Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

When Reward and Punishment are Both Present: Optimistic Learning Bias in Approach-Avoidance Conflict Reinforcement Learning

Poster Session C - Sunday, March 8, 2026, 5:00 – 7:00 pm PDT, Fairview/Kitsilano Ballroom

Haoyu (Allen) Zhang¹, Cendri Hutcherson¹, Yanxi (Elinor) Zhu², Charl Margaux Elcano¹, Zhen Wu², Rutsuko Ito¹, Andy Lee¹; ¹University of Toronto, ²Tsinghua University

Adaptive decision-making often involves approach-avoidance conflict (AAC), where one option can lead to reward and/or punishment, while another option typically results in a safer outcome. Such behaviour often requires the learning of each option’s associated outcome/s (e.g., reward and punishment probabilities), a process effectively described by reinforcement learning (RL). Previous research in humans and other mammals has shown that such learning can be characterized by an optimistic learning bias, with positive outcomes being weighted more strongly than negative ones, thereby producing optimistic value estimates for chosen actions. However, most work has examined reward and punishment learning separately, leaving open the question of whether this bias extends to AAC RL. Across three behavioural experiments (total N = 260), we addressed this question using a paradigm in which participants were required to choose between a conflict choice (higher probability for reward with the possibility of punishment) and a safe choice (lower probability for reward only). In one task version, reward and punishment were both monetary (gain vs. loss), while in another version, monetary reward was pitted against a subjectively equated aversive food punishment. RL modelling of choice data revealed that when both outcomes were monetary, optimistic learning emerged for both reward and punishment. Interestingly, when punishment was aversive food, the bias primarily influenced punishment learning but not reward learning. We discussed the implications of these findings, highlighting how shared neural mechanisms of punishment learning and AAC processing may contribute to this bias.

Topic Area: THINKING: Decision making