Three Kinds of Research Documents: Exploration, Explanation, Academic
Aug 2020 Edit: Changed “Clarification” to “Exploration”, thanks to a comment by Richard_Ngo
Epistemic Status: Low. This was a quick idea, but the grouping honesty doesn’t work as well as I’d like. I still think it could be useful to some people though. Ideas appreciated.
Recently I have started writing more and have been trying to be more intentional with what I accomplish. Different documents have different purposes and it seemed useful to help clarify this. Here is a list of three specific different types I think are relevant on LessWrong and similar.
Exploration
I see exploration posts as generally the first instance of information being written down. Here it is important to get the essential ideas out there and to create consensus around terminology among the most interested readers. In some cases, the only interested reader may be the author, who would use the post just to help cement their ideas for themselves.
Exploration posts may not be immediately useful and require later posts or context for them to make sense. This is typically fine. There’s often not a rush for them to be understood. In many cases, there is a lot of possible information to write down, so the first step is to ensure it’s out there, even if it’s slow, hard to read, or doesn’t much make sense until later.
I think of many of Paul Christiano’s posts as exploration posts. They’re very numerous and novel, but quite confusing to many readers (at least, to myself and several people I’ve talked to). Sometimes the terminology changes from one post to the next. I used to see this is somewhat of a weakness, but now it comes across to me as a pragmatic option. If he were to have tried to make all of this readable to the average LessWrong reader, there’s likely no way he could have written a portion as much.
One important point here is that if something is a exploration post, then the main relevant feedback is on the core content, not the presentation. Giving feedback on the readability can still be useful, but it should be understood and expected that this isn’t the main goal.
Explanation
Explanation posts seek to explain content to people. The focus here is on accessibility. Often the main ideas are already documented somewhere, but the author thinks that they could do a better job explaining them to their intended audience.
I would categorize some of the recent posts on Embedded Agency as being explanatory. Some of them have very nice diagrams and are elegantly laid out. I believe much of the content comes from earlier work that was a lot more fragmented and experimental. Zhukeepa’s recent overview of Paul Christiano’s work also is a good example.
Academic
Academic documents, as I interpret them, aim to be acceptable to the academic community or considered academic. Some attributes that typically go along with this include:
The academic article structure
Citations, generally of other academic works
Discussion of how work fits in with existing academic literature
A high level of rigor and completeness
An expectation that the main terms and ideas won’t change much
PDF formatting
There can definitely be a lot of signaling going on here. Many people see academic seeming articles as substantially more trustworthy and impressive than other works.
That said, I feel like there are some useful attributes to these works besides signaling. For one, it’s a format well suited to interfacing with the academic world. Interfacing with the academic world can be quite valuable, especially in domains with substantial academic work. Also, the format has become popular for some valid reasons around robustness and context.
As an example, MIRI’s official papers fit into this category.
Academic-oriented posts don’t need to be PDFs. I would consider my post on Prediction-Augmented Evaluation Systems to partially be in this category, and several EA Forum posts to partially be in this category (examples here, here, and here.)
There are some documents that do a good job being both “academic” and “explanation.” I think these should be considered a mix of both.
Further Thought
I think the main take away of this post is that some documents exist for the main purpose of exploration, and should be understood as such. I myself currently have a lot of ideas I want to write down and intend to focus on exploration posts for a while.
The distinction between explanatory and academic documents doesn’t seem as novel nor as elegant to me. I’d be really curious if readers can post in the comments with improvements on this ontology or better examples.
It seems backwards that the first kind of writing (where readability isn’t the main goal) is called “clarification”, since Google defines it as “the action of making a statement or situation less confused and more comprehensible” and gives “explanation” as a synonym.
After spending some time with a dictionary/thesaurus, I think “description”, “delineation”, or “sketch” might be a better term for it?
Yeah this through me quite a bit
Thanks for the ideas!
I was also hesitant to use “clarification”, but do kind of think about it as one “clarifying” their messy thoughts on paper, and clarifying them with a few people. I feel like “sketch” implies incompleteness, which is not exactly the case, and description and delineation are not descriptive enough.
