't the limiting factor any more.

Ok, that makes sense. Also, one problem with knowing a lot about programming is that you can easily write code that nobody else can understand or maintain.

What kind of projects are you working on? Do you have a personal homepage that you can share? (And why doesn't this blog let users put their homepage links on their user pages?)

I'm the sole frontend developer for and , as well as other stuff that isn't so well-known.

Tangent: you should see about optimizing for loading speed some more. I assume you're familiar with Yslow for firebug?

Yes, I am. Maybe we should do that. But we aren't global, and it's fast enough for Russia :-)

ETA: does OpenPhotoVR load fast enough for you? I.e., is it a Russia problem or a hosting problem?

OpenPhotoVR doesn't seem as resource-intensive, though I haven't played with it much. In general, I haven't noticed any Russia problems.

Thanks! Will investigate with the hosting provider tomorrow.


Propositional knowledge doesn't spontaneously turn into procedural, but it does help by directing you towards the more useful actions to take to acquire that knowledge.

For example, if I want to become a writer, simply reading a whole bunch of books on writing won't turn me into one, but they can direct me to the most useful ways of getting there (such as writing every single day and not worrying about the end result.) and help me avoid the ones that are counterproductive (such as waiting for inspiration to strike.)

I find as soon as I actually try to do something that I've read alot about, it becomes painfully obvious I'm not an expert at it. Perhaps the answer is just doing more things.

I agree with you, it's a superstimulus. Nowadays information is cheap and abundant(Internet) and since we are wired to seek information we are constantly chasing after the next bit. It's a coincidence, but before reading this article I was thinking about how the internet was supposed to make life easier since it makes all this information available for everyone. But the contrary has become true: now people waste countless hours chasing after meaningless(as in not applied in practice) stuff. The solution? Purposeful disconnect yourself and choose to practice whatever skills you want to improve. This is also advocated in some books, Tim Ferriss in his "4 hour work week" advises on minimizing email time etc... I know, I know, it's easier said than done.

I've read Tim Ferriss' 4-Hour Work Week, and his proposal of a low-information diet made me think. I cut down my newspaper-reading to zero. Not even online. And I am very grateful that I did — I don't miss it at all.

...and then I've run across this quote from Thomas Jefferson: "I do not take a single newspaper, nor read one a month, and I feel myself infinitely the happier for it."

I laughed out loud. The idea wasn't so new.

Then I leveraged the time I earned by overindulging in Lesswrong, Overcomingbias, Wikipedia and other precious sources. And reading more books.

Still propositional knowledge, but clearly more enlightening than the daily fix of murders and gossips.

There's a good point: part of the general issue is whether the information we're acquiring is relevant. Feedback from doing (whether procedural or propositional) is probably more relevant to the task you're trying to accomplish than information gleaned from a broad search, like reading newspapers, etc--and experience can greatly help to establish just how important the information is.

And information gained through feedback is probably also actionable unlike a lot of the stuff we read.

The problem is, excessively liking propositional knowledge over procedural is a bias that harms us every day. Though some information is directly useful, most of it is worthless.

Worthless to whom?

Passionate truth-seekers find enjoyment in their growing ability to see things as they are, in making the world make sense. It's an autotelic activity for them. If the value sought is simply to understand a phenomenon, it's not worthless.

Some people like watching TV; some people like reading about comonads. If you don't starve to death because of it, and if there is nothing else you'd rather be doing than the task in question, fine! Go for it.

Harm to whom?

The harm seems to appear when you have an objective goal, project or task which is important from the perspective of your utility function, and then you jeopardize deadline or quality by avoiding to follow the required steps to complete it.

You don't work objectively from the perspective of the end-product, consuming theory as necessary; instead, you diverge aimlessly. It may be a sign of risk-aversion and need for control — you can never screw up too much by reading books.

In any case, harm should be viewed in the light of your utility function.

fully agree with this.

Sounds like main problem in the examples you give is overconfidence about our ability to transform propositional knowledge into procedural knowledge.

Overweighting the propositional knowledge we have beforehand can cause us to discount ignore important information that we might pick up by doing--it can inhibit our ability to be empirical through the mechanisms of familiar biases. Planning fallacies, and problems of being unable to respond to unforseen circumstances--or recognize opportunity!--can come soon after.

Seems similar to the distinction Taleb makes between 'Practitioners' and 'Theorists', actually. I tend to be of the Not yet! I don't know enough! group, so this summer I've been trying to find ways to encourage myself to be more of a practitioner than a theorist in certain domains of life.

