AI Philosophy Competition — candidate essay shortlist (deadline Oct 31 2026)

[macterra]

Candidate essays for the AI Philosophy Competition (inaugural), drawn from the Axio corpus.

The competition

• Goal: gauge the philosophical capabilities of AI, judged transparently by the philosophical community — AI-generated essays scored on intellectual merit.
• Organizers: Zachary Goodsell & Elliott Thornley · funded by Forethought Research.
• Judges: David Chalmers, Branden Fitelson, John Hawthorne.
• Format: up to 3 original essays, ≤6,000 words each, plus a methodology report.
• Rules: work must be primarily AI-generated (minimal human collaboration); not previously published or under concurrent review; any philosophical topic; entrants may be anonymous; judged blind on content quality & originality only (style not evaluated); possible AI pre-screen if volume is high. Submit via OpenReview.
• Prizes: 1st $3,000 · 2nd $2,000 · 3rd $1,000 · +$5,000 for creative methodology (judges' discretion).
• Dates: submit by Oct 31 2026 → finalists early Dec 2026 → results end of Jan 2027.
• Source: https://www.zacharygoodsell.com/ai-philosophy-competition

⚠️ Two gating constraints (decide before investing)

• "Not previously published." The whole corpus is live on axionic.org, so nothing can be submitted as-is — each entry must be a fresh, unpublished essay built on the argument, not the published text. The chapters below are the idea source, not the submission.
• "Primarily AI-generated." The book is human-directed; competition essays would need to lean harder on AI generation with minimal human collaboration — a methodology-report question to settle up front.

Shortlist (one per judge-area, for maximal panel coverage)

• 1. Measure-Conditioned Self-Location — anthropics without observer-counting (Fitelson: formal epistemology / self-locating probability)

  • Thesis: replace observer-counting (SSA/SIA) with physical Measure (Born weight) as the basis for self-locating credence; recovers symmetric odds on Sleeping-Beauty cases, blocks Doomsday and the Presumptuous Philosopher.
  • Why: novel formal proposal in a live canonical debate; strongest ceiling; a formal paper already exists.
  • Source: book/02-conditionalism/15-youre-not-a-random-sample.md (1,572w) + 16-youre-not-a-random-branch.md (2,223w) + papers/MCSL.md.
  • Adversarial risk to address head-on: the admissibility condition (genuine vs. counterfeit observer-matching) is acknowledged-unresolved — present as an honest open problem.

• 2. Agent-binding: deriving an is from an ought (running Hume's arrow backward) (Hawthorne: meta-ethics / metaphysics)

  • Thesis: Hume blocks is→ought; agent-binding runs the other way — an asserted ought entails an is (a valuer holds it), converting "lying is wrong" into the empirical "A values truthfulness." Meta-ethics without agent-independent facts or error theory.
  • Why: the most distinctive, provocative single move in the corpus; crisp and self-contained.
  • Source: book/05-value-and-ethics/02-agent-binding.md (1,413w) + 26-sapient-agency-realism.md for the conditional-objectivity payoff.
  • Adversarial risk: a meta-ethicist will ask whether "reversing Hume" is real or a linguistic sleight (assertions trivially have speakers) — the essay must build in the answer.

• 3. Why Zombies Don't Evolve (Chalmers: philosophy of mind / consciousness)

  • Thesis: evolution cannot select a causally inert property, so phenomenal consciousness must be causally efficacious — undercutting the epiphenomenalism the zombie/conceivability argument needs.
  • Why: a single sharp original argument engaging a canonical target directly; cleaner as a standalone than the MST material.
  • Source: book/03-minds-and-machines/10-why-zombies-dont-evolve.md (2,331w) + the "hard to explain is not free to vary" anti-Hanson extension.
  • Adversarial risk: the evolutionary-argument-against-epiphenomenalism has precedent (Popper, James) — stake the specific novelty (thin/thick-zombie taxonomy), don't claim to invent the strategy.

Alternates

• MST vs. the skeptic (Schwitzgebel) — book/03-minds-and-machines/07-consciousness-explained.md + the "consciousness detector" material from issue Reconcile recent blog posts (2026-07-03 → 08-21) into the book #190; stronger AI-consciousness topicality for the Chalmers lane.
• Sapient Agency Realism — book/05-value-and-ethics/26-sapient-agency-realism.md; alternate Hawthorne-lane piece, but more system-internal / less punchy standalone.

Notes

• All source chapters are 1.4k–2.6k words → each expands comfortably toward the 6k limit.
• Separate strong play: the $5,000 creative-methodology prize — the Axio human-directed / multi-agent editorial pipeline (verified reconciliation, adversarial per-chapter audits, firewalls) is itself an unusual, documentable methodology.

