Reductionism Revisited
post by alkjash · 2018-03-20T06:00:01.379Z · LW · GW · 6 commentsContents
Day 28: Reductionism Revisited Mysterious Answers: A Brief Review Build Form by Cleaning Your Room Incremental Progress Reductionism vs. Procrastination Continuous Grading Daily Challenge None 6 comments
This is part 28 of 30 in the Hammertime Sequence. Click here for the intro.
The last three days of Hammertime, I’ll wrap up with some scattered thoughts to reinforce important principles.
Today, I’ll return to applications of reductionism to instrumental rationality.
Day 28: Reductionism Revisited
Mysterious Answers: A Brief Review
I had a conversation with a friend in which the topic of comedy popped up briefly. I will strawman his argument to make a point:
Friend: Well, there’s no step-by-step training drill to make someone funny. When I imagine a comedy coach, they probably just ask you to tell jokes and rate how funny they are.
Me: If you didn’t know math, would you say the same thing about studying math? That there’s no step-by-step approach to teaching induction. Instead, a math teacher just has to let the student try to prove things and rate how rigorous each proof is?
Friend: Point taken.
Irreducible and mysterious complexity, as we know, is a property of the map and not of the territory. It’s an easy cognitive mistake to make to believe that many skills, especially the ones you’re ignorant of, can’t be broken down with reductionism and must instead be learned organically and intuitively.
I think this is a symptom of a general cognitive error that can only be cured by rereading Mysterious Answers half a dozen times. It’s important enough to highlight again. The error goes like this:
In my subjective experience, my domain of expertise is concrete and gears-like, amenable to reductionism. I have a detailed mental model of how to go about solving a math problem or writing a blog post, step by atomic step. In my subjective experience, skills I don’t have are fuzzy, mysterious, and magical. Training them requires intuition, creativity, and spontaneity. From these defects in the map, I then incorrectly deduce that mysteriousness is an actual property of the territory beyond my competence, i.e. outside my comfort zone.
Mysteriousness is in the mind. Go forth with a Zeno‘s conviction that all of the territory breaks into infinitesimal pieces, each of which you can individually chew.
Build Form by Cleaning Your Room
One of the most important things to encourage in the early stages of a new skill is the development of good form. Once you have it, trying harder works, whereas if you don’t have it, trying harder just leads to a lot of frustration and discouragement. And of course, if you have bad habits right from the start, they’ll only going to get harder and harder to fix as you ingrain them through practice.
~ CFAR handbook.
One of the features of a reductionist approach towards instrumental rationality is this: hard problems break into small pieces. Small pieces are easy problems. Therefore, you can get better at solving hard problems by training your cognitive strategies on much easier problems.
True mastery starts with practicing cognitive habits to perfection on exceptionally simple tasks.
Counter-intuitive as this principle may seem, we already know it to be true. We know that students can’t move on to algebra until they have perfectly memorized their times tables. We know that before you practice writing you need to master typing or handwriting. In fantasy literature, this idea is ever-present: the novice must spend years levitating a pebble or kindling a flame with perfect control before he moves on to more ambitious magiks.
CFAR calls this principle building form, in the sense of physical exercise. (I’ve been told that) in the weight room, correct lifting form leads to better safety and muscle growth. Learning how to place your feet, tighten your glutes, and arch your back correctly are all important fundamentals to get down before you start benching several plates. All of these fundamentals are best trained on much lighter weights than your current maximum.
Jordan Peterson calls this principle cleaning your room. Start by solving the problems in your immediate domain of competence like dust bunnies and unwashed clothes (that reminds me … be right back). It you can’t the alignment problem of getting yourself to sleep and wake up on time, expect to hurt yourself trying to save the world.
Also, like the novice’s exercise of levitating the pebble, building form is not as simple as it appears. A friend of mine had plans to drop out and apply to work on AI at DeepMind. I told him to fix his sleep schedule first. Two months later, after numerous strategic meetings, he’s still working on this problem. At least he’s finally recognized its difficulty.
