First principle thinking is often presented as a foil to reasoning by analogy. The premise in first principle thinking is to go as deep as possible to reach proverbial intellectual building blocks / fundamental truths, and then use those to deductively solve the unique problem in front of us. This contrasts with reasoning by analogy, where we (often implicitly) assume that the problem we are working on is similar to one or more other problems we’ve seen before, and thus assume that whatever the best solution was for those other problems is the best solution to the one in front of us. While this seemed like a stable delineation to me for a long time, I’ve now come to believe that there is no firm boundary between the two - that so called “first principle thinking” is simply a different form of “reasoning by analogy”.
Let’s steel man this distinction between first principle and analogous reasoning before we try to tear it apart. In Michael Lewis’ The Premonition, he writes early on about the academic debate about the efficacy of non-pharmaceutical interventions (NPIs)1 in pandemic response, particularly when it comes to school closures2. As he writes,
“… the people inside the American government who would be charged with executing various aspects of any pandemic strategy thought all models of disease control were bullshit. They thought closing schools was a stupid idea. They believed none of these so-called non-pharmaceutical interventions would contribute anything but economic loss.”3
This view was heavily informed through reasoning by analogy. These people looked at past pandemics (particularly the 1918 pandemic), saw that NPIs did not appear to help, and concluded they don’t work. As Lewis lays out in his book, an important part of challenging this view was through computer modelling. While crude and laden with assumptions, some of which are surely incorrect or overly simplistic, these models suggested very strongly that school closures were an effective tool in slowing a potential or ongoing pandemic. This is an example of first principle thinking - taking our understanding of the foundations of how viruses spread and how people interact and using that to attempt to predict how certain interventions would influence pandemic development.
We’ll return to NPIs shortly, but let’s first consider another example of first principles thinking - friction. This is one of the mental models in The Great Mental Models Volume 2: Physics, Chemistry, and Biology by Parrish & Beaubien - mental models are basically generalizable first principles from different fields which may be applicable in other domains. They write about friction as follows,
“Friction is a force that opposes the movement of objects that are in contact with each other, such as the wheels of a pair of roller skates across the ground. For objects to move, they must overcome friction that pushes in the opposite direction. This requires extra energy…”4
This mental model is something I think about a lot when working with others. As a programmer, I frequently found ways to clearly signal that I didn’t want to be disturbed when working in an office. This ranged from wearing headphones to putting up the hood on my sweater to working away from my desk to be harder to find to glaring at people who might want to talk to me while I was working. The efficacy of this can be explained by the principle of friction - I introduced practical and social costs to trying to communicate with me, at least in person.
However, I wasn’t literally introducing friction. While I increased the energy others needed to expend to communicate with me, we can conceive of situations where others might actually distract me more as a result of making myself scarce (imagine being 13 trying to get your homework done and hiding away in the school library, increasing the motivation of your bullies to find and harass you). The relevant insight here is that in physics, an increase in friction slows the movement of something down, all else being equal. That is not necessarily the case outside of physics, despite the compelling logic (and bluntly, results) underlying my efforts to increase communication friction in my professional life. An increase in “friction” could actually increase, or minimally not influence, the rate of distraction.
Put simply, applying insights in physics such as friction to other areas is itself reasoning by analogy. As far as I can tell, the only true first principles are those in sciences relating to the rules which govern reality. Any other nominal first principles are akin to applying friction in contexts outside of physics. Instead, what first principles appears to be doing is inverting the direction of analogizing. Rather than looking at the specific context we are interested and looking for similar examples, we start with highly generalizable principles and see if / how they relate to the particular context at hand. I suspect that part of why first principles reasoning works is that it helps mitigates cognitive biases such as the narrative fallacy and anchoring more likely to appear in the opposite direction.
With that said, let’s return to the school closure question to defend (at least in part) non-first principle reasoning. To further push the argument in favour of NPIs, an academic paper was published looking more closely at these interventions in 1918. Per the paper,
“… cities in which multiple [non-pharmaceutical] interventions were implemented at an earlier phase of the epidemic had peak death rates ≈50% lower than those that did not…”
This makes perfect sense to me - the above analogous reasoning above that NPIs don’t work committed a statistical error - specifically failing to segment along an important variable (how early the NPIs are implemented), which obfuscates that they can be effective if done early enough. This is what statistical approaches such as regression analysis are attempting to get at - finding better high level analogies, which may be the best that we can do in situations where we can’t start from first principles due to a lack of theoretical understanding of the subject (say for instance that we lacked theoretical models of epidemiology to create the aforementioned computer simulations).
I ultimately think that first principle thinking will produce better reasoning on average, though it’s significantly harder and more time consuming. With that said, my point is that it’s not a panacea. We’re still analogizing, and we retain risks that the future will not look like the past (and so analogizing from the past will mislead us) or that we will improperly compare situations.
Preview image from Shashivarman on Unsplash.
In contrast with pharmaceutical interventions - vaccines.
This is, unsurprisingly, a charged subject. I want to be very clear that I’m not making any claims about the value of school closures on balance - i.e. the tradeoffs inherent in school closures. I’m writing exclusively from the perspective of whether closing schools is effective in slowing the spread of viruses.
The Premonition, Michael Lewis, pg. 99.
The Great Mental Models Volume 2: Physics, Chemistry, and Biology, Parrish & Beaubien, pg. 66.