Why can two smart choices add up to a bad outcome?

Because what's best for each person on their own can be worse for both together. That's the prisoner's dilemma, a classic puzzle of game theory. Two partners can either cooperate or betray each other. Betraying is the rational move for each of them, yet if both cooperated, each would end up better off.

Here's the classic story. Two gang members are arrested and questioned separately. If both stay silent, each gets one year. If one testifies and the other stays silent, the snitch walks free and the other gets three years. If both testify, each gets two years. Now think like prisoner A. If B stays silent, testifying gets you freedom instead of a year. If B testifies, testifying gets you two years instead of three. Either way, betrayal wins, and B reasons exactly the same. So both testify and serve two years each, when silence would have cost them just one.

The puzzle was invented in 1950 by Merrill Flood and Melvin Dresher at the RAND Corporation, and Albert Tucker later gave it the prison story. The hopeful twist came later: when the game is played over and over, cooperation can win. In Robert Axelrod's famous computer tournament, the champion was "tit for tat", just four lines of code, which starts by cooperating and then simply copies what the other player did last time.