Paradoxes of Rationality

We like to think that if we were only rational enough, the right choice would fall out cleanly. Most of the time it does. But philosophers have found a handful of situations where being careful and reasonable does not settle things — where two principles that each look obviously correct give flatly opposite advice. These are paradoxes of rationality, and they are worth studying precisely because they resist the easy answer.

The payoff is humility with teeth. Seeing that smart people still disagree about these cases teaches you to hold your own reasoning a little more lightly, to separate a principle from the label 'rational' you have stapled to it, and to notice when an argument's force comes from a hidden assumption. That skill transfers far beyond puzzles.

Two principles that can collide

Much of decision theory rests on two intuitive ideas. The expected-value principle says: weigh each option by how good its outcomes are and how likely they are, then pick the option with the best weighted average. The dominance principle says: if one option is at least as good as another no matter what happens, and better in some case, choose it — the state of the world is already fixed, so pick the action that wins in every column. Both sound unarguable. The trouble is that in a few carefully built situations they point in opposite directions.

When a predictor enters the room

The collision tends to appear when the setup involves a predictor — something that has already anticipated your choice. Once your decision is correlated with a past event you cannot change, the two principles come apart. Dominance treats the past as fixed and irrelevant to your choice. Expected value notices that your choice is evidence about that fixed past, and lets that evidence steer you. Deciding which principle deserves priority is the heart of the debate.

The limits of pure reason

A second kind of paradox is gentler: sometimes reason simply runs out of reasons. If two options are exactly, perfectly balanced, a chooser who insists on a decisive reason before acting has none — and may do nothing at all. This exposes that rational agency needs more than reasons for options; it needs a way to act when reasons tie.

Here is Newcomb's Problem. In front of you are two boxes. Box A is transparent and holds $1,000. Box B is opaque. You may take both boxes, or take box B alone. Yesterday a highly reliable predictor guessed what you would do. If it predicted you would take only box B, it put $1,000,000 inside it. If it predicted you would take both, it left box B empty. The prediction is already made; the boxes are already filled.

Dominance reasoning says: the money is set. Whatever is in box B, adding the visible $1,000 can only help — so take both. Expected-value reasoning says: people who take one box almost always find a million, and two-boxers almost always find a thousand, so taking one box has the far higher expected payoff — take only box B. Each argument is clean. They disagree completely. That disagreement, not either answer, is the paradox.

It is tempting to declare the puzzle a cheat: 'A perfect predictor is impossible, so ignore it.' But the paradox does not need perfection. Suppose the predictor is merely right 90% of the time — a modest, believable accuracy. Run the expected value: one-boxing still comes out far ahead of two-boxing, yet the dominance argument is completely untouched, because the boxes are still already filled. Weakening the predictor removes the science-fiction flavor and leaves the conflict exactly where it was, which is why the problem has stayed alive.

Newcomb's Problem is named for William Newcomb, a physicist at the Lawrence Livermore National Laboratory, who devised it in the 1960s. It reached a wide audience through the philosopher Robert Nozick, whose 1969 paper 'Newcomb's Problem and Two Principles of Choice' laid out the clash between the dominance principle and expected-value maximization. Nozick reported that when he described the case to people, opinion split sharply, with each camp finding the other's choice almost unintelligible — and that split has largely persisted.

The problem became a touchstone in decision theory, helping to motivate the divide between 'causal' decision theory (which leans toward taking both boxes, since your choice cannot cause the boxes to change) and 'evidential' decision theory (which leans toward one box, since your choice is evidence of the prediction). Both frameworks are seriously defended in the literature today. That is the honest state of play: not a solved riddle but a live fault line, valuable because it shows two respectable notions of rational choice pulling apart.

• 'There's an obvious right answer and the other side is just confused.' Trained decision theorists remain split; the confidence usually comes from not stating the opposing principle fairly.
• 'The predictor has to be magic.' The conflict survives with an imperfect, believable predictor, so appeals to impossibility do not dissolve it.
• 'Buridan's ass proves animals are irrational.' It is a thought experiment about a purely reason-governed chooser, not a claim about real donkeys, who simply pick one pile.
• 'Rational just means following expected value.' Dominance is equally intuitive; the puzzle is exactly that two reasonable standards of rationality disagree.

Whichever box you would take, write the single strongest sentence for the opposite choice — good enough that a thoughtful person could hold it — then notice how hard it is to call your own choice the uniquely rational one.

Think Like a Maester: When two sound principles disagree, the discipline is to name both fairly before you pick a side.

1. What makes a situation a paradox of rationality rather than just a hard choice?
2. State the dominance principle and the expected-value principle as they apply to Newcomb's Problem.
3. Why does weakening the predictor to 90% accuracy fail to dissolve the paradox?
4. In Buridan's ass, what exactly stalls the donkey, and what small addition prevents starvation?
5. Who formulated Newcomb's Problem, and who popularized it in 1969?

A paradox of rationality is a case where two principles that each look obviously correct recommend opposite actions. Newcomb's Problem, devised by the physicist William Newcomb and popularized by Robert Nozick in a 1969 paper, is the classic: the dominance principle says take both boxes because the money is already fixed, while expected-value reasoning says take one box because that choice is strongly correlated with a million-dollar prediction. Weakening the predictor does not settle it, and the split runs deep enough to help define causal versus evidential decision theory — an unresolved debate, not a solved riddle. Buridan's ass adds a quieter lesson: a chooser who demands a decisive reason for every act can stall when reasons tie, showing that rational agency needs a way to break ties as well as reasons to prefer. The maester's move is not to crown a winner but to state each side fairly and hold your own view with earned humility.

• Decision-Making Under Uncertainty
• Paradoxes of Self-Reference and Surprise