CompCog: Human planning in stochastic environments

New York University, New York NY

Investigators

Abstract

When people plan for the future, they often do so under uncertainty. Whether deciding a morning commute, saving for retirement, or planning for college, people must make decisions about the future while adapting to randomness in their environments. Yet despite the fact that uncertainty is nearly pervasive in real-world planning, little is known about how it affects their plans. This project studies how people change their planning behavior in response to varying levels of randomness in their environment. The project addresses this question by analyzing human behavior and eye movements to develop computational models of how people plan and act under uncertainty. Predictions are tested on multiple sources of uncertainty, including uncertainty in reward, uncertainty in actions, and volatile environments. Developing a deeper understanding of how uncertainty affects planning behavior has a wide range of potential applications, not just for basic science, but also for building effective interventions to help individuals and groups tackle problems for longer decision horizons. Multi-step planning is a hallmark of decision making. Every day, people are faced with tough decisions about the future, for example individually on career paths and collectively on large-scale societal problems. In real-world planning, these decisions often come hand in hand with some level of unpredictability or uncertainty. Studying how people adapt to this uncertainty is particularly interesting in cognitive science because it can give insights to the limits of human cognitive capacity; planning in uncertain environments involves making tradeoffs between cognitive load and maximizing reward while considering several possible future trajectories. Yet this effect has often been overlooked in the field. The central question of this proposal is how people change their planning when confronted with stochasticity. A combination of human behavior, eye measurements, and computational modeling are used to address this question. Computational models serve to identify the cognitive mechanisms leading up to a decision, while also producing predictions for eye fixations and eye movements. Throughout the proposed work, three forms of stochasticity are considered in parallel: unreliability, volatility, and transition noise. The proposed work is expected to substantially expand the field’s understanding of multi-step planning, as well as of the role of eye movements in planning. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.