What Changes Your Mind

Suppose you’ve spent six months and $80,000 developing a product. It will take another $20,000 to finish. Should you keep going?

Put this question at the end of a chapter about sunk costs and the reader knows what to look for. Put it inside a business and things become less convenient. Someone says you’ve come too far to stop. Someone else says you’re throwing good money after bad. Both sound like arguments. Neither tells you whether finishing is worthwhile.

A person who understands sunk costs should be able to say what’s missing: what the next $20,000 is likely to buy, and what else could be done with it. The money already spent cannot justify spending more. But it cannot, by itself, justify stopping either.

This suggests a more demanding test of learning than asking someone to explain a principle. Can they work out what would change their decision?

I’m building games and simulations to help people learn ideas about judgment. That makes this question practical for me, and makes games an answer I should be suspicious of preferring. It’s easy to imagine replacing a page of explanation with a decision. It’s harder to establish what the decision requires someone to understand.

Imagine turning the product question into an exercise. The learner can continue or cancel. If cancelling earns a point, we may have built a machine for teaching people to abandon projects. Add an explanation about sunk costs after each choice and they could acquire the vocabulary to defend this habit.

Even asking for a reason doesn’t entirely solve the problem. “Ignore sunk costs” can be memorized as easily as “cancel.” Someone might write the right sentence while still treating the amount already spent as evidence that the project should stop.

We could learn more by changing the situation. Keep the remaining cash, future costs, prospects, and alternatives identical, but change the money previously spent from $80,000 to $8,000. Should the decision change? Now keep the past spending fixed, but introduce credible evidence that customers will pay enough to make finishing worthwhile. Should it change this time?

These variations separate two things a single correct answer leaves tangled together. The learner must recognize a fact that should make no difference and a fact that should. Someone who always cancels will fail. So will someone who always perseveres. Repeating the principle won’t settle which information belongs in the decision.

Feedback could then address the particular mistake. If the learner changes their choice when only the historical spending changes, the explanation can point out that the available futures are still identical. If they refuse to reconsider when the prospects improve, it can show how they’ve turned a rule about irrelevant costs into a rule against continuing.

The useful thing about choosing before seeing an explanation is that it leaves something to compare the explanation with. There is a record of which facts mattered to you. Without that record, “Yes, that makes sense” tells you little about whether the explanation corrected anything.

But this is already assuming quite a lot. A beginner may not know how to compare the alternatives. They may not understand the difference between revenue and profit. Asking them to reason independently could mean asking them to struggle with several unfamiliar things at once, while we congratulate ourselves for making them think.

There’s no reason they must discover the whole method. In a study by David Klahr and Milena Nigam, children received either explicit instruction or an opportunity to discover how to design a fair experiment. More learned the method through instruction. Among those who learned it, the instructed children performed as well as the discoverers on a later task evaluating science-fair posters. That doesn’t settle how to teach adult judgment, but it gives us a concrete reason to question the idea that being told how something works prevents independent use. (PubMed)

A worked example might be exactly what our beginner needs. Show how to compare the available futures, including why a tempting fact has been left out. Then give them a case where the comparison is partly completed. Later, leave them to decide what belongs in it.

How much explanation is enough depends on where the learner gets stuck. If they cannot interpret the figures, explain the figures. If they can do the calculation but include an unrecoverable expense, examine that choice. And if they can solve the problem whenever it’s labelled “sunk costs,” giving them another explanation of sunk costs may miss the difficulty entirely.

The initial hypothesis needs refining. A learning experience shouldn’t require learners to do all the thinking an explanation does. Some of that thinking is worth demonstrating. But it should eventually hand over each part they will need to do without help. The important question is which decisions the teaching is still making for them.

One of the last to hand over is the decision to use the idea at all.

After ten exercises about whether to abandon a product, someone may become excellent at that exercise. To find out what they’ve learned, we might later present a different problem. They’ve bought a nonrefundable concert ticket, but now expect to enjoy an evening at home more. What should they consider? Put this among unrelated questions, without a chapter heading announcing the relevant principle.

Then change the ticket: it can be resold until five o’clock. Now ignoring its value would be a mistake. Going to the concert means giving up the resale money. A learner who has picked up “ignore what you paid” has to distinguish the historical price from an option that still exists.

The change of setting tests whether they can find the same relationship under different details. The refund condition tests whether they can notice when a superficially similar situation has changed in a relevant way. Both matter. Otherwise an idea can become either too narrow to use or broad enough to misuse everywhere.

Even this gives us only limited evidence. The person still knows they’re doing an exercise. Nobody is disappointed with them for staying home. They haven’t spent six months telling friends how successful their product will be. Recognizing a reason in a short problem and allowing it to govern a decision you care about remain different accomplishments.

Games introduce another difficulty: they need rules for what happens next.

Suppose our simulation rewards a sound decision to finish the product by making the launch succeed. It has made the lesson satisfying by removing something essential from it. Good decisions can turn out badly. If every sensible choice wins, we could teach the learner to expect vindication, or to regard an unlucky result as proof that their reasoning was wrong.

Random outcomes would help represent that uncertainty, but they wouldn’t automatically provide good feedback. A learner could make a careless bet, win, and feel confirmed. The exercise would need to examine what was reasonable before the result appeared: the evidence available, the alternatives considered, the risks they could afford, and the assumptions doing most of the work. Recording those judgments beforehand would make them harder to rewrite after the outcome.

A simulation can also replay a choice many times to show the range of possible results. This is a real advantage over a project that takes a year. But every replay comes from rules someone wrote. A strategy that reliably wins may reveal more about the designer’s assumptions than about business. Part of learning from a simulation should therefore be asking what it leaves out, and whether the decision survives different plausible assumptions.

By this point the distinction between books and games seems less useful than it did at the start. A book can stop before the solution and ask you to make the comparison. A discussion can reveal an assumption you didn’t know you were making. Writing can force you to state a prediction precisely enough to revisit it. A simulation can vary one fact while holding others still. A real project can expose facts and consequences its designer never thought to include.

The choice among them depends on what thinking we need to make possible, and what evidence we need that it happened. A game that rewards guessing the author’s intention has preserved much of the original problem. It has simply made guessing more entertaining.

To claim that an experience teaches useful judgment, we’d need evidence beyond completion. A later, unfamiliar problem without a hint would be a start. Comparing performance with people who received a good explanation would help establish whether the extra machinery earned its place. Evidence from consequential decisions would be stronger still, though much harder to interpret. We shouldn’t turn the difficulty of measuring improvement into permission to assume it.

Perhaps understanding can never be made compulsory in the literal sense. People can guess, imitate, and learn the peculiarities of a test. What we can do is make those substitutes less sufficient, while giving learners the help they need to move beyond them.

For the next exercise I build, I’d start with two questions: what could someone get right here without understanding, and what change to the situation would expose that? If the learner can succeed by spotting a phrase, choosing the virtuous answer, or repeating the previous move, the exercise still owes them some work. It should eventually ask them to decide what matters—and stop deciding for them.