My problem was more an emotional need to know as much as I can about the task and the context of the task before executing, or even practicing. I think this is a "nerd" tendency. Even excluding the epistemological pitfalls there is an opportunity cost for putting something off until later that needs to be part of the calculation.

I agree that in far mode we are more likely to accept propositions without sufficient evidence, and that much of our effective learning of procedures happens in near mode. Nevertheless, we can and do create valid propositions in far mode, and we do sometimes learn far mode procedures.

Fair enough, but can you give some examples of procedures that people successfully learn in far mode?

There are two issues that you have to separate. 1) Thinking that you are better than you are because you have read a lot. 2) Choicing to spend your time on reading.

You could rephrase 1) as being more optimistic than it's waranted. Sometimes that's damaging but at other time it might help you to get going.

1) isn't that big of a problem when you consider other alternative ways of spending the same amount of time like watching television. If you however don't spend enough time on actually doing things the problem happens to be procrastination and you just intellectuallise it as a different problem.

Knowing a lot about programming also gives you the benefit of better being able to mentally take the identity of being a progammer. Humans have the need to feel identity.

Perhaps you should have used the example of the bicycle. Remove any requirement for athletic skill, and the practical distinction between the propositional and procedural fades considerably.

For example, as a kid, I was able to ride a bicycle the first time I tried. Of course, it had training wheels.

Another example. As a young man, I was able to pilot (taxi out, take off, and fly) a small aircraft the first time I tried. (This was 1974, before the common use of simulators.) I had informed my instructor that although I had never even been in a small aircraft before, I had an engineering science education and had already studied the relationships of the basic instrumentation (especially the airspeed indicator) to a wing's angle-of-attack. After a certain amount of grilling, he let me go for it and he never had to touch the controls.

My point is that I can be successful without ever having to admit propositional knowledge is actually true. I just have to be able to take advantage of its utility in predicting outcomes. I consider this the distinction between engineering and science.

At this point I sincerely wish I had a recipe for overcoming this bias. Like pjeby, he always has a recipe.

No recipe for this one, sorry! (If you find one, let me know.)

Speculation within near-far dualism: Far mode is abstract.\ > Meta-level thought is abstract. > Meta-level thinking necessarily takes place in far mode > error correction is meta-level > Error correction takes place in far mode > far mode may be less reliable than near by default, but the most reliable thoughts must take place in far mode.

Error correction takes place in far mode

What? I'm pretty sure this is wrong. Errors get corrected by experience.

Language is an example of an area where propositional knowledge is pretty important. Sure, you'll progress faster by talking to people than by reading books, but memorizing vocabulary lists and grammar tables does pay off.

On the other hand, I'm not that sure that reading a foreign language textbook really is only about propositional knowledge - deciphering a sentence reauires more work than just reading one, and is more like "real" training.

Actually, you will have more fun talking, but I think you will learn more vocab in an hour by hitting the books and you will DEFINITELY do better with books if you have decent pronunciation and grammar but your goal is to go from conversational competence to actual fluency with a near-native vocabulary.


Propositional knowledge doesn't spontaneously turn into procedural knowledge, but it can enable the acquisition of procedural knowledge with no other ingredient than the will to become proficient. I am self-taught in Bayesian data analysis, and I like to think that I've reached journeyman status. If I have, it is purely through reading and applying what I have read.

For many subjects it's true, though even there the will is typically the limiting ingredient. (Also, will typically causes you to acquire knowledge, while knowledge doesn't typically cause you to acquire will.) But for many other subjects it feels false to me: whenever I start actually doing something, I'm invariably surprised by how little I'm helped by pre-acquired propositional knowledge. Not sure how to quantify that.

This video did a lot to help me distinguish near from far. I also like the subtle reference to how Grover's algorithm works.

ETA: I'm just being silly. The link is to the Sesame Street sketch where Grover explains (physical) near and far. But it's interesting to note that Scott Aaronson mentioned one time in a discussion on his blog (I'll find and link if necessary) that the sketch also accurately represents how Grover's (no relation) algorithm works.

Just have a good laugh, folks :-)

The video is "Sesame Street - Near and Far with Grover".

You should really mention little things like that.

Further such ruminations in much the same spirit :-)

I would be grateful for any accessible illustration of how Grover's algorithm works.