Can Active Behavior Rewrite Fate? Reinforcement-Learning Agents Learn to Lie Flat in the Talent-versus-Luck Model

Whether success comes from talent or from luck is an old debate. The Talent-versus-Luck (TvL) model drew wide attention with a blunt answer: luck decides who gets rich, and the richest are rarely the most talented. But its agents are passive.They stay put and take whatever luck reaches them, while real people move toward chances and away from trouble. To our knowledge we are the first to let TvL agents act and learn, using Double DQN, in a two-by-two design that crosses learning (off or on) with the kind of luck (asymmetric or symmetric). Two results follow. The new one is behavioral. Under TvL’s original luck, where a chance is more likely to cost than to pay, the agents learn on their own to avoid opportunity altogether, a model version of “lying flat”. It is a rational choice rather than laziness, and the proof is that the same reward makes them seek events once luck is made fair. The learning also gives the shape of that retreat, which the arithmetic does not: the agents keep moving, meet almost nothing, and talent stops leaving any mark on wealth. The second result is that TvL’s core findings survive. The richest are still the luckiest rather than the most talented, and talent predicts wealth only once luck exposure is held fixed, because talent is simply the rate at which luck is turned into gains. Within the bounded agency modeled here, acting protects what an agent starts with but cannot make effort beat luck.