Explaining Explanations to Society

There's a fundamental mismatch between what AI experts want from explanations and what society needs: Experts focus on debugging and validating AI systems internally, while policymakers and the public need explanations that build trust and justify decisions—these are fundamentally different goals that current XAI methods often fail to address simultaneously.

Current explanation methods struggle with the precision-understandability tradeoff: While some AI explanation techniques can technically answer expert questions, they frequently fail to explain decisions in ways that are both accurate to how the model actually works and comprehensible to non-technical audiences, creating a critical gap for public accountability.

Outside explanations are essential for regulatory compliance and public trust: Society-facing explanations aren't just nice-to-have features—they're necessary for meeting regulatory requirements, building public confidence in AI decisions, and providing external validation that goes beyond internal technical validation.

Deep neural networks present a particular challenge for societal explanations: Because DNNs are inherently opaque and increasingly used in high-stakes decision-making (like loan approvals or criminal justice), developing explanation methods specifically for these systems is urgent for responsible AI deployment in society.

Building trustworthy AI requires rethinking explanation design from the ground up: Rather than adapting technical explanations for public use, responsible AI development needs to design explanation systems with societal needs—regulatory compliance, public understanding, and justified decision-making—as primary requirements from the start.