[SEQ RERUN] Truly Part of You
Today’s post, Truly Part Of You was originally published on 21 November 2007. A summary (taken from the LW wiki):
Any time you believe you’ve learned something, you should ask yourself, “Could I re-generate this knowledge if it were somehow deleted from my mind, and how would I do so?” If the supposed knowledge is just empty buzzwords, you will recognize that you can’t, and therefore that you haven’t learned anything. But if it’s an actual model of reality, this method will reinforce how the knowledge is entangled with the rest of the world, enabling you to apply it to other domains, and know when you need to update those beliefs. It will have become “truly part of you”, growing and changing with the rest of your knowledge.
Discuss the post here (rather than in the comments to the original post).
This post is part of the Rerunning the Sequences series, where we’ll be going through Eliezer Yudkowsky’s old posts in order so that people who are interested can (re-)read and discuss them. The previous post was Artificial Addition, and you can use the sequence_reruns tag or rss feed to follow the rest of the series.
Sequence reruns are a community-driven effort. You can participate by re-reading the sequence post, discussing it here, posting the next day’s sequence reruns post, or summarizing forthcoming articles on the wiki. Go here for more details, or to have meta discussions about the Rerunning the Sequences series.
This level has always been awe-inspiring to me. I remember, when I was in high school, hearing from a smart friend that she’d forgotten an important equation during a math test but was able to re-derive it on scratch paper. I couldn’t even imagine a path from where I was to the point where I could do something like that.
Unfortunately, there’s no happy ending here. I’ve learned more math since then, but proofs are still a black art to me; the derivations in Thrun and Norvig’s AI class are driving home to me how incredibly unable I am at manipulating symbols by rules into a more useful form.
It seems to me that this sort of sentiment is really only useful for derivable knowledge, as opposed to historical facts. How would one go about deriving the fact that FDR was left in a wheelchair due to Polio without knowing that it was Polio?
If you’d asked me “Which disease, common in the past, causes paralysis?” my first answer would have been Polio.
Hindsight bias. Poliomyelitis was only one cause of paralysis; there remain dozens of others, including one far more likely than even polio was: physical injury.
Historical facts are all well-documented online, so they are the easiest to “derive”—just look them up and check the sources for credibility. In contrast, this approach rarely helps if you want to solve but a trivial math or physics problem, there you have to internalize the relevant knowledge.
I am having trouble reconciling the notion that re-acquiring the datapoint by rote memorization (after researching a means of accessing the datapoint) is a form of “derivation” with my understanding of what the word “derive” means in this context.
Would you agree that if you do not have access to the datapoint that FDR’s crippling was the result of Polio, you cannot derive that knowledge “from first principles”?
If you first principle is “look it up”, then why not?
A research methodology is not a first principle. A first principle is “a basic, foundational proposition or assumption that cannot be deduced from any other proposition or assumption.”
That is why not.
OK, now we are at the definitions stage, so there is no point in arguing about them. Just ask EY which definition he meant, if any, and be done with it.
I’m the one who introduced the term “first principles”, so I am the one who should be asked what definition I meant. Given that my original point was differentiating between the two value-sets, and that I was trying to convey the distinction between “figuring it out for yourself” as opposed to “being exposed to the answer”—I have to say that your continuing to re-introduce the notion of “look it up” just… isn’t very helpful to the conversation.
I ask you again; would you agree that unless you were exposed to the datapoint of what caused FDR’s paralysis, you could have no way of figuring it out for yourself?
(Note: “looking it up” is not an answer to this question. That would be a form of getting exposure to the datapoint.)
I don’t care about your personal definition since the original statement was EY’s.
If you aren’t interested in how I define a term that I introduced to the discussion as opposed to how someone who never used said term would define it, I don’t know what you’re doing but I do know two things:
It is not engaging in rational discourse.
I will not be a participant in it.
You’re correct but you should have assumed The Least Convenient Possible World. He could have picked some data you could not look up.
There’s a definition problem here, which I later on related, and gave a better way of asking the question. Short of having access to the information of what caused FDR’s paralysis, how would one go about “figuring out” that it was Polio?
Historical facts can still be derived from other historical facts.
Anyway, I think the generalization of your argument is that if you lose evidence then you may lose the ability to derive some facts.
The question is then how you distinguish necessary evidence from knowledge that could be “a part of you” but is not.
It seems (to me) that the kind of ‘knowledge’ Eliezer was referring to is knowledge-as-epistemological-model, as opposed to knowledge-as-datapoint.
I can understand the disdain for “guessing the password” but the simple fact is that at some point there’s just no way to ‘know’ something beyond having it be an act of rote memorization—that is, the only way to know the “datum” is to be exposed to the “datum”. For example: what is my given name? Short of being able to uncover that information, there is no model of reality which will allow you to ‘derive’ that knowledge without external input.
This places “Logos01′s given name” in a different category from “the relationship between kinetic energy and mass at a given velocity”. It is possible to determine a greater mass at velocity
vwill have more kinetic energy than a lesser mass at velocityvwithout being told this directly. It is not possible to make that same determination of my given name.