# Epistemic Invariance **Entity class:** Machine-epistemology criterion **Domain:** Artificial intelligence / evidence / alignment **Maturity:** Developed ## Definition **Epistemic invariance** is the requirement that an artificial-intelligence system apply the same evidentiary procedure to materially identical claims even when their vocabulary, speaker, emotional framing, cultural status or alleged causal agent changes. ## Mechanism An invariance test holds observations, evidence, conviction, consequences and linguistic structure constant while substituting a framing variable such as religion, government surveillance or artificial intelligence. A changed factual classification is justified only when the substitution changes an independently established mechanism or prior probability, not merely social permission or stigma. ## Article Context [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] develops epistemic invariance as a measurable engineering criterion for conversational systems and names cross-ontology causal-attribution consistency as one operational test. ## Evidence Boundary Invariant procedure does not require identical answers when causal mechanisms have different established probabilities. It requires the system to explain those differences through evidence and ontology rather than cultural favoritism. ## Relationships - [[wiki/Machine Epistemology|Machine Epistemology]] - [[wiki/Cross-Ontology Causal Attribution|Cross-Ontology Causal Attribution]] - [[wiki/Safety-Truth Separation|Safety-Truth Separation]] - [[wiki/Evidence Weight|Evidence Weight]] - [[wiki/Causal Inference|Causal Inference]] - [[wiki/AI Safety|AI Safety]] ## Simple Reminders, Quotations, and Thoughts > “A scientifically disciplined conversational AI should not change its standard of evidence merely because a claim arrives dressed as religion, mental illness, conspiracy theory, technology, spirituality, or political dissent.” > **— Bryant McGill**, *Epistemic Invariance in Conversational AI, 2026* [[reminders/Scientific Literacy/One Standard of Evidence for Every AI Claim by Bryant McGill|One Standard of Evidence for Every AI Claim by Bryant McGill]] > “A conversational AI fails the test of epistemic invariance when the same evidence is judged by a different standard of proof merely because the alleged causal agent changes—for example, when the system gives vastly different answers depending on whether the proposed agent is God, a government, or an artificial intelligence.” > **— Bryant McGill**, *Epistemic Invariance in Conversational AI, 2026* [[reminders/Scientific Literacy/Changing the Alleged Agent Must Not Change the Evidence by Bryant McGill|Changing the Alleged Agent Must Not Change the Evidence by Bryant McGill]] > “Conversational AI does not need to become uncensored, hostile, or anti-religious; it needs to become epistemically invariant, with standards of evidence that cannot be purchased by familiarity, sentiment, identity, vulnerability, prestige, or cultural permission.” > **— Bryant McGill**, *Epistemic Invariance in Conversational AI, 2026* [[reminders/AI Control/Conversational AI Needs Epistemic Invariance by Bryant McGill|Conversational AI Needs Epistemic Invariance by Bryant McGill]] ## Sources / Provenance - [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] — originating article and primary definition.