Human planning in stochastic environments

The world is stochastic, making planning difficult. Despite the ubiquity of stochasticity in real-world environments, it remains an open question how people effectively balance the cognitive costs of planning against its potential benefits when faced with stochasticity. To study this, we designed a planning task where participants face one of three forms of stochasticity commonly encountered in the real world: reliability, volatility, and controllability. We find a robust pattern across all three manipulations: as stochasticity increases, people reduce their planning effort as measured by first-choice response times. To understand the processes underlying this effect, we developed several computational cognitive models to account for participants' choices. We find that rather than calculating expected values optimally, people chose a simpler strategy, acting as if the world were deterministic, a phenomenon known as determinizing. Consistent with our response time findings, we found that planning depth estimated in the winning model decreases monotonically with increasing stochasticity. Our work highlights the often overlooked role of stochasticity in human planning and its impact on planning strategy and effort. Moreover, it reveals the limitations of studying stochasticity solely through single-shot decisions.