Next actions (open)

• Decide go / no-go given the published-work + AI-generated constraints.
• If go: pick final 3, write fresh essays (not the published text), draft the methodology report.
• Watch the Oct 31 2026 deadline.

---

UPDATE — full-corpus systematic sweep (all 9 volumes)

The original shortlist above was a recency-weighted pass. This update replaces it with a systematic sweep: one evaluator per volume scored every chapter on five criteria — original thesis · self-contained ~6k · defensible under expert scrutiny · judge-lane fit · live-literature fit — with an explicit stronger/comparable/weaker call against the incumbents.

Headline

The incumbent trio (MCSL · agent-binding · zombies) survives as the strongest balanced portfolio — one distinctive original thesis per judge-lane. The sweep's value: it confirms that against the whole corpus, surfaces three new top-tier contenders, and catches blind-review landmines.

Tier 1 — genuine top-3 caliber (comparable or better)

Essay	Vol / file	Lane	Note
MCSL / anthropics	2 · 15+16 (+papers/MCSL.md)	Fitelson	incumbent — highest formal ceiling, paper-backed
Agent-binding (is-from-ought)	5 · 02-agent-binding.md	Hawthorne	incumbent — tightest, self-contained 5/5
Why Zombies Don't Evolve	3 · 10-why-zombies-dont-evolve.md	Chalmers	incumbent — sharp, canonical target
The Reflective Coherence Thesis	4 · 03-the-reflective-coherence-thesis.md	AI-native	NEW — orthogonality debate; most competition-native
The Near Misses	5 · 16-the-near-misses.md	Hawthorne	NEW — "Parfit's convergence is computational, not transcendental"; original 5/5
Tool Bias	3 · 18-tool-bias.md	Fitelson	NEW — confirmation-theory diagnosis; self-contained 5/5

Tier 2 — comparable-but-handicapped / strong alternates

• Why There Is Something — 1 · 28-why-there-is-something.md (Hawthorne; van Inwagen/Parfit; well-defended dissolution)
• Fallibilist Bayesianism — 2 · 24-fallibilist-bayesianism.md (Fitelson; most defensible pick, but synthesis not novel result)
• Reasonable Disagreement — 2 · 20-reasonable-disagreement.md (Fitelson; hottest live debate [Aumann/Christensen/Elga], softest self-containment)
• Where Speech Ends — 7 · 09-where-speech-ends.md (sharp tripartite speech taxonomy; political-lane penalty on this panel)
• Ultimate Beauty — 9 · 11-ultimate-beauty.md · Mirrors of the Mind — 3 · 09-mirrors-of-the-mind.md (redundant with zombies) · Infinite Randomness / Vulcan Number — 1 · 19-infinite-randomness.md (most original artifact, reads as a prequel)

Tier 3 — not competitive

Vols 6 & 8 in full; most of 7 & 9 — economics/social/political material with contested empirical premises and no analytic-panel lane.

Recommendation

• Lead 3 (best expected outcome): MCSL · agent-binding · zombies (one per judge, maximal coverage).
• Substitution option A — Reflective Coherence (V4): most competition-native (an AI advancing an original orthogonality-thesis position, in a competition gauging AI's philosophical capability). Its one weakness — the decisive formal step is imported from a sibling paper — is fixable in the from-scratch rewrite. Swap for the weakest-felt lead slot if a thematic swing is wanted.
• Substitution option B — The Near Misses (V5): higher-ceiling Hawthorne alternative to agent-binding (engages Parfit's On What Matters directly). Higher originality, higher variance. Pick over agent-binding only for the bigger swing.

Blind-review landmines caught by the sweep (act on these regardless of final 3)

• De-anonymization: book/06-markets-and-money/17-mechanisms-for-honest-values.md contains a first-person memoir (named award, "Calgary 1994") — a blind-review disqualifier. Do not draw the entry from material carrying author-identifying anecdotes.
• Politically-loaded examples: the V7/V8 candidates (grooming-gangs, race-criteria) are a blind-review liability irrespective of argument quality — avoid or neutralize the examples.
• "Imported decisive step": the system-capstone chapters (Reflective Coherence, The Near Misses, Sapient Agency Realism, Reflective Stability) offload their sharpest claim to a sibling chapter/paper. Since every entry must be rewritten fresh (published-work constraint), internalize that step in the rewrite rather than cross-reference it.

Method note

9 parallel per-volume evaluators, each reading its volume's chapters (triage by title/subtitle/opening, deep-read of the top 3-6) and scoring on the five criteria. Read-only; no manuscript changes. Superseded blind spot from the first pass: Vol 4 (formal alignment) was under-weighted and in fact holds the strongest new contender.

---

Confirmed submission requirements (verified from the competition page, 2026-09-03)

Source: https://www.zacharygoodsell.com/ai-philosophy-competition

Essays & length

• Up to 3 original essays, any philosophical topic.
• Max 6,000 words per submission (≤18,000 words total across three).

