The Craft and the Community: Post 28 – 30

28. Practical Advice Backed By Deep Theories

If Yudkowsky had merely written that “we overestimate the degree to which other people understand our thoughts and folks in general overestimate how much they understand what the other person is saying” I wouldn’t have been very impressed.

But he elaborated on this advice with several posts (such as illusion of transparency, inferential distance, and double illusion of transparency) and explained in detail the underlying, general cognitive mechanisms which allowed me to develop a deeper understanding. Which in turn makes it more likely that I remember this advice and also use it.

This is the signature style I want to convey from all those posts that entangled cognitive science experiments and probability theory and epistemology with the practical advice—that practical advice actually becomes practically more powerful if you go out and read up on cognitive science experiments, or probability theory, or even materialist epistemology, and realize what you’re seeing.  This is the brand that can distinguish LW from ten thousand other blogs purporting to offer advice.

He concludes:

…practical advice really, really does become a lot more powerful when it’s backed up by concrete experimental results, causal accounts that are actually true, and math validly interpreted.

29. Less Meta

Yudkowsky is afraid that the recent posts have gotten a bit too meta. Therefore he (contrary to Newsome 😉 ) recommends to talk about the object level, i.e. the boring and practical stuff.

30. Go Forth and Create the Art!

(This is the last post of the last Sequence and boy, I’m glad it’s over! Don’t get me wrong, I mostly enjoyed reading the Sequences, but after a while writing comments and summaries really got on my nerves. 🙂 )

After he laid the groundwork, Yudkowsky hopes that others will be able to improve their rationality on their own, and maybe even create different styles and new rationality skills.

I suspect—you could even call it a guess—that there is a barrier to getting started, in this matter of rationality.  Where by default, in the beginning, you don’t have enough to build on.  Indeed so little that you don’t have a clue that more exists, that there is an Art to be found.  And if you do begin to sense that more is possible—then you may just instantaneously go wrong.  As David Stove observes—I’m not going to link it, because it deserves its own post—most “great thinkers” in philosophy, e.g. Hegel, are properly objects of pity.  That’s what happens by default to anyone who sets out to develop the art of thinking; they develop fake answers.

He concludes:

…My last essay on having a secret identity was not well-received, so let me try again:  I want people to go forth, but also to return.  Or maybe even to go forth and stay simultaneously, because this is the Internet and we can get away with that sort of thing; I’ve learned some interesting things on Less Wrong, lately, and if continuing motivation over years is any sort of problem, talking to others (or even seeing that others are also trying) does often help.

But at any rate, if I have affected you at all, then I hope you will go forth and confront challenges, and achieve somewhere beyond your armchair, and create new Art; and then, remembering whence you came, radio back to tell others what you learned.

In essence:

Step 1. Reading LessWrong.

Step 2. Rationality and stuff. Or something.

Step 3. Achieving godhood.

Easy as pie!

 

5 comments on “The Craft and the Community: Post 28 – 30

  1. -

    If going meta causes you problems, you need to go meta on going meta. Saying that meta arguments are flawed is a step in the right direction, by being a meta-meta argument. You still need a few more levels, though. I’m currently at 8 or so. Also, confusing reference to an obscure theologian and/or almost-reasonable concept in an unreasonable context.

    (I do plan on writing more than Newsome-style mindfuckery about that eventually, as part of the meta-morality posts, but dunno when. As a meta-meta-meta argument, I’m not yet convinced that this is really important for anything besides philosophy (in a broad sense), but I’m serious about the main point – if meta isn’t helping, maybe you’re not using enough of it.)

    • -

      Additional point, based on the recent meta-discussion fatigue on LW. (I occasionally read it, even though I’m still technically on vacation.)

      There’s a difference between exploring a meta-level more and going up to a new meta-level. When people get tired with meta arguments (assuming they aren’t literally incapable of decent abstract thinking), I suspect they really are bored by the *current* level, mostly because of diminishing returns. It might be a good idea to go down one (or more) levels at that point, but often finding new things to abstract away can be even better.

      So for example, imagine a bunch of people trying to learn a language. They might struggle with exercise 4, 8, 15, 16, 23 and 42. That’s our object-level.

      That looks like 6 difficult problems until someone has a clever idea: “Those problems are about vocabulary, so we should find out to learn vocab in general!” That’s the first meta-level.

      Discussion might look like this:

      “We should use drills and rote repetition as God intended!”
      “No, fill-in-the-gaps!”
      “Lots of those words are household objects. We could label our own stuff as a reminder!”

