First principles thinking

How to think from first principles, from first principles.

I really like thinking from first principles, but while thinking about how to help my daughter develop this skill, I realized I have neither a formal understanding of nor a repeatable mechanism for how I got here. What’s more, I find that any attempts at explanation of how to actual go about doing this are quite hand-wavy at best. The process is often reductively explained by analogy (how ironic) to physics, with a vague nod to some fundamental laws here and a sprinkling of cleverness there.

Here’s my attempt to formalize this process — for myself, mostly, but hopefully it helps you too.

Thinking from first principles is about finding the right governing principles.

In an article by James Clear on the subject, Clear talks about how Elon Musk used first principles thinking to reduce the cost of launching a rocket by 10x by going down to first principles — i.e. by considering the cost of raw materials used to build rockets.

But my great complaint about this account is that (while I greatly respect Clear’s work) it doesn’t really help me understand how to think from first principles. The first principle Elon gets to is only really obvious in hindsight. Yes, Elon was able to get to a very fundamental concept — the price of raw materials for rocket production. But he could’ve just as easily arrived at a different fundamental concept — the necessity of those raw materials in satisfying the physical constraints imposed by rocket design. Or, gaining ownership over raw material sources to cut a deal with rocket creators. There are a number of “first principle” solutions, so while the first principle seems obvious in hindsight, it’s not nearly so obvious in the process of trying to find it.

And this is what I’ve found to be the biggest misunderstanding around first principles thinking: to do it well is not about your deductive capacity. It’s more often about defining what the relevant first principles are from which to begin your deductive reasoning. It’s certainly not always the correct solution to go down to physics¹, for example, if you’re talking about something decidedly more abstract, like product or engineering or relationship problems. So there must be some means of finding the correct level of abstraction at which the salient properties of the system can be explained. Identifying the proper principles for the system you are attempting to understand is key.

… and objective-seeking is next.

In reality, problems that I’m trying to solve from first principle usually have a very dubious objective. While it’s important to identify the principles that govern your system around that objective, it’s equally important to refine that objective as needed.

If we revisit the Elon example, for example, one can easily imagine how an initial objective would have to evolve. Imagine if his objective started as “negotiate cheaper rocket prices”. Regardless of what principles you use to describe the rocket-price system, reasoning in the realm of this objective would never amount to the solution Elon got to — building rockets himself. The objective adds a constraint to the principles you can reason around. The key, then, is to iterate on the objective: “launch a rocket as cheaply as possible” is the correct improvement, widening the aperture to include the prospect of building one from scratch.

Here’s another example: I want my daughter to learn piano. That was my objective, to start. The core, first parameter to consider at that point was the quality of the teacher — I believed in the principle that the quality of teaching vector largely defined the experience that she would have. I then, of course, found her a teacher and started to bring her to lessons.

But on practicing with her, I quickly discovered that, while I was impressed with the progress she was making, she was getting frustrated while practicing. I revised my set of initial principles, now believing that the quality of her experience came from multiple parameters, including:

• The quality of the teacher.

• The quality of her experience during practice.

  • The quality of her practice.

  • Her level of enjoyment.

I revised my objective accordingly: I want my daughter to gain enough piano skill to be able to experiment with and love music. The action to take toward that objective and in light of the fundamental governing principles was clear: practice with her only enough so that she isn’t frustrated, and then let her play around thereafter. Maximize piano skill against the constraint that her love for music cannot decrease.

It was an iterative process. My objective changed, and as it did, so too did the first principles that mediated that objective. The way to reason from first principles, then, isn’t to choose some a priori principles that seem to be fundamental and comprehensive, but to continually scrutinize your objective and what you believe to be the governing principles behind that objective until both, in conjunction, effectively describe the core problem you’re trying to solve.

Some other things that have helped me: sacred cows

One final comment here: one of the biggest things that gets in my way when trying to think from first principles are sacred cows — the things I want to be true, for reasons orthogonal to my principal objective. The insidious, destabilizing bias trying to creep in.

This is something that the startup world is rife with. When building a new product, for instance, it’s quite natural to start to love the product you’re building. This, of course, imparts a bias which can undermine your initial choice of objective.

Here’s a view of how first-principles problem solving unfolded when we first pivoted our company:

I like this particular because it has two clear messages:

1. It’s clear that sacred cows are abound at every step.

   The sacred cow in the above example was that I wanted the existing business to work, so I’d already started the framing of the problem under the assumption that choice of business was not a parameter that I could play with.

2. It makes clear the utility of revising one’s objective.

   Without revision, we would never have executed the pivot.

Final comments

One of my advisors frequently reminds me that, though I am naturally inclined to think of everything from first principles, first principles thinking is not always the most optimal solution. Oftentimes copy-pasting an existing solution can lead you to a more optimal solution, particularly if you lack the data to arrive at the correct conclusion from first principles yourself. Moreover, this sort of reasoning by analogy is often much faster. (Of note: pattern recognition and wisdom are the mirror skill sets, which is articulated well in the quote below:)

“Frequently, when I am faced with what would appear from the outside to be a challenging problem, the grinding mental computation is somehow circumvented, rendered, as if by magic, unnecessary. The solution comes effortlessly, seamlessly, and seemingly by itself. I seem to have gained in my capacity for instantaneous, almost unfairly easy insight. Is it perchance that coveted attribute … wisdom?”

- Elkhonon Goldberg, The Wisdom Paradox

But I’ve found that, in instances where you are indeed doing something entirely novel — operating at the forefront of scientific understanding (research), for instance; or building a product that has never been seen before — first principles thinking, while it should never be your default means of operation, is the only way to do something truly novel. But of course, apply sparingly.

1. As much as [we] physicists want you to believe everything comes from physics, that’s hardly a practical worldview. One can only explain two elements in chemistry through basic quantum mechanics principles (hydrogen and helium, sort of).