I misunderstood the nature of research for most of my life, and this prevented me from doing any. I thought significant research came from following the scientific method until novel discoveries popped out. I'd never contributed something new to human knowledge before, so being a researcher—which required replicating this outcome—felt impossibly far out of reach.
But it turns out the novel discovery is just a side effect. You don't make novel discoveries by trying to make novel discoveries.
Instead, research is simply a continuation of something we already naturally do: learning. Learning happens when you understand something that someone else already understands. Research happens when you understand something that nobody else understands yet.
This hit me during COVID quarantine when I escaped to the mountains of Santa Fe to learn probability theory from a dear mathematician friend. He began by asking me to teach him something simple, like combinatorics. I said that seemed pointless, as certainly he already knew combinatorics, but he assured me he'd forgotten much and that it'd help. I nervously brushed up, preparing. As I began whiteboarding, he started asking questions, and pretty soon I felt like he was actually learning from me—exhilarating.
At some point, he asked a question about multinomial distributions I couldn't answer. I realized I wouldn't be able to find the answer in any textbook or paper, and in fact, if we wanted the answer, we'd have to work it out ourselves. We hunkered down and began playing with the problem. A day later, having filled the whiteboard several times over, I understood multinomial distributions far more deeply than I ever imagined I would.1
Offhand, mired in the complexity of the problem, he remarked: "Honestly, I don’t think humanity knows the answer to this question yet."
Wow. I felt as if my whole life I'd been hiking in a dense fog-ridden forest and suddenly stumbled upon a lush valley saturated with sunshine and stunning clarity. This was research. I wasn't straining to discover something new. I was accidentally doing it because I was curious, because my friend had asked a question I couldn't answer and it seemed nobody else had figured it out either.
Research, I realized, is what happens as a byproduct when you try to understand something and hit the bounds of what humanity currently knows.2 At that point, there's suddenly no one who can tell you the answer.
If you care enough about the question, you have to figure out how to answer it yourself, and that's when you start running experiments and developing hypotheses. That rote process of science we're taught in school—to start with a question, generate hypotheses, test with experiments, draw conclusions—it's a good tool, but it doesn't capture the most important element: actually wanting to know the answer to the question!
When our goal is to understand something, we start getting curious: why does this work this way, and not that way? Why doesn't this do what I expected? Strange! These observations of strangeness, where pieces of the understanding puzzle don't fit, lead to exploring beyond what's known.
In retrospect, this seems a bit obvious, especially if I imagine Newton or Darwin. I doubt they went around asking: "What new thing can I discover?" Rather, they made incredibly detailed observations about what confused them and tried very hard to piece together the laws that described what was going on. At some point, after having gone extraordinarily deep in an uncharted direction, they were each able to turn their understanding into an explanatory model, resulting in a tremendous novel discovery. While from the outside, we believe the discovery to be a magical eureka moment—an apple falling on Newton's head—from the inside, it feels more like a painstaking process of incrementally deeper understanding.
Put another way, the generating function for novel work is trying really hard to deeply understand something until you pass through the edge of current human understanding, and then continuing imaginatively onwards.
Of course, that understanding needs to be augmented with a model of what’s already known. Without such a model, you may think you're discovering new things at the edge, but you're actually in the center rediscovering what's known. With a good model, in many areas it's amazing how quickly you shoot beyond the edge.3
When we see research as a process of understanding rather than a process of producing novel discoveries4, it becomes clear that many curious, bright people who don’t believe they can be researchers already have the foundational skills needed. We're naturally curious; we already know how to learn, how to notice when something doesn't make sense. We ask questions, and we work out the answer. At some point, we'll ask questions where nobody has the answer. At that point, we’re starting to do research.5
Viewed as a way to understand reality more deeply, suddenly research shifts from something intimidating to something beautiful, this process of slowly feeling the inner workings of the universe come into focus, clicking into place, a blossoming of clarity suffusing, igniting the jungle of neurons traversing my mind, until I can finally see.
Footnotes
[1] This required a big mentality shift from the habits I internalized in school—I had to repeatedly squash my urge to "just understand things enough" (enough for the test, of course) and move on. That externally-motivated approach to understanding blocked my ability to play with the problem. The playfulness came from inside, from curiosity, enjoyment, humor, amazement.
[2] I used to model understanding as a data download, a process of writing what people already knew into my brain. Now I realize that it's a creative process, a process of actively constructing explanations of how things work.
[3] Note that under this definition, many things we think of as research would not be considered research! For example, publishing papers looks like research, but in machine learning, many papers get higher performance on a benchmark but don't necessarily increment our understanding. I'm not clear on whether this is an advantage or a shortcoming of this definition. ¯\(ツ)/¯
[4] Ph.D. advisors are helpful because the good ones are essentially a walking model of what's known, which makes finding the edge much easier. There are some additional interesting questions here, like: how do you get to the edge quickly? How do you do that across multiple fields? What do you do if the field seems misdirected, like much of psychology? At some point I'd like to write more about this, so if you have any thoughts, please send them my way :)
[5] There are two more general lessons here: first, not understanding how any given process works reduces your sense of agency, and it's worth identifying other such black boxes in your mind. And second, it's common to mistakenly set the outcome of a process as your goal, as I did by believing I should try to make novel discoveries. The thing to emulate is the process, not the outcome.
Thank you Josh Albrecht, Laura Deming, Sneha Shah Jain, Michael Nielsen, and Erik Torenberg for many helpful insights and comments on drafts of this essay, and thank you Sebastien Zany for prompting this insight about research in the beautiful mountains of Santa Fe.
For comments and discussion, see this Twitter thread.