Incremental Progress
Reductionism vs. Procrastination
If you have a procrastination problem, here’s a simple cognitive shift based on reductionism that helps. It’s a variation on the only piece of “classic self-help advice” I ever found useful. Every time you catch yourself delaying a task to a future date, ask yourself the question:
How much of this task am I willing to do right now?
Answer it honestly. Then, do that much.
Maybe instead of getting exercise, all you’re willing to do is go outside for a minute. Maybe instead of filing your taxes, all you’re willing to do is organize the relevant forms in a folder. Maybe instead of writing that paper, you can at least tolerate typing in the title and section headers.
The discerning reader will notice this script is essentially a TAP to apply a microscopic CoZE experiment at every task aversion. That’s exactly right.
Continuous Grading
Despite how disappointing the project was, I really liked Duncan’s Dragon Army Retrospective. One of the tangential reasons for this liking is his use of grades instead of a cruder pass/fail system. Grades imply a smooth, continuous success function which is much easier to optimize.
Human beings are not built to make Fails turn into Passes. Human beings are built to make numbers go up [citation needed].
Score yourself continuously, and you’ll have an easier time measuring incremental progress and mentally rewarding yourself for it. Score yourself not for whether or not you did something, but for how much you did and how well you did it.
Daily Challenge
Just now I described a microscopic version of CoZE to apply at the five-second level. How many of the other Hammertime Techniques can you build TAPs to apply minified versions of?
6 comments
Comments sorted by top scores.
comment by Qiaochu_Yuan · 2018-03-20T17:18:15.650Z · LW(p) · GW(p)
In my subjective experience, my domain of expertise is concrete and gears-like, amenable to reductionism. I have a detailed mental model of how to go about solving a math problem or writing a blog post, step by atomic step. In my subjective experience, skills I don’t have are fuzzy, mysterious, and magical. Training them requires intuition, creativity, and spontaneity.
The way this used to feel to me is that I would think of my own skills as "basically just thinking really hard," whereas when I saw other people with skills that involved interacting more directly with the world (say, starting companies, or building robots) I would accord them really high status, so they had this shiny "wow, I could never interact with the world like that" quality that made them seem extra ineffable. This actually tied back into an aspect of my Giant Bug around fear of being judged for interacting with the world poorly, or something like that.
The idea that intelligence wasn’t monolithic, that I could be much worse than them in proofs and theorems but still be their equal in other areas, was hugely liberating to me, but it took me a very long time to accept it, to believe that I really was as valuable a human being as they were.
And when I tried to analyzed my certainty that – even despite the whole multiple intelligences thing – I couldn’t possibly be as good as them, it boiled down to something like this: they were talented at hard things, but I was only talented at easy things.
It took me about ten years to figure out the flaw in this argument, by the way.
I was doing this constantly - when I read SPARC applications I would be really impressed (in retrospect, clearly overimpressed) by students who did things like start companies, in the total absence of any kind of gears model for how to start a company.
One of the things that started shifting my views around this was a conversation I had with a robotics startup that was trying to sound me out for recruitment. They saw a Facebook status I had written about RL that they liked, and I asked them why that was enough to get them interested in me considering my relative lack of actual experience in robotics, and they said (paraphrased) "we're bottlenecked on people who can think well. It's easy to find people who can build things well." And I began to see the way in which I had been both undervaluing "basically just thinking really hard" and underestimating my own ability to pick up these other skills that seemed ineffable to me.
comment by steven0461 · 2018-03-21T16:47:03.018Z · LW(p) · GW(p)
It you can’t the alignment problem of getting yourself to sleep and wake up on time, expect to hurt yourself trying to save the world.
I disagree with this example. Bad wake and sleep times are often a physiological problem. Such problems, like any other problems, can sometimes be solved with competence and good decision making, but this post suggests sleeping at the right times is just a matter of playing the good decision making game on easy mode, and that definitely hasn't been my experience. (I've gone from a terrible sleep cycle to a great sleep cycle through strategies such as aging.)
