Emergent, not Immanent: A Baradian Reading of Explainable AI

Explainability is not simply "revealed" but actively created through interaction: Rather than XAI methods uncovering pre-existing meanings hidden inside AI models, interpretations emerge from the specific relationships between the model, the people using it, the context, and the tools designed to explain it. This means the same AI system may be legitimately "explained" in different ways depending on who is asking and why.

Current XAI approaches rest on hidden assumptions that limit their effectiveness: Many popular XAI methods assume meaning exists within the model itself and that explanations can be objectively extracted through computational techniques. The paper reveals these assumptions are often unexamined, which can lead to explanations that feel authoritative but may not capture the full picture of how an AI system actually works in practice.

The explainer cannot stand outside the system being explained: Traditional XAI treats the person seeking explanation as separate from the AI system, but this separation is artificial. Humans, their questions, their values, and their interpretative tools are all entangled with the AI system itself—meaning responsibility for explanation is shared rather than one-directional.

Better XAI design should support multiple, situated interpretations: Rather than aiming for a single "correct" explanation, XAI interfaces should be designed to acknowledge that different stakeholders (developers, users, regulators, affected communities) may need different kinds of understanding. This supports more honest and contextually appropriate accountability.

Ethical responsibility in AI requires rethinking how we approach transparency: If explanations emerge from specific entanglements rather than existing objectively in the model, then responsible AI governance must focus on how explanations are co-produced and participates in that process, not just on making systems more technically transparent.