Under AI Conditions 2.3A.1-3

When the Information Environment Is Automated

The epistemic integrity disciplines taught in this unit — provenance tracking, verification, assumption management — were designed for an information environment in which claims arrive through channels that may obscure their origins. Under AI conditions, the problem intensifies categorically.

AI-generated content introduces three structural challenges to individual epistemic integrity that the pre-AI information environment did not produce at comparable scale.

Provenance collapse. AI systems generate text, images, audio, and video that have no human author in the traditional sense. The "source" of an AI-generated claim is a statistical model trained on data whose composition the user cannot audit. Provenance tracking (Question 2 of the Personal Provenance Protocol) encounters a structural limit: the claim originated in a system whose reasoning process is not inspectable in the way a human author's reasoning can, in principle, be examined. This does not make the claim false. It means the provenance chain terminates in a process the individual cannot independently verify — a qualitatively different epistemic condition than a claim whose provenance terminates in a human judgment one could, in principle, evaluate.

Verification flooding. AI systems can produce plausible-sounding content at volumes that overwhelm individual verification capacity. The verification discipline (from the article) asks: can I check this independently? When the volume of AI-generated claims exceeds the individual's bandwidth for independent checking — and that threshold has already been crossed for most information environments — the discipline must shift from claim-by-claim verification to source-level assessment: evaluating the systems, institutions, and channels through which AI-generated content arrives, rather than verifying each output individually.

Assumption injection. AI systems can produce content that embeds assumptions without declaring them — not through deception, but through the nature of statistical pattern-matching, which reproduces the assumptions present in training data without flagging them as assumptions. The assumption management discipline (from the article) becomes critical: when AI-generated content informs a decision, the assumptions register must include not only your assumptions but the assumptions the AI system may have embedded in the content it produced.

The Adjusted Practice

Under AI conditions, the Personal Provenance Protocol acquires a fourth question: Was this generated or mediated by an AI system, and if so, what do I know about that system's training, governance, and incentive structure? This is not a binary filter — AI-generated content is not inherently less reliable than human-generated content. It is a provenance marker that changes the verification strategy. When the answer is "yes" or "I don't know," the verification discipline shifts from checking the claim against independent evidence to checking the claim against multiple independent sources that are not derived from the same underlying model.

The self-custody practices from sidebar (2.3A.1-2) also adjust under AI conditions. AI-mediated identity verification — voice authentication, facial recognition, behavioral biometrics — introduces new attack surfaces that traditional credential hygiene does not address. The custody assessment (Practice 3) must now include: which of my access mechanisms are vulnerable to AI-enabled impersonation? The current answer, for most people, is "more than I assumed."

The honest limit remains the one identified in Phase I's stress test (1.3.9): the architecture cannot expand the biological capacity for volitional attention. Under AI conditions, the volume of information requiring attention does not merely grow — it grows at a rate determined by computational capacity rather than human production. Epistemic integrity's structural tools — provenance, verification, assumption management — help the individual allocate their finite attention more effectively. They do not make that attention infinite.