16:30 - 18:00
Tue—HZ_13—Talks6—66
Tue-Talks6
Room:
Room: HZ_13
Chair/s:
Chris Donkin
Exploration in Domains with Gains or Losses
Tue—HZ_13—Talks6—6601
Presented by: Ludwig Danwitz
Ludwig Danwitz *Ann-Katrin HoschBettina von Helversen
Universität Bremen
To minimize losses or maximize gains, individuals must understand the rules and patterns in their environment. This study examines how environmental factors influence exploration and exploitation behaviors, comparing environments where participants can either only win (with varying gain sizes) or only lose (with varying loss sizes). Participants completed a Multi-Armed Bandit task across these conditions in two online experiments. Results showed reduced exploration behavior but better performance in gain environments compared to loss environments.

The relative richness of the environment—whether an environment yields more or fewer points than expected based on prior blocks—also influenced exploration. Richer-than-expected environments reduced exploration, particularly in the gain domains, where this reduction harmed performance as participants overlooked lucrative opportunities.

Computational modeling revealed three processes underlying exploration behavior: optimism under uncertainty, which biases participants toward uncertain options; random exploration, where participants occasionally choose suboptimal options to gather information; and unsystematic noise, reflecting erratic choice behavior.

Increased exploration in poorer environments stemmed from noisier behavior, while decreased exploration in richer environments was driven by relatively reduced optimism under uncertainty. These findings highlight the nuanced interplay between environmental context and cognitive mechanisms in shaping exploration and exploitation behaviors.

Keywords: Exploration-Exploitation, Loss aversion, Noise, Reinforcement Learning, Uncertainty, Decisions from experience