This Paradox Splits Smart People 50/50

Newcomb’s paradox—where a near-perfect predictor offers a choice between taking one “mystery” box or taking both a mystery box plus $1,000—splits people almost evenly because two reasonable ways of calculating “what you should do” treat correlation and causation differently. The setup is simple: the predictor has already placed $1,000,000 in the mystery box if it expects you to take only that box; if it expects you to take both, it leaves the mystery box empty. Since the prediction happens before you enter and your decision can’t change what’s already been arranged, the puzzle becomes a test of how people reason under uncertainty.

One camp—often called “one-boxers”—leans on evidential decision theory: the predictor’s past accuracy is evidence about what’s in the mystery box. If the predictor is highly reliable, then choosing the one-box option is strongly correlated with walking away with $1,000,000. In that view, expected value calculations favor one-boxing once the predictor’s accuracy is even slightly above random. The argument is essentially: “My choice is a signal of what the predictor already expects, and the predictor’s track record makes that signal meaningful.”

The other camp—“two-boxers”—leans on causal decision theory: the decision you make now can’t cause the contents of the already-set boxes to change. With that assumption, the mystery box’s contents are fixed, so taking both boxes guarantees an extra $1,000 on top of whatever the mystery box contains. In expected-utility terms, two-boxing dominates one-boxing because it adds $1,000 without reducing the chance of getting the $1,000,000. The paradox persists because both camps treat different probability models as the “right” one: one-boxers use probabilities tied to evidence about the predictor’s accuracy; two-boxers use probabilities tied to what the choice can causally influence.

The disagreement isn’t just math—it points to deeper questions about rationality and free will. If a predictor is perfect, then the only way to get the $1,001,000 outcome would be to be the kind of person who one-boxes in advance, then switch at the last second. That raises whether free will exists in a way that can matter once prediction is already locked in. The discussion connects this to “Why Ain’tcha Rich?”—the observation that if one-boxing is the rational strategy for maximizing money, then one-boxers should systematically win more—and to philosopher arguments (including Gibbard and Harper) that sometimes “irrationality” can be rewarded when prediction is strong.

Finally, the conversation reframes the paradox as a problem about rules, not moment-to-moment choices. If people could pre-commit—through reputation across repeated trials, through mechanisms that let choices affect the past, or through explicit commitments—then the “worse” option (one-boxing) could become the best rule to live by. The same logic is compared to deterrence and game-theory commitments in Cold War “assured destruction” (MAD): stability can depend on committing to an outcome that you hope never occurs. The core takeaway is that Newcomb’s paradox isn’t meaningless; it forces clarity about whether decisions should track evidence or causation, and what “rational” behavior means when your action is entangled with prediction.