Replies from: alkjash↑ comment by alkjash · 2018-03-21T18:36:25.915Z · LW(p) · GW(p)
Certainly sleep issues can be a result of deeper physiological problems, but I'm doubtful as to whether this is often the case. The most likely cause is "lack of conscientiousness and beliefs that sleep times matter." I would guess that 9 out of 10 LWers who sleep irregularly probably could fix the problem with attention and instrumental rationality.
comment by igshipilov · 2024-05-13T11:45:18.584Z · LW(p) · GW(p)
Record the bugs
Minimum action: Strike a note in Todoist, as is, albeit drafted (note: I always have Todoist on hand).
Yoda Timer
Minimal action: Start a timer on your phone.
PTD/TAP
Minimal action: Whenever you think of adding a repetitive thing to your life, give yourself 30-60 seconds and answer the question, "What is the simplest thing I can do regularly in this area?"
For example:
- To get accustomed to reading, is it ok for me to read at least one paragraph a day?
- To pump up my English, is it normal to watch one 10-20 minute TEDx talk in the original every morning?
Design
Minimal action: Once you notice that you are not happy with something in your environment or on your computer or phone, then answer yourself the question, "Can I fix it in 1-2 minutes?". And if the answer is, "Yes!" then just do it.
Factorization of Avoidance
Minimal action: When you notice that you don't feel like doing the same thing over and over again, then think for 10-30 seconds, "Why don't I want to? Is something scary going to happen? Or does this deed not seem meaningful?"
Avoidance Factorization
Minimum action: When you notice that you don't feel like doing the same thing over and over again, then reflect for 10-30 seconds, "Why don't I want to? Is something scary going to happen? Or does this deed not seem meaningful?"
Belief in sunk costs
Minimal action: When you feel like quitting, remember how you washed your lids for a week (clarification: this was my "homework" on restoring my belief in sunk costs).
Time Calibration
Minimum action: At the beginning of any task, give yourself 5-10 seconds and intuitively estimate how long that task will take. At the end of the task, look at the actual time taken and remember your prediction. How different is it from reality?
Murphyjitsu
Minimal action: Take exactly 1 minute to think of one possible cause of failure and immediately think of how you can prevent it or reduce its likelihood or deal with its consequences cheaply.
Push where effective
Minimum action: Give yourself 5 seconds and then intuitively, from System 1, suggest where to push where it would be most effective.
Search for new things in the past
Minimal action: When you learn something new and are about to apply it, whether you are reading or being told, take exactly 5-10 seconds to think: Have you done something similar before?
Cancel superstition
Minimum action: When you have successfully completed System 1, intuitively answer the question, "Which of my actions or habits contributed most to this victory?"
Focusing
Minimal action: For five seconds, notice where in the body something has contracted, where a chill, a tingling sensation has appeared.
TDT for people
Minimal action: Say to yourself, "Next time it will be 10% easier."
Friendship
Minimal action: Just think: Who could be on a call with you while you solve this problem? Diman? Someone from the learning chat rooms? Or a programmer? Maybe my girlfriend?
Silence
Minimum action: When you feel dreary and seem to want to "go nowhere" and not experience anything at all, just let your thoughts roam free - don't hold them back. You don't have to write them down, just let them wander and express whatever they want. Promise the thoughts that you won't judge them - they can speak freely.
Inner root of disagreement (IFS)
Minimal action: Give yourself 30-60 seconds and name as many body sensations and your feelings about this situation as you can.
Reductionism
Minimum action: Give yourself 10-30 seconds and intuitively, from System 1, answer the question, "What is the ONE skill behind this case? (meaning: 'ONE of the skills')""
For example:
- What one skill (one of) lies behind the ability to argue and defend your point of view? (in person, in voice, not online in text)
- What one skill lies behind writing music?
comment by mraxilus · 2018-04-29T15:48:23.665Z · LW(p) · GW(p)
We know that students can’t move on to algebra until they have perfectly memorized their times tables.
That's how most people progress into algebra, but I don't think you need perfect memorization of the times tables, simply the ability to reproduce it.
comment by TAG · 2018-03-21T16:58:50.245Z · LW(p) · GW(p)
Irreducible and mysterious complexity, as we know, is a property of the map and not of the territory.
You don't know that, and the linked article doesn't demonstrate it. Instead, it is mostly an objection to complexity as concept of limited usefulness.