Behind every good choice is a bad guess: a simulation of feedback-based learning
Author
Faculty Advisor
Date
2026
Keywords
learnability, Rescorla-Wagner learning equation, learning theories, reward frequency, Student Research Day
Abstract (summary)
Learning begins with a mistake: but what happens when there is nothing to learn? In two-choice reward tasks, individuals repeatedly select between options that can lead to either reward or loss. When contingencies are learnable, expectations gradually align with underlying probabilities. However, when feedback is random, no stable expectation can form. This raises a key question: does task learnability alter the computational signals that underlie reward processing? Learning theories propose that behaviour updates through prediction error, the difference between expected and actual outcomes. Larger prediction errors reflect greater surprise and drive updating. This independent study developed a laboratory tutorial in Octave to simulate behaviour in learnable and unlearnable versions of a two-choice task using the Rescorla-Wagner learning equation. Overall reward frequency was held constant while manipulating whether outcome contingencies could be learned. Simulations revealed that prediction errors remained larger and more variable when feedback was random. These findings suggest that unpredictability may amplify computational surprise signals, even when structured learning is possible. By formulizing how value representations evolve across trials, this work provides a computational framework for interpreting behavioural adaptation and its relationship to neural markers of reward processing measured with electroencephalography.
Publication Information
DOI
Notes
Presented on April 23, 2026, at Student Research Day, held at MacEwan University in Edmonton, AB.
Item Type
Student Presentation
Language
Rights
All Rights Reserved