To take satisficing seriously, you first have to kill an idol: the belief that with enough discipline you could optimise, and satisficing is just the lazy shortcut you settle for. That’s wrong. For an enormous class of real decisions, optimising isn’t merely hard — it is impossible in principle, no matter how smart, patient, or well-resourced you are. This lesson makes that concrete on three fronts: the problems explode too fast to search (combinatorial explosion), the options can’t even be listed, and thinking itself costs something you’re trying to conserve. Once you feel all three, satisficing stops looking like a compromise and starts looking like the only move on the board.
Before you read — take a guess
A chess player wants to play the truly optimal move: the one that leads to the best outcome assuming perfect play by both sides forever. Roughly how feasible is computing that move for a midgame position?
Combinatorial explosion: the numbers detonate
The first wall is arithmetic. Many decisions are combinatorial: the number of possible answers grows not by adding but by multiplying as the problem gets bigger — and multiplication runs away from you with terrifying speed.
Take the travelling salesman problem: visit N cities once each and return home by the shortest total route. To optimise — guarantee the single shortest tour — you must, in the worst case, compare all the distinct orderings. With 5 cities that’s a handful. With 15 it’s already over 43 billion. With 25 cities the number of tours exceeds the number of seconds since the Big Bang by a wide margin. Nothing changed except adding a few more dots on a map, yet the search space detonated.
| Problem size (N) | Number of arrangements to check | Feel |
|---|---|---|
| 5 cities | 12 (½ of 4!) | trivial |
| 10 cities | ~181,000 | a computer’s snack |
| 15 cities | ~43,600,000,000 | now it hurts |
| 20 cities | ~6 × 10¹⁶ | supercomputer territory |
| 25 cities | ~3 × 10²³ | more than seconds since the Big Bang |
| Chess (game tree) | ~10¹²⁰ | more than atoms in the universe |
This is combinatorial explosion, and it is not a failure of effort or hardware. It’s the mathematics of the problem itself. A brute-force optimiser doesn’t get slow as N grows; it hits a wall where the computation cannot finish before the sun burns out. And crucially, the decisions we care about — routing, scheduling, packing, choosing a sequence of career moves, planning several steps ahead in anything — are shot through with this structure.
Why 'just use a computer' doesn't save you
Computers don’t beat combinatorial explosion by optimising faster — they beat it by refusing to optimise. Chess engines prune whole branches, evaluate positions with learned heuristics, and stop searching at a “good enough” depth. Route planners use approximations that get close to the shortest tour without proving it’s shortest. Even our most powerful machines are satisficers wearing a lab coat. The best they promise is “very good, fast,” not “provably best.”
Unlistable options: you can’t rank what you can’t see
The second wall is subtler and, for human decisions, more important. The optimiser’s whole procedure begins with a step that sounds innocent: consider all the options. But for most real choices, the full set of options cannot be enumerated at all.
What are all the possible careers you could pursue? All the people you could marry? All the business ideas you could start, all the ways you could spend the next decade? These aren’t big lists — they’re open, generative, and partly unknowable. Many of the best options don’t exist yet (the job title, the company, the person you haven’t met). You cannot compute the maximum of a set you cannot write down. The optimiser’s first step is already impossible.
This is where bounded rationality bites hardest, and it connects straight to your circle of competence (a prerequisite for this course): the reason you can’t optimise outside what you understand is that outside your competence, you can’t even see the option set clearly — you don’t know what you don’t know. Optimisation assumes a god’s-eye view of a closed menu. Reality hands you a fog and a flashlight.
The menu you can see vs. the menu that exists
Choosing a cereal from ten boxes on a shelf is a rare, tame decision: the options are finite, visible, and fixed. You could, in principle, optimise. But choosing a career, a partner, or a strategy is nothing like that shelf — the options are unbounded, many don’t exist yet, and the set changes as you move through it. Treating an open-ended, generative decision as if it were a cereal shelf is one of the most common ways smart people paralyse themselves.
Why is 'choose the optimal career' fundamentally different from 'choose the cheapest of ten visible cereal boxes'?
The cost of computation: thinking isn’t free
The third wall is the one classical theory ignored entirely: deliberation has a cost. Every minute spent gathering information, comparing options, and running the numbers is a minute (and often money, and attention, and emotional energy) you don’t get back. Time spent deciding is time not spent doing — and sometimes the option you were analysing expires while you analyse it.
This reframes the whole problem. The “rational” thing is not to compute the best answer regardless of cost. It’s to account for the cost of computing. A decision procedure that finds a slightly better answer but takes ten times as long can be worse, all things considered, than a quick rule that gets you 95% of the way there. This is exactly the logic of the value of information (another prerequisite): more searching is only worth it while the expected improvement it buys exceeds what the search costs. Past that point, gathering more is not diligence — it’s waste.
Herbert Simon called the ideal that ignores this “Olympian” rationality — a god’s-eye optimiser with unlimited computation — and argued it was useless as a model of any real agent, human or machine. What agents actually have is “procedural” rationality: being sensible about the process of deciding, given that the process itself is scarce and costly.
Simon's scissors
Simon’s most famous image: rational behaviour is shaped by two blades of a pair of scissors — the cognitive limits of the mind, and the structure of the environment. You cannot explain how someone cuts by studying one blade alone. A decision that looks “irrational” against the Olympian ideal is often perfectly sharp once you account for the real limits and the real world it’s cutting through. We’ll meet the second blade again in Lesson 4, where the environment’s structure is exactly what makes simple rules work.
Three walls, one conclusion
Put the walls together and the verdict is inescapable. Optimisation fails not because we lack willpower, but because:
- The search space explodes — for combinatorial problems, no achievable computation can check every option (chess, routing, scheduling, multi-step planning).
- The options can’t be listed — for open-ended choices, there is no closed menu to take the maximum of (careers, partners, strategies, business ideas).
- Computation itself is costly — even when optimising is possible, the time, money, and attention it burns can make it the wrong thing to do.
So the perfect optimiser isn’t a high standard we fall short of. It’s a fiction that describes no real decision-maker, not even a supercomputer. And that’s liberating, not depressing: it means the goal was never “find the best.” The goal is to make a good decision at a reasonable cost — which is precisely what satisficing is engineered to do, and what the rest of this course builds out.
Sort each decision by whether a true OPTIMISER could, in principle, find the single best option — or whether optimisation is fundamentally blocked (explosion, unlistable options, or prohibitive cost).
Place each item in the right group.
- Choosing the highest-interest account from 5 listed savings accounts
- Selecting the single best career you could ever have pursued
- Finding the provably shortest route through 30 delivery stops
- Choosing the objectively ideal life partner from everyone alive
- Buying the cheapest litre of one milk brand across 3 nearby shops
- Computing the perfect chess move from a midgame position
- Picking the cheapest of 8 clearly-priced flights shown on one screen
Simon distinguished 'Olympian' rationality from 'procedural' rationality. What's the key difference?
The one thing to remember
Optimisation isn’t just impractical — for combinatorial problems, open-ended option sets, and costly-to-compute decisions, it’s impossible in principle. No achievable amount of effort or hardware finds “the best,” because the search explodes, the menu can’t be listed, or the computing itself costs more than it’s worth. The perfect optimiser is a fiction. That’s why we need a different kind of rationality — one that sets a bar for “good enough” and stops. That’s satisficing, and it’s next.