table of contents
A couple of months ago, the topic of group conversation turned to credit card rewards. Someone mentioned that they found it interesting how credit cards functioned as a wealth concentrator in that rewards points are effectively subsidized by poorer users. Immediately my mind jumped to something I’d remembered Patrick McKenzie write on the topic. I say “something” because I knew he’d addressed that exact argument, but I couldn’t remember the substance of his actual point. In fact, I couldn’t conjure up even the faintest hint of what he’d actually said. The conversation moved on while my brain revolved at terminal velocity in desperate recall mode.
Turns out my memory was a bit off that day — it was actually a podcast episode (opens in new tab), not a blog post — but I was correct enough: enough to recall that an expert had shared a great many thoughts on the subject. If only I’d been able to recall any of it!1
My initial diagnosis was that I’d forgotten the point because I hadn’t written it down. As supporting evidence I weighed up all the times I’d intentionally summarized an interesting piece of information, and remarked on how easily I could recall them months or years later. I even knew that I knew that writing things down makes them easier to recall. This aligns with my mental model for learning — as Leslie Lamport (probably) said, “If you’re thinking without writing, you only think you’re thinking”. So why didn’t I do that when it came to credit cards rewards?
First, I gave myself retroactive permission to be lazy. Empirically speaking, maintaining a knowledge base seems to be hard work! Many people seem to run into problems with formal PKM maintenance — think curation issues and byzantine, unusable network graphs. But that wasn’t really my issue, as I’d never experienced any problems keeping things clean. My initial architectural conditions seemed fine, or at the very least they hadn’t backfired at scale. Next, I revisited my 2020 piece about using Obsidian as a “second brain” (opens in new tab). Looking back at that post, I see an unguarded optimism that I don’t share today, in large part because I eventually stopped writing much of anything in Obsidian. There are entire months without notes; months where I must have read hundreds of Hacker News-linked blog posts and journal articles and the like, but apparently didn’t bother to commit that text into any more permanent form. My brain will remember this, I must have thought, or maybe I know this already so there’s no benefit to writing it down. When you’ve got a vague intuition for how things work, you’re more comfortable taking the cognitive shortcut of telling yourself that you already understand it well enough to skip the summary.
So this shortcut turned out to be a bad choice, but maybe it’s a forgivable one. Much like homework, summarization is subject to temporal discounting: the immediate cost of summarizing information is extremely concrete, whereas the long-term value of a summary is less clear-cut. You can’t predict that someone will want to talk about credit card rewards in the coming six months, so the benefit of summarizing it now feels intangible. And if that’s true, I must not have believed in or internalized the long-term value of summarizing a given article.
The other conclusion I’ve been tempted to draw is that maybe the knowledge base wasn’t doing what I wanted in the first place. It’s all well and good to consult your system when drafting an email or responding to a query. But this does you little good in the heat of the moment, unless you plan on telling your interlocutor to hold that thought while you pull out your phone and regurgitate an Obsidian summary. And yet somehow I’d hoped that the system would enable precisely that kind of real-time activity. If you want intelligent things to say on a wide variety of topics, writing them down might make them stickier, but trying to surface them with a tool designed to be deliberately searched and queried — something that inherently takes time — might not have been the most direct approach.
I suppose this is the paradox at the heart of knowledge management. Writing things down can help you internalize them, but the mere act of transcribing knowledge into a PKM system isn’t magic. There’s an intellectual step between storage and synthesis that was clearly missing in my credit card case, and there’s no shortcut to social proof of intelligence. At some point you’ve got to do the homework!
Footnotes
-
Armed with the benefit of both hindsight and the Internet, I can now report that card schemes’ profit objective functions are indeed more complex than “poor users subsidize rich users”, as North American card segmentation pretty cleanly suggests. If anything, McKenzie argues that most credit card users are actually subsidizing those with middling credit scores. ↩