Source review: first high-risk batch
What changed
All three PDF-dependent entries received a direct source check and an independent read-only review. The largest correction was to the simulation argument: Bostrom’s first horn is a near-zero fraction of human-level technological civilizations reaching posthumanity, not “almost all civilizations go extinct.” The paper’s third horn is a possible consequence, not an assumption.
- The Simulation Argument: The article now separates the assumptions about consciousness and computing from the three-way conclusion. It also makes clear that Bostrom’s personal-credence step is conditional on the “bland indifference principle” and relevant evidence. Bostrom’s paper argues for a disjunction; his 2005 reply to Weatherson says it does not establish that the simulation branch is more likely. The article notes the debate without claiming it is settled. Weatherson’s PhilArchive abstract was consulted; the full paper was not accessible in this review.
- Pascal’s Mugging: The entry now states the conditions Bostrom actually uses: unbounded utility, linear happy-day valuation, a finite promised reward, and no risk aversion or temporal discounting. It no longer presents the puzzle as a general result about every improbable claim. It credits Yudkowsky’s 2007 post with naming the problem while distinguishing his example from Bostrom’s. A later response by Hiller and Hasan is described as one proposal, not a consensus solution; their peer-reviewed article proposes considering symmetric rewards available if Pascal declines.
- The Orthogonality Thesis: The definition now uses Bostrom’s “capacity for instrumental reasoning,” preserves “more or less” and its qualifications, and separates final goals from instrumental subgoals. Two later arguments are presented as debate, not settled refutations: Müller and Cannon’s critique of a common inference to AI existential risk, and Sharadin’s analysis of goal-promotion accounts used in instrumental-convergence arguments.
Review state and limits
Under the approved source-review standard, all three entries are source-checked for the material claims listed in the batch ledger. Independent reviews found no remaining material source issues. This label means claim-to-source traceability in the recorded scope; it does not mean the philosophical arguments are true, settled, exhaustively researched, or reviewed by a human subject-matter expert. No human expert review is claimed.
This batch covers the three entries singled out for PDF-dependent priority review, not all 57 newer entries identified in the repository’s quality report. The other entries remain outside this audit. No source-review status has been added to data/experiments.json or the site. Dataset changes add source references and support descriptions, update the Pascal summary and the orthogonality citation title, and record source-consultation dates.
Repository validation passed, all 18 unit tests passed, and git diff --check passed. The work remains on the local audit/source-review-high-risk branch. Nothing has been committed or pushed.
Files in this batch