Where to Start Research?

Link post

When I began what I called the knowledge bootstrapping project, my ultimate goal was “Learn how to learn a subject from scratch, without deference to credentialed authorities”. That was too large and unpredictable for a single grant, so when I applied to LTFF, my stated goal was “learn how to study a single book”, on the theory that books are the natural subcomponents of learning (discounting papers because they’re too small). This turned out to have a flawed assumption baked into it.

As will be described in a forthcoming post, the method I eventually landed upon involves starting with a question, not a book. If I start with a book and investigate the questions it brings up (you know, like I’ve been doing for the last 3-6 years), the book is controlling which questions get brought up. That’s a lot of power to give to something I have explicitly decided not to trust yet.

Examples:

When reading The Unbound Prometheus, I took the book’s word that a lower European birth rate would prove Europeans were more rational than Asians and focused on determining whether Europe’s birth rates were in fact lower (answer: it’s complicated), when on reflection it’s not at all clear to me that lower birth rates are evidence of rationality.
“Do humans have exactly 4 hours of work per day in them?” is not actually a very useful question. What I really wanted to know is “when can I stop beating myself up for not working?“, and the answer to the former doesn’t really help me with the latter. Even if humans on average have 4 hours, that doesn’t mean I do, and of course it varies by circumstances and type of work… and even “when can I stop beating myself up?” has some pretty problematic assumptions built into it, such as “beating myself up will produce more work, which is good.” The real question is something like “how can I approach my day to get the most out of it?”, and the research I did on verifying a paper on average daily work capacity didn’t inform the real question one way or the other.

What would have been better is if I’d started with the actual question I wanted to answer, and then looked for books that had information bearing on that question (including indirectly, including very indirectly). This is what I’ve started doing.

This can look very different depending on what type of research I’m doing. When I started doing covid research, I generated a long list of fairly shallow questions. Most of these questions were designed to inform specific choices, like “when should I wear what kind of mask?” and “how paranoid should I be about people without current symptoms?”, but some of them were broader and designed to inform multiple more specific questions, such as “what is the basic science of coronavirus?”. These broader, more basic questions helped me judge the information I used to inform the more specific, actionable questions (e.g., I saw a claim that covid lasted forever in your body the same way HIV does, which I could immediately dismiss because I knew HIV inserted itself your DNA and coronaviruses never enter the nucleus).

I used to read a lot of nonfiction for leisure. Then I started doing epistemic spot checks– taking selected claims from a book and investigating them for truth value, to assess the book’s overall credibility- and stopped being able to read nonfiction without doing that, unless it was one of a very short list of authors who’d made it onto my trust list. I couldn’t take the risk that I was reading something false and would absorb it as if it were true (or true but unrepresentative, and absorb it as representative). My time spent reading nonfiction went way down.

About 9 months ago I started taking really rigorous notes when I read nonfiction. The gap in quality of learning between rigorous notes and my previous mediocre notes was about the same as the gap between doing an epistemic spot check and not. My time spent reading nonfiction went way up (in part because I was studying the process of doing so), but my volume of words read dropped precipitously.

And then three months ago I shifted from my unit of inquiry being “a book”, to being “a question”. I’m sure you can guess where this is going- I read fewer words, but gained more understanding per word, and especially more core (as opposed to shell or test) understanding.

The first two shifts happened naturally, and while I missed reading nonfiction for fun and with less effort, I didn’t feel any pull towards the old way after I discovered the new way. Giving up book-centered reading has been hard. Especially after five weeks of frantic covid research, all I wanted to do was to be sat down and told what questions were important, and perhaps be walked through some plausible answers. I labeled this a desire to learn, but when I compared it to question-centered research, it became clear that’s not what it was. Or maybe it was a desire to go through the act of learning something, but it was not a desire to answer a question I had and was not prioritized by the importance of a question. It was best classified as leisure in the form of learning, not resolving a curiosity I had. And if I wanted leisure, better to consume something easier and less likely to lead me astray, so I started reading more fiction, and the rare non-fiction of a type that did not risk polluting my pool of data. And honestly I’m not sure that’s so safe: humans are built to extract lessons from fiction too.

Put another way: I goal factored (figured out what I actually wanted from) reading a nonfiction book, and the goal was almost never best served by using a nonfiction book as a starting point. Investigating a question I cared about was almost always better for learning (even if it did eventually cash out in reading a book), and fiction was almost always better for leisure, in part because it was less tiring, and thus left more energy for question-centered learning when that was what I wanted.