Some other related terms. Blueprint, survey, first pass, embryonic, pioneering, preliminary, germinal, prosaic.
Ok, that makes some sense, but the usage still conflicts strongly with the everyday usage of “clarification”, in a way that’s likely to cause a lot of confusion. Consider if someone says “Please read this clarification of my idea.” Wouldn’t the listener naturally (if they didn’t know the speaker was using “clarification” in your sense) think that the document was written with readability as a main goal?
I think in contrast, the downsides of “description” and “delineation” are not nearly as high as that of “clarification”. That is, if the listener didn’t know the specific intended meaning of “description” or “delineation” at least they won’t end up confused.
Agreed that “clarification” is confusing. What about “exploration”?
I like exploration. Could also see synonyms of exploration. “discovery”, “disclosure”, “origination”, “introduction”
I finally got around to changing this. I came back to this article, also was confused by “clarification” when I first skimmed it. I agree more now that it was a pretty poor word to use originally, apologies!
There are good non-signaling reasons for academic documents being the way that they are. Consider the following properties of academia:
A field is huge, such that it is very hard to learn all of it
The group of people working on the field is enormous, requiring decentralized coordination
Fields of inquiry take decades, meaning that there needs to be a way of onboarding new people
Consider how you might try to write explanatory posts for such a group that are shorter than books, and I suspect you’ll recover many of the properties of academic articles (perhaps modernized, e.g. links instead of citations).
I had a thought. One way to make this taxonomy more robust might be to use your RAIN framework to talk about the types of tradeoffs each document is trying to make (while assuming that all posts are trying to optimize for importance)
Clarification posts: aim to optimize novelty at the expense of robustness and accessibility.
Explanatory posts: aim to optimize accessibility and robustness at the expense of novelty.
Academic posts: aim to emphasize robustness at the expense of accessibility and novelty
Some other categories this suggests:
Meme spreading/popularization: emphasizes accessibility at the expense of novelty and robustness.
Original research: aim to emphasize robust and novelty at the expense of accessibility.
Exploratory/Speculation posts: Aim to optimize novelty and accessibility at the expense of robustness.
I really like this formulation, will consider the specifics more. It kind of reminds me of Pokemon types or Guilds in Magic.
I’ve been working on mapping explanation-space and have identified 3 areas of decomposition, the how-why-what axis corresponding to levels of analysis (Marr, et al), the temporal axis delineating past, future, and now as a special moment of causal intervention, and the variance-invariance axis.
That sounds interesting, I look forward to eventually reading more about it, if that is published online.
I would challenge this post that a good document should do all three. Preferably at the same time (although not always possible or easy).
What about short technical reports / forum posts like the original “All Mathematicians are Trollable: Divergence of Naturalistic Logical Updates”? It’s short, doesn’t have any references, doesn’t hold your hand with much background like academic articles do. On the other hand it contains more detailed information than “An Untrollable Mathematician Illustrated”.
Is this a clarification post because it’s the first example of an idea?
Good question. It’s more technical than most of the ones I was considering and I have a harder time judging it because I haven’t really gone through it. I think with pure-math posts the line is blurry because if it’s sufficiently formal, it can be hard to make more explanatory, for dedicated readers.
I imagine that there are very math-heavy posts on Agent foundations that are both optimized for readability from other viewers, and ones more made as a first pass at writing down the ideas. That specific post seems like it’s doing a good amount of work to be clear and connect it to other useful work, so I would tend to think of it as explanatory. Of course, if it’s the first time the main content is online, then to viewers it would fulfill both purposes of being the original source, and also the most readable source.
Curious how you arrived at this ontology. It doesn’t seem obvious to me as a natural grouping.
Good to know. These are the main three types I’ve been thinking about working on so it seemed personally useful, I assumed the grouping may be useful to others as well. I could definitely imagine many other kinds of groupings for other purposes.
Makes sense. Thanks!