Evidence at the Physical Limit: Why Trustworthy Records of Non-Replayable Computation Are Possible

Two objections dominate the debate on holding AI systems accountable: the inference that produced an output cannot, in general, be replayed, and the internal state that produced it cannot be meaningfully inspected. This paper argues that neither objection supports its conclusion, by examining the regime in which both hold as physical law. Quantum mechanics forbids re-running any individual measurement event (its statistics can be repeated on fresh preparations; the recorded event cannot be restaged), forbids reading the content of an unknown state without disturbing it, and forbids complete records of any system. Yet quantum experiment is among the most trusted evidence-producing activity in science, and six decades of nuclear monitoring and safeguards practice have produced evidence fit for the highest stakes about events that can never be re-run and objects that may never be opened. Exactly three moves survive the prohibitions: replaying the judgment rather than the event; verifying structure rather than content; and declaring the scope of every record rather than claiming totality. The paper states this as a design thesis for evidence systems, states and machine-checks a minimal separation result (Lean 4, zero axiom dependencies) showing that judgment replay is independent of event replay, grounds the determinacy of the past in thermodynamics, and derives a three-question audit applicable to any AI evidence architecture, illustrated with an autonomous-agent worked example and a case study of the 2021 retraction of a Nature quantum-computing result. The central proposition: trustworthy evidence does not require reproduction of the event that generated it; it requires a verifiable relationship between the event, its record, the rules of judgment, and the declared limits of the claim. No new physics is proposed; the contribution is the assembly.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24422123
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Evidence at the Physical Limit: Why Trustworthy Records of Non-Replayable Computation Are Possible

raza sharif sharif
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

Evidence at the Physical Limit: Why Trustworthy Records of Non-Replayable Computation Are Possible

raza sharif sharif
preprint en

Abstract

Two objections dominate the debate on holding AI systems accountable: the inference that produced an output cannot, in general, be replayed, and the internal state that produced it cannot be meaningfully inspected. This paper argues that neither objection supports its conclusion, by examining the regime in which both hold as physical law. Quantum mechanics forbids re-running any individual measurement event (its statistics can be repeated on fresh preparations; the recorded event cannot be restaged), forbids reading the content of an unknown state without disturbing it, and forbids complete records of any system. Yet quantum experiment is among the most trusted evidence-producing activity in science, and six decades of nuclear monitoring and safeguards practice have produced evidence fit for the highest stakes about events that can never be re-run and objects that may never be opened. Exactly three moves survive the prohibitions: replaying the judgment rather than the event; verifying structure rather than content; and declaring the scope of every record rather than claiming totality. The paper states this as a design thesis for evidence systems, states and machine-checks a minimal separation result (Lean 4, zero axiom dependencies) showing that judgment replay is independent of event replay, grounds the determinacy of the past in thermodynamics, and derives a three-question audit applicable to any AI evidence architecture, illustrated with an autonomous-agent worked example and a case study of the 2021 retraction of a Nature quantum-computing result. The central proposition: trustworthy evidence does not require reproduction of the event that generated it; it requires a verifiable relationship between the event, its record, the rules of judgment, and the declared limits of the claim. No new physics is proposed; the contribution is the assembly.

Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon