{"path":"research/corpus-graph/readers/2026-10-10-ai-race-and-sensemaking-notes-as-a-reader-returned-it.md","content":"# Older notes that bear on ten recent notes about the AI race, shared sensemaking and communities (2026-10-10)\n\nOne of three read-only readers (Sonnet) sent on 2026-10-10, after the founder asked for the corpus graph to\nbe filled: *\"If it looks promising, fill up the graph with more substantial connections that you see\nclearly after reading through the corpus.\"* This reader was given ten recent notes that hold unruled\nproposals and was asked which older notes bear on a specific passage of each and are not named by it. It\nread all ten in full (its own count), with 107 tool calls. Its report is below, unchanged.\n\nWhat was kept from it, after both passages of each finding were opened and the quoted words checked by code,\nis in `../found.jsonl` (lines whose `where` names this file). Why readers were sent, and what the graph does\nwith what they find: [../README.md](../README.md) § 9.\n\nIn the report, T9 is understanding-after-the-proof-machines.md, the ninth of the ten targets given to the\nreader.\n\n---\n\n### F1\ntarget: docs/research/ai-risk-for-mathematicians-and-the-coordination-game.md:214\ntarget_words: \"So the essay's own success condition, Conjecture 2 becoming common\"\nolder: docs/research/empirical-deliberation-convergence-evidence.md:62\nolder_words: \"The ~20-point gap barely closed, and participants attributed much of what movement occurred to real-world events\"\nkind: contradicts\nwhy: The older note reads the Forecasting Research Institute adversarial collaboration on this very debate (22 skeptics and concerned experts, 8 weeks, moderated crux calls). It found no convergence, no failure of mutual understanding, and disagreement rooted in different anchoring priors and in how much weight multi-step theoretical arguments deserve; the near-term cruxes were empirically empty. A session writing § 3d would not present a shared assessment among those who race as the thing mapping can deliver. It would cite this as a measured best case where understanding was complete and the assessment still did not converge, and it would soften the § 2d claim about walking a skeptic through S1-S20.\nstrength: strong\n\n### F2\ntarget: docs/research/mapping-the-machine-consciousness-debate.md:109\ntarget_words: \"none of the kinds has a name for that floor\"\nolder: docs/research/synthetic-stress-suite-and-ontology-reflection.md:11\nolder_words: \"a descent from an evaluative dismissal to the phenomenal floor — the first live `is_bedrock` terminus\"\nkind: contradicts\nwhy: By 22 August the corpus had a horizon kind, `is_bedrock`, for exactly the phenomenal (what-is-it-like) floor, and a live descent typed with it. Prediction 6 repeats the July gap from deep-descent-experiment.md as if still open. A session would instead ask whether the access-to-experience step is an `is_bedrock` leaf (O7 in the ledger proposes renaming it `is_open`), and would reuse the tasting-machine descent as the precedent.\nstrength: strong\n\n### F3\ntarget: docs/research/schmachtenberger-and-the-ai-race.md:353\ntarget_words: \"nothing, the claim loses a little, and that is what makes the test worth running.\"\nolder: docs/research/dogfood-run-6-israel-palestine-cross-domain.md:98\nolder_words: \"What died is the assumption that the shipped machinery can detect it.\"\nkind: narrows\nwhy: Run 6 was already the first cross-domain auto-connect test. It found claim-level cosine cannot see structural homology across domains (402 within-debate candidates against 4 for the rest of the corpus) and ruled that a null does not count against the metacrisis claim. § 6 should say that a null from auto-connect only counts if the test runs on the later concept-route and structural-kinship generators (auto-connect-upgrade.md § 2b, built 2026-08-20), not on embeddings.\nstrength: strong\n\n### F4\ntarget: docs/research/shared-sensemaking-what-deliberus-holds-and-owes.md:705\ntarget_words: \"sense mismatch and the keeper's verdicts, read and never set\"\nolder: docs/research/nativism-thread-and-false-dissolution.md:41\nolder_words: \"After the dissolution, does anything about what to actually do change?\"\nkind: supplies-a-missing-piece\nwhy: V8 proposes computing \"talking past\" from sense mismatch alone. The older note shows two things V8 lacks. Failure to join issue also occurs with no sense mismatch (a party pivots to another level of argument), and a claimed dissolution can be a \"leveling dissolution\" that leaves the decision unchanged. V8 should carry this checkable test, and should not score a debate as well-connected on topical edges between parties who are not engaging.\nstrength: strong\n\n### F5\ntarget: docs/research/a-cognitive-ecology-and-the-terms-of-the-relation.md:308\ntarget_words: \"Find the premise a set of claims shares, store it as a claim that\"\nolder: docs/research/the-load-bearing-unsaid.md:21\nolder_words: \"Its diagnostic feature is invisibility — doxa is what orthodoxy and heterodoxy\"\nkind: narrows\nwhy: A premise both camps share is doxa-like, not an ordinary omitted crux. The older note says implicit-premise recovery suits a crux and is the wrong tool for a hinge-like commitment, that pseudo-sacred versus hinge is found out by trying, and that turning doxa into orthodoxy is political and not neutral (Bourdieu). A session proposing to store the disputed premise as a claim would add this test and warning. This applies equally to mapping-the-machine-consciousness-debate.md § 5j item 6.\nstrength: strong\n\n### F6\ntarget: docs/research/ai-risk-for-mathematicians-and-the-coordination-game.md:388\ntarget_words: \"Values and capabilities are independent.\"\nolder: docs/research/heylighen-value-complexity-and-deliberus.md:9\nolder_words: \"intelligence and values are not independent dimensions\"\nkind: already-answers\nwhy: S3 is the orthogonality thesis, and the corpus already holds a source that attacks it at the root: values originate in self-maintenance, and relevance realization is value-laden. A hand-made map of S1-S20 should attach this as an attack on S3 (and on the S1 caveat) rather than leave S3 unchallenged. The note's own read-depth is two structured extractions of the essay, not a full read.\nstrength: strong\n\n### F7\ntarget: docs/research/mapping-the-machine-consciousness-debate.md:413\ntarget_words: \"The step had no way to say *the source reports this party as not holding the position*.\"\nolder: docs/research/the-remaining-case-for-the-stance-guard.md:35\nolder_words: \"Add it once (asserts · rejects · reports without taking a position)\"\nkind: narrows\nwhy: On 13 September the founder ruled that reported speech needs a three-valued stance field (asserts, rejects, reports) on the claim and on the REPORTS edge, because no module stores the reporter's stance. The `not_held_by` repair is a one-case patch of that same gap. A session would fold it into the ruled stance field and test the five reported-speech-with-stance-lost cases from that note.\nstrength: moderate\n\n### F8\ntarget: docs/research/understanding-after-the-proof-machines.md:42\ntarget_words: \"A machine-made map is born an unsolved exposition problem too.\"\nolder: docs/research/formalized-fermat-and-the-oracle-boundary.md:80\nolder_words: \"Nobody will read 13 million lines, and nobody needs to, because the kernel ratified every one.\"\nkind: contradicts\nwhy: The corpus's own reading of Anthropic's 11-day, 13-million-line Lean proof of Fermat's Last Theorem took the \"nobody needs to read it\" side. It is the sharpest specimen of Sanderson's claim that a machine proof is born an unsolved exposition problem. A session would use that proof as the worked case, and state that the older note's position holds only where an oracle exists and leaves the \"checked is not yet understood\" gap open.\nstrength: moderate\n\n### F9\ntarget: docs/research/a-cognitive-ecology-and-the-terms-of-the-relation.md:182\ntarget_words: \"So the ecological picture moves the risk from intention to selection.\"\nolder: docs/research/heylighen-value-complexity-and-deliberus.md:12\nolder_words: \"The principal existential danger may come not from making AI too intelligent, but from making it too much like life.\"\nkind: supports\nwhy: The note rests the intention-to-selection step on one source (Szathmary), read from an abstract and two sections. Heylighen reaches the same boundary independently (tools selected for use against systems selected for their own persistence) and grounds value in autopoiesis. That also supplies a second warrant for § 1's \"a machine that feels has something to lose\", and makes the paragraph less single-sourced.\nstrength: moderate\n\n### F10\ntarget: docs/research/ai-risk-for-mathematicians-and-the-coordination-game.md:91\ntarget_words: \"the crux is stated on one side and left implicit on the other (run 3F)\"\nolder: docs/research/frontier-extraction-experiment.md:40\nolder_words: \"Yudkowsky never states the orthogonality thesis or instrumental convergence\"\nkind: same-idea-earlier\nwhy: Run 3F's AI-risk batch found the decisive clash between two unstated premises: Yudkowsky's orthogonality and instrumental convergence, against Andreessen's \"goal-directedness requires being alive\". That is the psychedelics batch's one-sided pattern only in the second batch, so the gloss here is off. He's S2 and S3 state the first pair explicitly, so a map of this essay is the place to wire them to that opposing implicit node.\nstrength: moderate\n\n### F11\ntarget: docs/research/blobs-community-sandboxes-and-the-advice-process.md:259\ntarget_words: \"the one signal that would redirect the whole build\"\nolder: docs/research/interaction-modalities-and-the-pointing-test.md:33\nolder_words: \"A null under one interface is data about that interface.\"\nkind: narrows\nwhy: On 17 August the founder ruled that \"do people point at claims\" is only a signal across a pre-registered set of modalities, and the verdict is \"did not pay under the modalities tried\". Blob 1 (and Blob 2) should name its modality set in advance and log any expansion after a null, or the falsifier cannot fire in either direction.\nstrength: moderate\n\n### F12\ntarget: docs/research/the-post-that-states-the-problem-from-inside.md:112\ntarget_words: \"deciding it is out of scope because no platform can manufacture it.\"\nolder: docs/analysis-and-attunement.md:149\nolder_words: \"conversations **deploying arguments alone had no effect**, while otherwise identical conversations\"\nkind: contradicts\nwhy: The older note's field experiments (Kalla and Broockman, 6,869 voters, a stranger's non-judgmental exchange of narratives, durable at four months) show that the active ingredient is manufacturable without a prior bond, and that arguments alone were null. A session would not frame the \"ninth condition\" as love or leave it out of scope. It would instead reframe it as narrative exchange, and note that the post's by-hand descent is the attunement arm, not the argument arm. The Pettigrew and Tropp contact meta-analysis in empirical-deliberation-convergence-evidence.md supports the same reframing.\nstrength: moderate\n\n### F13\ntarget: docs/research/four-cuts-of-the-ai-debate.md:70\ntarget_words: \"the compressed form is the one that travels\"\nolder: docs/research/the-residual-error-taxonomy.md:242\nolder_words: \"Compression drops the scope clause first, because the scope clause is the least memorable part\"\nkind: same-idea-earlier\nwhy: The note calls this \"the corpus's finding\" but cites no home. The home gives the mechanism (the scope clause goes first) and a seconds-long headline test that can be run on Romero's and O'Connor's renderings. It explains why \"the epistemic bare minimum\" became \"advocates for acceptance\", and gives V7 an instrument.\nstrength: moderate\n\n### F14\ntarget: docs/research/understanding-after-the-proof-machines.md:63\ntarget_words: \"A map view that chooses what to show\"\nolder: docs/research/scaffolding-versus-difficulty.md:17\nolder_words: \"The demand for clarity was trained by the supply of it.\"\nkind: contradicts\nwhy: The older note records a live objection that supplying clarity (rubrics, exemplars, next steps) trains demand for more clarity and removes the difficulty learning needs. It proposes the rule \"show the door, never walk through it\". A motivation check on reading order, or discovery fiction read out of the record, is a way of walking through. T9's counter-voices omit this, and a session should either pass the candidates against that rule or note the tension.\nstrength: moderate\n\n## Looked for and not found\n\n- An older internal note on the game theory of the AI race (stag hunt, assurance, common knowledge of the risk). The only earlier text is the passage in non-zero-sum-economics-and-civilizational-cooperation.md § 6, which the coordination note already corrected and cites. Aumann's agreement theorem, with its common-prior condition, is held at moral-convergence-metaethics-and-psychology.md:65. I folded it into F1 rather than reporting it separately.\n- Anything before October on machine consciousness or moral patienthood beyond deep-descent-experiment.md and the tasting-machine stress test (F2). The corpus has no earlier map of that debate, so the consciousness note's \"nothing on the subject\" is accurate.\n- A prior statement of the contributors-pay idea for whose-tokens-pay-for-mapping.md, as the note itself says. I found only the-residual-error-taxonomy.md class 2 (ingestion balance) and two partial run-to-run measurements (cross-source-premise-run.md:92, scheme-set-exhaustiveness.md:190). All were too thin to report. I also dropped three finds because the target already names the note: memetic-immune-system.md § 10 (an answer to shared-sensemaking § 4.3), legibility-under-power-red-team.md on Brandolini and persistence (against the blobs note's \"persistence buys nothing\" row), and microtasks-in-broad-daylight.md.\n\nTargets read in full: 10 of 10.\n"}