Learning from scattered resources
By learning from scattered resources I mean the kind of learning that effective altruists or rationalists do to catch up on the state-of-the-art in EA/rationalist thinking, or the kind of learning that Duncan Sabien mentions of how he learned parkour: "Speaking as someone who pieced together the discipline of parkour back in 2003, from scattered terrible videos (pre Youtube) and a few internet comment boards—pulling together a cohesive and working practice from even the best writeups is a tremendously difficult task."
This kind of learning, where one (1) actively goes searching for many resources (each of which contains only a small amount of information) and (2) receives relatively little feedback from people who know about the topic, seems different from the kind of learning that happens in other situations:
- in school or apprenticeships, there is usually a teacher/mentor or textbook that contains the vast majority of the information that is to be learned
- when learning on the job, there are again coworkers/bosses, but also there's constant feedback on job performance
I think maybe a lot of learning that happens in various subcultures is like this. As of 2019, I think learning how to use spaced repetition software is like this.
Various tricky things that can happen in this kind of learning:
- there are multiple people giving little bits of info, and different people can have different opinions and can contradict each other
- depending on the discipline, the majority of the people can be untrustworthy or unreliable in other ways
- the identities of the people giving the info are numerous and difficult to verify, so one must rely on one's assessment of the object-level details (as well as cues like grammar and punctuation)
- many topics are not explained well, and have to be pieced together by the learner
- there can exist a "grapevine" or various private discussions where people give frank thoughts, in contrast to the public discussions (note: even in more established disciplines this happens, e.g. search "secret paper passing network")
- sometimes there can be a huge amount of information out there, but very little in an organized format, so the problem is how to prioritize which things to read/understand (e.g. AI safety as of 2020 is like this for a newcomer who doesn't have a ton of background in machine learning, math, CS in general)
- sometimes there can be very little information at all, so part of the problem becomes asking people for information or trying to invent the material oneself