This post was inspired by an interaction I recently had with Charly Marie, who was giving feedback on a manuscript we’re coauthors on. It went something like this:
- Charly: We could provide more concrete recommendations about reproducibility in the Discussion section. For example, tools like papaja ensure that a paper’s Results section can be directly linked to the code that produced it.
- Me: Good idea! I’m a fan of tools that help users follow good practices. Thank you for mentioning papaja, I was not aware of it. I have been using targets myself.
- Charly: I didn’t know about targets, so that’s a fair exchange!
The last response came as a nice surprise to me. In my actual email, I had not bothered to link to the package because I had assumed that, since Charly was aware of one R package that supports computational reproducibility, it was likely that he was already aware of other members in that category. In fact, when writing my response, I even had doubts about bringing targets up: “surely,” I thought, “he already knows about this tool; he probably chose not to mention it because it is not relevant to the context at hand and I don’t want him to feel like I am pushing my own preferences over his own suggestion.”
I spent the rest of that day thinking about how close I had been to missing that positive interaction because of an incorrect assumption about my interlocutor’s knowledge, and I had to wonder how many other such mutually beneficial exchanges I might have missed in the past due to similar unexamined assumptions of mine.
Now, before I proceed, there’s a very important character I must introduce.
Meet the protagonist
In the picture below, there are two shapes. If I gave you the labels bouba and kiki, which label would you associate most closely with which shape?
The bouba/kiki effect is a well-studied phenomenon in linguistics, whereby people from very diverse backgrounds all tend to associate the round shape with the bouba label and the spiky one with kiki. Now that I know you know this, we can get back to the main story.
Are we kiki or are we bouba?
Over the days following the aforementioned email exchange, the thoughts it had prompted in me coalesced into one persistent idea: human communication had a spiky vibe to it. There was something that felt so intuitively correct about the notion, that I couldn’t help but to try and put some more concrete meaning on it. The boubafied-kiki model of communication (BKMC) is what came out of that attempt.
Under this model, a person is represented by a shape that conveys the knowledge they hold across various domains: the further out the shape extends, the deeper the knowledge one has in that domain (I attribute the seed of that idea to the visualization in Matt Might’s The Illustrated Guide to a PhD, which I came across several years ago). I posit that, given the immense size of the space of human knowledge, there’s only a very small subset of things that we can hope to know a fair bit about. Therefore, by the time we’re full-grown adults, all of us appear to be kiki-shaped entities when projected onto the model’s space:
The thing is, though, that we don’t know the precise boundaries of a fellow kiki until we start interacting with them. My third and final postulate is that we tend to assume that having knowledge about one thing makes a person more likely to have knowledge about closely related things. If we visualize interactions as probes being sent from one kiki to the other, this means that an initial exchange will very likely cause us to massively under- or overestimate the knowledge that a fellow kiki has about a certain domain:
Effectively, our initial probing has produced a heavily boubafied approximation of the kiki we are actually communicating with.
Closing remarks
Given the elaborate scientific diagrams presented above, you may think I am deeply attached to the BKMC. I am not. But I do think it has some explanatory power: people with some programming knowledge likely have been in a situation where they get pegged as “the computer person” and are suddenly expected to know how to remove malware, reset routers and fix printers. In any case, I think there’s a couple practical insights that I can extract from the model:
- Just because someone appears to be very knowledgeable about some aspect of a topic doesn’t mean I should assume they know all there is to know about it. Therefore, I should not blindly trust in their recommendations, and I should not keep myself from sharing my own perspectives on the matter.
- Conversely, when someone does not seem familiar with one particular aspect of some thing, that is not sufficient reason to assume they are completely ignorant about the broader topic. Therefore, it is a good idea to probe a few more times around related ideas lest judgment be too hastily passed.
- Finally, in terms of one’s own outward presentation: dwelling too much on your shortcomings is likely to cause others to underestimate your overall ability, while a “good first impression” can lead them to assume you’re highly competent across the board (which could actually be detrimental in certain contexts! e.g. kids who get good grades early on receiving less guidance on how to form effective study habits).
And that’s most of what I had to say about this. I think it is safe to assume that similar ideas have already been expounded on with far more rigor and thoroughness elsewhere, so I’d be happy for those in the know to point me towards those sources. Until then, there’s other tasks that I must continue getting spikier on.
BONUS: Taking the model too seriously
As I was writing this post, it occurred to me that the spikiness of our knowledge-space projections can probably be derived from more basic principles: let the volume of the shape represent something akin to learning effort and set some minimum radius that needs to be exceeded before knowledge of a particular domain can provide us with some tangible returns. Seems to me that spikiness would be a natural consequence of such a setup. The boubafication-of-the-other component seems like it could be readily linked to Bayesian optimization, but I’ll leave it at that before I nerd snipe myself. And since I’m already referencing xkcd, here’s one that is actually related to the spirit of the main post: Ten Thousand.