Picture a small, dull room with two boxes on a table. One is clear and holds a thousand dollars you can see. The other is opaque. Someone has already decided what goes in it, and that someone is very good at guessing what you will do. You may take the opaque box alone, or take both. Nothing about the table changes while you think. Only your reasoning does, and the reasoning splits in two.
The setup and why it is hard
This is Newcomb's problem. In the standard version, box A is transparent and always holds $1,000. Box B is opaque. If the predictor expected you to take only B, it holds $1,000,000. If the predictor expected you to take both, it holds nothing. The prediction is made first and the choice second, and the predictor is right with near-certainty (Wikipedia).
The payoff table is short. Both boxes against a correct prediction of both pays $1,000. Both boxes against a prediction of only B pays $1,001,000. Only B against a prediction of both pays nothing. Only B against a correct prediction pays $1,000,000. When the predictor is near-certain, one-boxers collect about a million per game and two-boxers about a thousand. That gap is what makes the problem sting.
The trouble is that two respectable principles read this table differently. Each one feels like plain sense until the other speaks.
The case for taking both boxes
The first argument is almost bookkeeping. By the time you choose, box B is full or empty. Your choice cannot reach back and alter it. If B is full, taking both gets you $1,001,000 instead of $1,000,000. If B is empty, taking both gets you $1,000 instead of nothing. Either way, the second box adds a thousand dollars. This is the strategic dominance principle, and it is hard to argue with on its own terms.
Causal decision theory gives it a formal home. The Stanford Encyclopedia entry says rational choice must attend to an act's causal consequences, not merely to evidential correlations, and that the probabilities used should depend on the option causally (Stanford Encyclopedia of Philosophy). On that view, taking one box does not make it any more likely that B was filled. The money is already where it is.
The case for taking only box B
The second argument counts what tends to happen. If the predictor is nearly always right, people who take only B nearly always leave with a million. People who take both nearly always leave with a thousand. Expected utility is a probability-weighted average of outcomes, and here the probabilities depend on the act. Weighted that way, one-boxing wins once the predictor is reliable enough.
Robert Nozick is credited with first analyzing the problem, in 1969, and the sources say he defended one-boxing in the limit case, where the predictor is infallible. They do not lay out his argument in detail. Martin Gardner then brought the puzzle to a wide audience in his Scientific American column. Sources disagree on the date, giving March 1973 in one place and March 1974 in another. Many solutions focus on the limit case, because very high probability can seem, for decision purposes, no different from certainty. Whether that holds is part of the quarrel.
Why people still split
Nozick observed that people divide almost evenly, with each side thinking the other is being silly. Professional philosophers show the same split. In the 2020 PhilPapers Survey, which covered 1785 English-speaking philosophers, 39.0% chose to take both boxes and 31.2% chose one box. Two-boxing leads, but not by enough to call the matter settled.
Three questions do most of the work. Does my choice causally affect what is in box B, or does it only correlate with it through the predictor's forecast? The causal theorist says the contents are fixed, so the correlation is beside the point. The one-boxer says the correlation is the whole situation.
Next, how reliable must the predictor be before the answer changes? Does near-certainty differ from certainty in a way that matters? The sources do not give the causal theorist's full reply for the merely near-certain case, so that reply stays open here. One more caution comes from Wolpert and Benford, who argue that paradoxes like this arise when not every relevant detail of the setup is specified, such as how the prediction is made.
What the record supports is modest. Both answers have serious defenders, and claiming either is obvious ignores the evidence. The better habit is to decide which question you think the table is really asking, then see which box that question points to.
Sources
- Causal Decision Theory (Stanford Encyclopedia of Philosophy)
- Newcomb's problem - Wikipedia
- Decision Theory (Stanford Encyclopedia of Philosophy)
- Newcomb's paradox - HandWiki
- PhilPapers Survey 2020
- Sentience Institute | Simulations and Catastrophic Risks
- 24.118_S19 Paradox and Infinity, Lecture Note 10: Newcomb's Problem | Paradox and Infinity | Linguistics and Philosophy | MIT OpenCourseWare
- Santa Fe Institute
Marcus Adeyemi-Holt is a house writer for Deeperpectives, where he examines how ordinary routines reveal larger patterns in work, memory, and place. He draws on years of reporting on municipal records and oral histories to trace the quiet forces that shape everyday life.
