Bias You Can Notice — and Bias You Can’t: Exploring Fluency, Trust, and AI in the Classroom

Generative AI is often discussed in terms of visible bias such as data imbalance, hallucinations, or political framing. But large language models also encode structural bias: fluency preferences, reward shaping, and the optimization of coherence over correctness.

 In this interactive session, we will examine how these design choices shape trust, assessment, and reasoning. Through live demonstrations and discussion, we will what happens when prompts change, models disagree, and when fluent answers mask gaps in understanding.

Presenter: Roger Gale is a technology architect and cybersecurity educator at BCIT, blending mathematics, psychology, and systems design to understand how systems shape our world.