Deciding Under Deep Uncertainty: Choosing Well When You Can't Know the Odds

A weather app tells you there’s a 70% chance of rain. Annoying, but easy: you know the odds, so you weigh a light umbrella against a soaked commute and decide. Now a different question. Which job platform, if any, will dominate hiring in 2040? What’s the probability your industry still exists in its current shape in fifteen years? You don’t have a number. Nobody does. And here’s the trap this whole course exists to defuse: those two questions feel similar — both are “the future is uncertain” — but they are made of completely different stuff, and the tools that crush the first one fall apart on the second.

The first is risk: a gamble whose odds you actually know. The second is what we’ll call deep uncertainty: a gamble whose odds are unknown, contested, or genuinely unknowable. You’ve spent this whole latticework learning to tame risk — expected value (probability times payoff), base rates, fat tails, margin of safety. This is the honest sequel. It asks the question every one of those tools quietly needs you to already have answered: what do you do when you can’t put numbers on the odds at all?

Why expected value isn’t enough anymore

Expected value is a beautiful machine, and it has served you well: list the outcomes, multiply each payoff by its probability, sum, and pick the biggest number. But look at what the machine eats. It runs on probabilities. Feed it good odds — a fair die, a mortality table, a well-understood market — and it hums. Feed it decisions where the odds themselves are unknown, and it doesn’t stop working; it does something worse. It keeps producing a crisp, confident answer built on numbers you invented, and hands you false precision: a spreadsheet that looks like knowledge and shatters the instant reality lands in a cell you never modelled.

The decisions that matter most live in exactly this territory. Which technology wins the decade. Whether a novel pandemic spreads. How a market no one has seen before will behave. Whether your one big bet holds up against a future you can’t forecast. For these, the probabilities are not just hard to compute — they are, in Frank Knight’s century-old distinction, uncertain rather than merely risky. And the strategy that wins under risk (compute the odds, optimise the average) is the strategy that gets you killed under deep uncertainty.

Robustness over optimality

Here is the whole model in a sentence: when you can’t reliably compute which action is best, stop trying to be right and start refusing to be ruined. That one move reorganises everything. Instead of a single forecast, you picture a spread of futures. Instead of the action that scores highest in your best guess, you look for the one that scores acceptably in all of them — and, above all, the one that survives even the bad ones. Because the game is played through time, and one ruin ends it: there is no comforting long-run average to collect when you’re out of the game for good.

That reframing is the seed of a whole toolkit you’ll build over the next five lessons: satisficing (good-enough beats a fragile best), minimax-regret (minimise how much you’ll kick yourself in the worst case), the ruin-avoidance / precautionary principle (never risk what you can’t afford to lose), and the practical machinery of buying back your own ignorance — reversibility, option value, redundancy and slack, scenario thinking, and the pre-mortem. None of it requires you to know the odds. That’s the point.

The map of the course

Five teaching lessons, then a final exam you can’t undo. The route up:

1. Risk vs. Uncertainty — Knight’s distinction made precise: the dice you can compute versus the future you can’t. What deep uncertainty actually is, why the two demand different tools, and how to tell which one you’re facing.
2. When Expected Value Lies — why a point-estimate average quietly assumes you know the odds, how fat tails and unknown unknowns break it, and the deepest cut of all: ergodicity and ruin — why the average of many parallel bets is a lie about the fate of one person betting through time. You’ll meet the many-worlds explorer and watch the optimiser blow up.
3. Robustness Over Optimality — the core switch: satisficing, robust decision-making, minimax-regret, and the ruin-avoidance / precautionary principle. The strategy that does acceptably everywhere beats the one that’s perfect in your single guess.
4. Buying Back Uncertainty — the practical toolkit: reversibility and option value, margin of safety, small reversible experiments, redundancy, slack and antifragility, scenario thinking and the pre-mortem. How to act well without the odds.
5. The Deep-Uncertainty Playbook — the heuristic that ties it together, and the traps that catch even experts: false precision dressed as rigour, treating deep uncertainty as mere risk, over-hedging into paralysis, and mistaking robustness for plain pessimism. Ends with a whole-course recap.

Then a Final Exam — graded, one question at a time, one-way: once you answer, it locks. No back button, no retries.

How to use this course

One habit does most of the work: before you optimise a decision, ask whether you actually know the odds — and if you don’t, switch from “what’s best?” to “what survives?” When you hit an exercise, commit to an answer in your head before revealing anything; the small sting of being wrong is what makes the idea stick. The exercises are the lesson; the prose just sets them up.

Next up: lesson 1, where we draw the sharp line between the risk you can compute and the uncertainty you can’t — and learn to tell, in the moment, which one you’re standing in.