Contextuality 101: What “POM Success Probability” Means for AI

This article introduces a hierarchy of contextuality within GPT systems, framed through the lens of resource theory. It explores how contextuality can be quantified using “monotones” like classical excess, and applies the parity oblivious multiplexing (POM) game as a practical measure. Together, these approaches provide both a theoretical and operational foundation for understanding contextuality in generalized probabilistic theories.