On This Problem Rational People Do Worse

You enter a room with a supercomputer and two boxes. One is open, containing a visible $1,000. The other is sealed. The supercomputer has correctly predicted thousands of people's choices in this exact scenario. You can take both boxes or only the mystery box. Here's the twist: before you walked in, the computer predicted your choice. If it predicted you'd take only the mystery box, it put $1 million inside. If it predicted you'd take both, the mystery box is empty. The boxes are already set up. The computer isn't trying to trick you — its only goal is accurate prediction.

When philosopher Robert Nozick introduced this problem, he noted that «to almost everyone, it is perfectly clear and obvious what should be done. The difficulty is that these people seem to divide almost evenly on the problem, with large numbers thinking that the opposite half is just being silly.» A 2016 Guardian poll of over 31,000 people found 53.5% were one-boxers (mystery box only) and 46.5% were two-boxers (both boxes). The Veritasium audience poll of 24,000 responses skewed even more heavily: two-thirds chose one box. Yet each camp is utterly convinced the other is making an obvious mistake.

Philosophers Gibbard and Harper admitted that two-boxers fare worse, but argued they still made the rational choice — the game is simply rigged to reward irrationality. But this reveals something profound: sometimes being a rational person requires acting irrationally. The parallel with the prisoner's dilemma is instructive. In a single round, the rational act is to defect. But a rational society is full of cooperators. Rationality at one level isn't compatible with rationality at another.

This creates three scenarios where a committed two-boxer would switch to one-boxing: first, if your choice now can actually change the past (like a wormhole to the future). Second, if there are multiple trials where reputation matters. Third, if you can pre-commit before the prediction is made. In all three cases, the commitment itself changes the structure of the game. The question shifts from «what should I choose now?» to «what kind of person should I be?»

On August 29, 1949, the Soviet Union detonated its first nuclear weapon, launching an arms race that left the US with over 30,000 warheads by the mid-1960s. US Secretary of Defense Robert McNamara advocated «mutually assured destruction» — the commitment to retaliate with total annihilation if attacked. This commitment is what kept the peace. But imagine you're the US president and you've just received confirmed word that Soviet missiles have launched. If you retaliate, everyone dies — possibly triggering a nuclear winter that kills nearly everyone on Earth.

The rational act in the moment is not to launch. But the rational pre-commitment is to always launch, because that's what prevents the attack in the first place. As depicted in «Dr. Strangelove,» the Soviets built a doomsday device that automatically triggers if attacked or tampered with — including tampering by the Russians themselves. The whole point is to remove the option of changing your mind. In game theory terms, you visibly throw your steering wheel out the window in a game of chicken, forcing the other driver to swerve. Pre-commitment to a worse option produces the best outcome.