On Knowing What You Don't Know

9 April 2026 · 2 min read

epistemologythinking

There’s a particular kind of confidence that comes not from certainty, but from a clear-eyed understanding of where your knowledge runs out.

Most people treat uncertainty as a failure state — something to be minimised, hidden, or apologised for. But I think that gets the relationship backwards. Knowing that you don’t know something, and knowing why, is often more valuable than a brittle confidence that hasn’t been tested.

The Dunning-Kruger framing is too simple

The popular version of Dunning-Kruger — beginners are overconfident, experts are humble — is a useful cartoon, but it misses something. The interesting question isn’t just how much you know, but what shape your ignorance takes.

A beginner doesn’t know what they don’t know. An intermediate practitioner has a map of their ignorance — they know which questions they can’t yet answer. An expert has refined that map further: they know which of those questions are answerable, which are unanswerable, and which are not yet well-posed.

That last category is the most interesting. Some questions we treat as hard are actually confused — they need to be dissolved rather than solved.

Uncertainty as information

In financial services, there’s a concept called a confidence interval. Rather than saying “the return will be 8%”, you say “the return will be between 5% and 11% with 90% probability.” The interval is the honest part. Collapsing it to a point estimate isn’t precision — it’s false precision.

The same applies to beliefs generally. When I say I’m “pretty sure” about something, that vagueness isn’t sloppiness. It’s load-bearing. It tells me and the person I’m talking to: I hold this loosely, new information can move me.

Practical upshot

I try to maintain a rough internal categorisation of my beliefs:

  1. High confidence, high stakes — worth defending, worth updating only on strong evidence
  2. High confidence, low stakes — fine to assert without hedging
  3. Low confidence, high stakes — flag uncertainty explicitly, seek more information
  4. Low confidence, low stakes — hold loosely, don’t spend much energy here

The goal isn’t to never be wrong. It’s to be wrong in ways you can update from — and right in ways you can actually trust.

This is the kind of thing I find myself returning to. If you have a different view, I’d be interested to hear it.

← All posts