Toward A Causal Framework for Modeling Perception

Perception occurs when individuals interpret the same in-formation differently. It is a known cognitive phenomenonwith implications for bias in human decision-making. Per-ception, however, remains understudied in machine learning(ML). This is problematic as modern decision flows, whetherpartially or fully automated by ML applications, always in-volve human experts. For instance, how might we accountfor cases in which two experts interpret differently the samedeferred instance or explanation from a ML model? Address-ing this and similar questions requires first a formulation ofperception, particularly, in a manner that integrates with ML-enabled decision flows. In this work, we present a first ap-proach to modeling perception causally. We define percep-tion under causal reasoning using structural causal models(SCMs). Our approach formalizes individual experience asadditional causal knowledge that comes with and is used bythe expert decision-maker in the form of a SCM. We definetwo kinds of probabilistic causal perception: structural andparametrical. We showcase our framework through a series ofexamples of modern decision flows. We also emphasize theimportance of addressing perception in fair ML, discussingrelevant fairness implications and possible applications.