      And so on. Eventually, the group gets tired of all this meta-discussion. But going back to the object-level would be a huge waste! If they went meta on learning words, they start asking themselves how to remember facts in general (meta-meta) and discover Spaced Repetition.

      And then ask themselves how they figured out it was about vocabulary. Are there are other important features? How would you know? (And discover morphological analysis, how to order facts and find something like Direct Instructions.)

      And because meta-levels aren’t really linear, instead they could ask themselves how the material was actually chosen. How do you get vocab lists anyway, especially good ones? (And find stuff like subs2srs and other automatic tools.)

      And how do we know that we found good solutions? (And discover empiricism.)

      (And then go meta on language-learning, and ask yourself why you do it, and discover evopsy, cogsci, etc., get stuck in a never-ending quest for knowledge, never getting anything done and join LW. And the go meta on that, ask yourself why you pursue knowledge anyway, discover signalling theory and become a cynical asshole and/or sociopath. Or gods. I forgot when the gods come in.)

      Anyway. tl;dr: When you ask “How do I solve X?” and A, B and C aren’t working, try asking “How do I figure out how to solve X?”. Repeat ad nauseam ad maiorem Dei gloriam (Bad muflax! No theology jokes in a meta post!).

      • -

        Yeah, meta is awesome.

        The only objection that came to mind is that you need to make sure that you don’t get lost in ever more meta-levels and eventually lose touch with reality.

        But then you wrote:

        “And how do we know that we found good solutions? (And discover empiricism.)”

        Which is a good counter-argument. However, it also makes the distinction between meta-levels and the object-level kinda fuzzy (or does it?).
        I mean, empiricism works because some guys actually do the basic and boring object-level stuff and test which theories are true and which are bullshit. To use the language-example: We only know that spaced repitition is great cause folks like Ebbinghaus and Spitzer did the dull practical research on the object-level.

        Of course, you could reply that I’m not being meta enough. 😉
        I could try to convince others to test my theories for me. Or I could try to convince others that they should convince others to test my theories for me. Or something. (Is this even genuinely meta?)

        “And then go meta on language-learning, and ask yourself why you do it, and discover evopsy, cogsci, etc., get stuck in a never-ending quest for knowledge, never getting anything done and join LW. And the go meta on that, ask yourself why you pursue knowledge anyway, discover signalling theory and become a cynical asshole and/or sociopath. Or gods. I forgot when the gods come in.”

        Haha, nice one. But this brings up an even bigger problem: Does the ascending ever stop? Is there a “last” meta-level or an infinite regress? Or is there some circular magic a la “where recursive justification hits bottom?”

        tl;dr: Can I remain a cynical asshole or should I try to become a god? And why?

        Pretty confusing topic.

  2. -

    “Can I remain a cynical asshole or should I try to become a god? And why?”

    False dichotomy: do both. 😉

    But yeah, the object-level does matter, and it’s to some degree relative, in that you can take meta-levels as objects themselves, so whether something is meta or not depends somewhat on your framework and goals.

    I think (~70%) that going meta is bounded in both directions. In the object direction, raw perception and choice are likely as close to the bottom as it gets, and I currently fail to think of anything less meta than them. In the meta direction, I suspect that power stops increasing at some point and you reach a level that is it’s own meta-level.

    So basically like the Chomsky language hierarchy with Turing-complete languages at the top – once you have that level, it’s incredibly hard to find something more powerful, maybe even impossible. Similarly with theories, once you are at a Tegmark / modal realism / Omega level, I doubt you can actually go any more meta.

    I don’t have a formal argument for that, but intuitively, “going meta” takes not a single instance, but a class, a generalization of instances as its object. At the lower levels, the resulting frameworks aren’t themselves part of the class you took your instances from. (“Spaced repetition” is not a word itself, but about words.) However, once you have sufficient meta-power, then whatever features you choose, the resulting abstraction is itself an object of its domain. (Turing machines can simulate themselves, formal system can prove theorems about formal systems, etc.) So you can’t go “more” meta, but you can still explore that meta-level because there are always new features you can look at, and Gödel seems to forbid you from having the ultimate abstraction that leaves nothing out.

    And then the problem doesn’t so much become a search for more powerful tools, but rather features you haven’t tried abstracting yet. (Or this is all insanity, like 6 Degrees of Kevin Bacon, only for meta-heads.)

    • -

      “False dichotomy: do both.”

      Haha, good idea!

      But yeah, your theory makes sense. Now I just have to use this knowledge and achieve godhood.

Leave a Reply