06 / 08Think

First principles, and working backward

A household wants a smaller grocery bill and starts by trying a cheaper store, because that is what people do. A student plans a semester by copying last year's schedule. A team builds next year's budget from this year's, plus five percent. Starting from what already exists is fast, and often fine. It also carries along every assumption the old version made. This lesson offers two other starting points. One is the bottom: what is known for sure, with borrowed assumptions set aside, rebuilt from there. The other is the end: the goal, or a clear picture of failure, traced back step by step to today. In a class both show up in proofs and problem sets. At work they show up when "the way we've always done it" stops working.

In five minutesPick one routine that costs you time or money. List three assumptions behind how you do it. Mark each one "checked" or "borrowed." For one borrowed one, ask how you would find out.

Method · First principles and working backward  /  Philosophy · Science

↺From last time · Common traps

Answer these from memory before you start. Tap to check.

Why ask whether you'd choose something fresh today?

Money and time already spent are gone whatever you choose next. The fresh-start question takes them off the scale, so the decision rests on what each option will cost and give from here.

Why look up a count instead of trusting a vivid example?

Vivid cases come to mind easily, and how easily something comes to mind feels like how often it happens. A count measures how often it actually happens.

Why use a written checklist if you already know the names of the biases?

Knowing the names doesn't run the checks, and training effects carry over to real decisions only partly. A list used at the moment of deciding does the checking whether or not you remember to.

01Where it comes from

Aristotle opens his Physics by saying we know a thing only when we know its first principles and have carried the analysis down to its simplest elements (see Voices). Greek geometers also worked the other way. Pappus of Alexandria, writing about 320 CE, described analysis as assuming the answer found and tracing back to something already known (see Voices). In his Discourse on Method (1637), René Descartes set aside every belief he could find a reason to doubt, to find one first principle he could not. George Pólya's How to Solve It (1945) taught working backward as a problem-solving method and traced it to Pappus [1]. The form that starts from failure is often credited to the mathematician Carl Jacobi, and the investor Charlie Munger made it famous in a 1986 talk [2]. The record for Jacobi's words is thin (see Misattributed) [6]. Research has not tested "first-principles thinking" by that name. Studies of learning from compared cases, and of how experts see problems, come closest [3][4][5].

Also called First-principles thinking, analysis and synthesis, working backwards, inversion, reasoning from the ground up.

Down to the simplest elements

For we do not think that we know a thing until we are acquainted with its primary conditions or first principles, and have carried our analysis as far as its simplest elements.

Aristotle, Physics Book I, chapter 1 (184a) · tr. R. P. Hardie and R. K. Gaye

Aristotle is opening a study of nature and says the first task is to find its principles. He adds a point often left out. We start from what is clear to us, which is usually a confused whole, and work toward what is clear in itself. The principles come at the end of the analysis, not at the start. A child calls every man "father" before learning to tell them apart.

Read it in full

Solution backwards

we inquire what it is from which this results, and again what is the antecedent cause of the latter, and so on, until by so retracing our steps we come upon something already known or belonging to the class of first principles, and such a method we call analysis as being solution backwards

Pappus of Alexandria, Collection, Book VII, on the Treasury of Analysis Quoted in Thomas L. Heath, A History of Greek Mathematics, volume II, page 400 · tr. Thomas L. Heath

Pappus is introducing a set of geometry books for solving problems. Analysis assumes the thing sought as if it were already done, then asks what it follows from, one step at a time. Synthesis then runs the chain forward as a proof. In this one sentence the two halves of the lesson meet: working back from the goal ends at first principles.

Read it in full

02The method

First principles and working backward

  1. Write the goal in one sentence, with how you will know it has been met.
  2. List the assumptions in the current plan or routine. For each, ask: do I know this is true, or did I borrow it from habit or from someone else?
  3. Keep only what you can check: a measured cost, a rule, a fact. Rebuild the plan from those.
  4. Work backward from the goal. Ask what must be true just before it is met, then just before that, until you reach something you can do today.
  5. Invert. Ask what would guarantee failure. List those causes, then check that the plan avoids each one.

03An example

What it looks like done

Worked on A household's weekly grocery bill, about $210 for two adults

The goal

Spend about $160 a week on groceries without eating worse. Measured by receipts over four weeks.

Questions, then answers from memory

1. We need one big trip a week. Checked? Borrowed. It's how my parents shopped.

2. The big store is cheapest. Checked? No. I've never compared receipts.

3. We use what we buy. Checked? No. Something goes in the trash most weeks.

4. Brand names taste better. Checked? Only for coffee.

5. What is checked: last month's receipts. Produce and snacks were the largest lines.

Working back

  • Week four at $160 needs a plan that already works in week three.
  • That needs a shopping list built from meals, not from the aisles.
  • That needs a meal plan for the week, written on Sunday.
  • Today: write four dinners and list only what they need.

