Thinking tools
The map is not the territory
Every model, metric or plan is a simplification, and the parts it leaves out can matter more than the parts it shows.
Alfred Korzybski, in a 1931 paper presented to the American Association for the Advancement of Science, reprinted in Science and Sanity (1933).
A map is useful because it leaves things out. A street map that showed every pebble would be as big and as hard to read as the city. Korzybski's point was that we reason with representations (words, numbers, diagrams, models) and then forget that they are representations. A dashboard is a map of a business. A test suite is a map of correct behaviour.
Keeping the distinction in mind tells you where to look for trouble. A map is accurate only for the conditions its makers expected. When the territory changes, or when people start steering by the map itself, the gap between the two grows quietly until something breaks.
Rejecting all models because they are incomplete leaves you with nothing to reason with. The useful habit is to keep checking the map against the ground: ask what a given map leaves out, and whether that omission matters for the decision in front of you.
Example
Google Flu Trends estimated flu levels in the United States from search queries. In February 2013, Nature reported that it was predicting more than double the proportion of doctor visits for influenza-like illness that the Centers for Disease Control and Prevention (CDC) measured from lab and clinic reports, even though the tool had been built to predict those very reports. A 2014 paper in Science by David Lazer and colleagues called this big data hubris.
Where it applies
An evaluation benchmark is a map of model quality. When a large language model (LLM) improves on the benchmark but users complain more, believe the users and go and look at real transcripts.
Think about it
Which number do you watch most closely at work, and when did you last check it against what is actually happening?
Applied it to something real? Log it.
Test yourselfWhat should you ask about any model, metric or plan you rely on?
What it leaves out, and whether that omission matters for the decision you are making. Then check it against the real thing.
Related models
- First principlesBreak a problem down to what you know is true and reason up from there, instead of copying what others have done.
- Circle of competenceKnow where your real understanding ends, and keep your big bets inside that line.
- The Feynman techniqueTest whether you understand something by trying to explain it in plain words to a beginner; where the explanation breaks, your understanding has a gap.
- Goodhart's lawWhen a measure becomes a target it stops being a good measure, because people optimise the number instead of the thing it stood for.
- EmergenceSimple local rules among many parts produce patterns at the level of the whole that no part contains or intends.