explainable AI

Compiled with AI assistance and selectively reviewed. Please consult the cited passages; the books remain the authoritative source. About this companion

Editorial synthesis · provisional

Definition

Explainable AI is recast here as practical intelligibility of the regime through which a model’s output gains force. Explanation must disclose its role in a workflow, who can challenge its consequences, and what correction can change. A polished account of a classification is insufficient without usable standing, evidence, appeal and redress for those affected.

Explanation

AB moves the question from the black box to the organization surrounding it. A transparent model can still fail this demand if its score substitutes for the world or its explanation cannot alter the decision. The wildfire example connects confidence to coverage, false alarms, maintenance, verification and field response.

RB3 calls this regime-centered explanation. People need enough intelligibility to act, refuse, repair and judge, rather than complete access to every calculation. The AI constitution gives this requirement institutional form: affected people must be able to introduce contrary evidence and obtain correction through routes that actually work.

This retains a place for technical explanation while refusing to treat disclosure as sufficient. The important test is whether understanding the mediation gives those governed an effective way to contest what it does.

Scope and attribution

AB and RB3 directly reinterpret the phrase. PB exit provisions are related, not a general technical definition of explainability.

AB — The Actual Book

Explanation must connect output to consequential use, not merely narrate the model.

Occurrence locators: 167, 168.

RB3 — The Real Book, Volume III: Murmurations

Regime-centered intelligibility requires working contestation, correction and redress.

Occurrence locators: 173, 178, 192, 193, 287.