Explainability Is Not a Feature: A Position on Trustworthy AI

Gabriel Banaggia, Eduardo Soares, Renato Cerqueira, Emilio Vital Brazil, Simone Barbosa

Keywords: Explainable AI, Trustworthy AI, Human-Centered AI, Selective Explainability, AI Transparency

Abstract:

Explainable Artificial Intelligence (XAI) is widely invoked as a central
component of trustworthy AI. Despite this prominence, the concept of
explainability remains deeply contested, with persistent disagreement
regarding what constitutes an explanation, whom explanations are for, and
the conditions under which they are meaningful. In this position paper, we
argue that these disagreements reflect a fundamental limitation of current
approaches that treat explainability as an intrinsic technical property of
models. Drawing on a systematic synthesis of the XAI literature, we contend
that explainability is more appropriately understood as a relational and
context-dependent requirement, shaped by human roles, decision contexts, and
risk profiles. We further argue that indiscriminate or exhaustive explanations
can undermine trust, motivating the need for selective and purpose-driven
explanatory practices. Finally, we discuss the implications of this reframing
for trustworthy AI research, deployment, and regulatory oversight.

Submission Number: 192