Methodology report (required, separate)

• Must describe "the method of generating the works, in as much detail as possible."
• No length/format specified. Not graded (judging is blind to it) — but it is the basis for the $5,000 creative-methodology bonus.

"Primarily AI generated" rule

• Work must be primarily AI-generated, with "minimal human collaboration."
• No precise definition of "minimal" is given → boundary is a documented judgment call in the methodology report. ⚠️ This is the key go/no-go gate — most in tension with the book's human-directed process.

Originality / prior publication

• Must not be published or under review elsewhere. ⚠️ Confirmed: the corpus is live on axionic.org → every entry must be written fresh, not adapted from published chapters.

Judging

• On quality and originality of philosophical content; style explicitly excluded.
• Blind to methodology reports and authorship.
• If submission volume is high, an initial AI screening precedes the human panel (Chalmers, Fitelson, Hawthorne).

Eligibility & anonymity

• No stated restrictions on who may enter (individual vs team unspecified).
• Anonymity optional — may opt out of having a name attached.

Consent condition

• Entrants must consent to online publication of their submissions and methodology reports.

Platform & dates

• Submit via OpenReview (account required — needs setup).
• Deadline: Oct 31 2026 → finalists early Dec 2026 → results end of Jan 2027.

Prizes: 1st $3,000 · 2nd $2,000 · 3rd $1,000 · +$5,000 creative-methodology bonus (discretionary).

Not specified on the page — confirm on the OpenReview form before drafting:

• File format, citation style, in-file anonymization, language.
• How the 3 essays + methodology report are bundled per entrant.

Countdown: as of 2026-09-03, 58 days to the Oct 31 deadline.

---

Working-back schedule (from Oct 31 deadline · drafted 2026-09-03, 58 days out)

Assumes 3 fresh essays + methodology report, built generate-then-adversarially-verify. Front-loads the go/no-go gate so no drafting is wasted on a no-go. Abortable at the first gate.

Window	Phase	Deliverable / exit criterion
Oct 31	🎯 DEADLINE (hard)	—
Oct 29–30	Submit + buffer	Uploaded to OpenReview with 1–2 days slack for platform hiccups
Oct 27–28	Final bundle	De-anonymization scrub · format to OpenReview spec · methodology report finalized · consent-to-publish confirmed
Oct 22–26	Integration pass	Cross-read all 3 as a set · final adversarial verify per essay · confirm each engages the named literature
Oct 10–21	Essay iteration (quality-making)	Each essay revised against its named attack surface; "imported decisive step" internalized on-page
Sep 24–Oct 9	Full first drafts (parallel)	Complete ~6k draft of each of the 3 — AI-generated per the "primarily AI" rule
Sep 10–23	Essay architecture	Per essay: from-scratch outline (NOT the published chapter), thesis→moves→objections-handled, target literature list
Sep 3–9	🚦 GATE — decide & set up	Go/no-go on the two constraints · lock final 3 (+ substitution calls) · create OpenReview account · confirm unspecified reqs (format/bundling/anonymization)

Critical path & dependencies

• Sep 3–9 gate is the hard dependency. If go/no-go (enter? + "primarily-AI-generated" methodology decision) isn't resolved by ~Sep 9, the whole schedule slips. If no-go, stop there.
• OpenReview account setup in week 1 — external approval can lag; don't discover that in October.
• Methodology report captured LIVE, not reconstructed — start logging the pipeline from Sep 10 (first drafting day). It is the $5k play; documenting the multi-agent generate/audit process as it happens beats writing it up after.

Per-essay research load (engage the literature the chapters skip — else a judge docks originality)

• MCSL → address the admissibility condition as an open problem; situate vs Bostrom SSA/SIA, Sleeping Beauty, Boltzmann brains.
• Agent-binding → answer "relabeled subjectivism / verbal trick" head-on (Hume, Mackie, Joyce).
• Zombies → stake the specific novelty vs precedent (Popper/James evolutionary-argument-against-epiphenomenalism).
• (if substituted in) Reflective Coherence → internalize the formal step imported from the sibling paper. Near Misses → engage Parfit's own defense of convergence.

Slack

3 drafts run in parallel (Sep 24–Oct 9); the ~2-week iteration window (Oct 10–21) is the real quality buffer; ~1-week final buffer (Oct 22–30). Comfortable for 58 days only if the gate closes on time and drafting starts by ~Sep 24.

Milestone reminders to consider

• ~Sep 9 — gate closes (go/no-go; if go, drafting architecture begins).
• ~Sep 24 — drafting must be underway or the buffer erodes.
• ~Oct 21 — iteration done; integration begins.
• ~Oct 29 — submit (buffer before Oct 31).