Abstract for:Can learning and markets solve challenges to dynamic decision making?
Dynamic tasks, where current outcomes depend both on current and past choices, are ubiquitous. Significant discrepancies between individual and optimal choice exist in these settings. Yet dynamic tasks are inherently complex and even best algorithms require long training periods. Moreover, market forces can alleviate individual-level learning failures. In a task highlighting the tension between short and long-term outcomes we experimentally remove challenges to learning due to ambiguity, large state space, exploration, and memory and updating. We find that none explains the observed learning failures. Instead inability to trade off short-term gains against investing in promising system states is a persistent challenge. The gap between human learning and algorithmic alternatives widens as the task is simplified. Surprisingly, markets exacerbate failures and may induce wrong learning.