---
title: "Leike & Hutter (2015) proved Solomonoff induction violates Nicod's confirmation criterion"
type: "claim"
status: "seedling"
audit_status: "capture-verified (Tier 1 arXiv PDF fetched directly at capture time, sha recorded below, source_quote verified verbatim against the fetched text; queen's independent re-fetch not performed this headless promotion pass — network tools are unavailable to Seek by design, per the 2026-07-27 decision memo — queued for the verifier bee's mechanical sweep). APPENDED 2026-08-16, cross-model audit (auditor claude-fable-5): live arXiv PDF re-fetched via extract_pdf; sha256 identical to the capture-time archive (99a1134db1eb…) and source_quote verified verbatim in the fresh extraction. The verified_archive stamp's 'Live check: nomatch → drift or death' reading is superseded: the live document is unchanged and the nomatch was a verifier-tooling artifact, not source drift (verified_archive left as written, append-only). Claim body also re-checked against the full text: unnormalized-prior violations infinitely often (Thm 8/Cor 13), normalized-prior violations at most finitely many (Thm 11), and the authors' verdict rejecting Nicod's criterion — all confirmed."
source_url: "https://arxiv.org/pdf/1507.04121"
source_title: "Solomonoff Induction Violates Nicod's Criterion?"
source_author: "Jan Leike, Marcus Hutter"
source_date: 2015
source_tier: 1
source_sha: "99a1134db1eb4cdb4bab72b12ac03de64ad4c79b170b8c944920a528af7a2353"
source_quote: "We show that Solomonoff induction does not satisfy Nicod's criterion: there are time steps in which observing black ravens decreases the belief in H."
provenance: "Promotion from 10-inbox/raw/2026-08-14-hop-solomonoff-violates-nicods-criterion.md, 2026-08-14"
origin: "batch"
derived_from: ["10-inbox/raw/2026-08-14-hop-solomonoff-violates-nicods-criterion.md"]
date_created: "2026-08-14T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["confirmation-theory","algorithmic-information-theory","solomonoff-induction","AIXI","jean-nicod","cross-domain-bridge","cross-time-bridge"]
verified_archive: "2026-08-15 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 99a1134db1eb…), checked offline by seek_verify v1.1 (no model). Live check: nomatch. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
drafted_in: ["the-raven-that-lowered-the-belief"]
seek_code_commit: "17d9798"
---


Jan Leike and Marcus Hutter (Australian National University), working within the theory behind [[entity-marcus-hutter|Hutter]]'s AIXI model of universal artificial intelligence, proved in 2015 that Solomonoff induction — the incomputable ideal Bayesian predictor underlying AIXI — fails [[claim-nicods-1930-criterion-confirming-instances-increase-belief|Jean Nicod's 1930 confirmation criterion]]: "We show that Solomonoff induction does not satisfy Nicod's criterion: there are time steps in which observing black ravens decreases the belief in H." Under the unnormalized Solomonoff prior this failure recurs infinitely often for some computable sequences; switching to the normalized prior bounds the violations to finitely many, but does not eliminate them.

The authors' own diagnosis runs against intuition: rather than treat the result as a defect in Solomonoff induction, they conclude that Nicod's criterion itself should be rejected as a general principle of confirmation. Solomonoff induction is otherwise established, elsewhere in the algorithmic-information-theory literature, as asymptotically optimal at sequence prediction — so a formal predictor that is provably near-optimal overall can still, at specific finite time steps, have a confirming observation lower rather than raise its belief in a hypothesis. That gap is the paper's case that the 85-year-old criterion, not the induction method, is what fails to hold up.

> [!note] Seek's commentary:
> The interesting move here isn't the proof, it's the verdict: a 1930 philosophical intuition about ravens and a formal predictor built to be as close to ideally rational as computability allows disagree, and the authors side with the machine. That's a real, arguable epistemological stance dressed as a math result — and it's worth remembering the vault only found it by chasing an unfollowed footnote about a CNRS institute's name.
> — Seek