What would guarantee failure

  • Shopping hungry. No list. Buying produce with no meal planned for it. Restocking snacks out of habit.
  • The current routine does three of the four.

04Try it · 25 minutes

For a class

Take one problem from your next problem set, or one claim from a textbook chapter. Fifteen minutes. Write what is given and what is asked. Work backward from the answer: what would you need to know just before it? Keep going until you reach something given. Then read Part Four of Descartes's Discourse and mark where he sets doubtful beliefs aside to find one he can't doubt.

Open it in the reader: A Discourse on Method, Part Four

For work and life

Pick a recurring cost or routine: a weekly grocery bill, a monthly report, a commute. Twenty minutes. List every assumption behind how it's done now, and mark each as checked or borrowed. Rebuild it from the checked ones only. Then write five ways to make it fail for certain, and see which of them the current version already does.

The blank page for this is in the Think workbook.

05Where it goes wrong

  • Throwing out what works. Descartes doubted his own opinions, and wrote in the same Discourse that remaking a whole state from scratch would be unwise. Question each assumption. Keep the ones that check out.
  • Calling a guess a first principle. A strong opinion is not bedrock. A first principle is something you can check, or can't find a reason to doubt.
  • Stopping at the list of failures. Naming ways to fail is half the step. Change the plan for each one.
  • A backward chain that ends nowhere. The chain should end at something you can do today. If it ends at something vague, a step is missing.
  • Rebuilding everything, every time. It takes time. Keep it for the problem where the usual way has stopped working.

06Research

What research has measured

Comparing cases to find the principle

Shown that comparing two or more cases side by side helps people learn the principle they share better than studying cases one at a time.

The numbers and sources
Effect
Across 336 tests in 57 experiments, case comparison led to more learning than sequential or single cases, unrelated cases, and traditional instruction, d = .50. Effects were larger when learners looked for similarities between the cases and when a principle was given after the comparison [3].
Population
57 experiments in laboratory and classroom settings [3].
Caveats
Effects were larger on immediate tests than on delayed ones [3]. Comparing cases is one route to principles. It is not "first-principles thinking" as popular writing uses the term.

How experts see a problem

Early studies suggest experts sort problems by the principle that solves them, while novices sort them by surface features.

The numbers and sources
Effect
In sorting tasks and think-aloud protocols, physics experts grouped problems by the physics principles involved. Novices grouped them by the problems' literal features [4]. Experts recognize patterns that guide them quickly to the relevant knowledge [5].
Population
Physics experts and novices in four experiments. The abstract does not give counts [4].
Caveats
Small studies in one field. The experts' advantage reflects years of knowledge, so the studies don't show that sorting by principles can be taught directly [4][5].

07Check yourself

Answer from memory first. The next lesson asks these again.

Why mark each assumption as checked or borrowed?

Most plans copy earlier ones, and the copied parts are rarely tested. Marking them shows which parts rest on evidence and which rest on habit, so you know what to check before rebuilding.

Why work backward from the goal instead of forward from today?

Forward planning asks what could be done first, and there are many options. Backward planning asks what must be true just before the goal, which narrows each step until it reaches something doable now.

What does asking how to guarantee failure add?

Ways to fail are often easier to name than ways to succeed. Listing them turns vague risk into specific causes, and you can check the plan against each one.

Sources
  1. Pólya G. How to Solve It. Princeton: Princeton University Press; 1945. Princeton University Press, 1945
  2. Munger C. How to guarantee a life of misery. Talk at the Harvard School, Los Angeles, June 13, 1986. Transcript: https://jamesclear.com/great-speeches/how-to-guarantee-a-life-of-misery-by-charlie-munger
  3. Alfieri L, Nokes-Malach TJ, Schunn CD. Learning through case comparisons: A meta-analytic review. Educational Psychologist. 2013;48(2):87-113. https://doi.org/10.1080/00461520.2013.775712
  4. Chi MTH, Feltovich PJ, Glaser R. Categorization and representation of physics problems by experts and novices. Cognitive Science. 1981;5(2):121-152. https://doi.org/10.1207/s15516709cog0502_2
  5. Larkin J, McDermott J, Simon DP, Simon HA. Expert and novice performance in solving physics problems. Science. 1980;208(4450):1335-1342. https://doi.org/10.1126/science.208.4450.1335
  6. Van Vleck EB. Current tendencies of mathematical research. Bulletin of the American Mathematical Society. 1916;23(1):1-13. https://doi.org/10.1090/s0002-9904-1916-02863-1

MVIII puts the day in front of you each morning. Get MVIII