{"path":"research/perceiving-possibility-and-the-analogy-organ.md","content":"# Perceiving Possibility and the Analogy Organ — a founder-shared thread read against the corpus\n\n**Date**: 2026-09-02 · **Type**: source intake (a public Facebook post by Manu Herrán, 2026-08-31,\n\"inspired by Adam Ford\", with comments by Steve James and Rama Ganesan) read against the analogy\nmachinery already designed here · **Status**: reflection; two proposals at the end, unruled.\n**Read-depth caveat**: the Hofstadter/Copycat, error-management and recognition-primed-decision\nreferences are from memory of well-known work, not re-read this session — verify before citing\noutward.\n\n---\n\n## 1. The thread, compressed\n\nHerrán: the \"humans create, AI recombines\" distinction does not survive scrutiny — human creativity\nis recombinatorial too. The interesting distinction underneath is **detecting structure while it is\nbarely visible**: weak signals, intentions from incomplete evidence, acting before the pattern is\nexplicit. Courtship is the revealing case (subtle, often unarticulable signals, yet the interaction\ndepends on reading them); the animal world is full of it; evolution rewarded organisms that detect\n*what is about to matter* before it is obvious. So some creativity is *perceiving possibility*\nrather than producing novelty. And AI becomes an **experimental contrast** for biological minds:\nif it reproduces these abilities we learn which mechanisms suffice; if it consistently struggles,\nthe failures are equally informative.\n\nSteve James: Hofstadter said this decades ago — creativity is variations on a theme. Rama\nGanesan: when she talks to AI, *she* brings the unexpected connections; the AI, with the whole\nlibrary, does not make them, and \"in some ways it SHOULD NOT\" — human creativity is idiosyncratic\nbecause individuals are.\n\n## 2. The distinction is one the corpus already holds, with a name on each side\n\n**Recombination** is the reuse flywheel: Kauffman's adjacent possible, Arthur's combinatorial\nevolution, and assembly theory's index — construction from existing parts is not a deficiency but,\nabove a complexity threshold, the only construction available\n([complexity-transitions-and-why-now.md](complexity-transitions-and-why-now.md),\n[assembly-theory-and-the-reuse-mechanism.md](assembly-theory-and-the-reuse-mechanism.md)). Herrán\nreaches \"recombination is not an objection\" from the creativity side; the corpus reached it from the\nphysics side. Same conclusion, independent route.\n\n**Perceiving possibility** is the second term of assembly — not the index (how hard to make) but\n*selection* (which combination will be copied), anticipated before the copy number arrives. The\ncorpus's objects for it, already built or designed:\n\n- the **analogy daemon**, the graph's gestalt organ — *\"intuiting what belongs\"* is Herrán's faculty\n  named for a graph ([the-analogy-daemon.md](the-analogy-daemon.md));\n- the **dancer** in Zurn & Bassett's curiosity styles — connects two structures so both change — the\n  daemon's human twin ([curiosity-as-connection-zurn-bassett.md](curiosity-as-connection-zurn-bassett.md));\n- the **hinge** and `worth_asking` — which unopened question will swing the root: perceiving where\n  the structure will crystallize, one step ahead; Terenin's reservation values are its non-myopic\n  form ([derivability-of-missing-considerations.md](derivability-of-missing-considerations.md) § 6);\n- the **implicit-crux finding** — the crux of a real debate is stated on one side and implicit on\n  the other, and implicit-premise recall is the pipeline's sharpest quality axis. Reading the\n  unstated crux *is* weak-signal detection as a pipeline metric;\n- the **scheme nursery** — clusters of `does_not_fit` are the signal of a category that does not\n  exist yet, detected before the category does: perceiving possibility built into the taxonomy\n  machinery ([scheme-set-exhaustiveness.md](scheme-set-exhaustiveness.md));\n- and the project itself: the idea preceded its parts by a decade, prototyped once before the\n  adjacent possible contained what it needed — perceived possibility, with a documented failed\n  first attempt as the receipt.\n\n## 3. Herrán's own argument is an analogy argument — run the four checks on it\n\nThe post's warrant is a cross-domain analogy: courtship ↔ scientific regularity before the evidence\n↔ musical possibility before the genre ↔ opportunity before others see it. This is exactly what the\ndaemon's second pass would receive, so run its checklist\n([convictions-across-scales.md](convictions-across-scales.md) § 1):\n\n- **Recurrence** — passes: the faculty is independently named outside the post. Klein's\n  *recognition-primed decision* (firefighters sensing a floor about to collapse before any\n  articulable cue) and Haselton & Buss's *error management theory* — whose founding 2000 paper is\n  literally about courtship mind-reading under asymmetric error costs — name it from two fields.\n- **Mechanism identity** — **lowers the transfer**. Courtship is a *cooperative signalling game\n  under deniability*: the sender wants to be read and wants to be able to deny it (Pinker, Nowak &\n  Lee on indirect speech). A scientific regularity is *non-agentive pattern in noise*; nobody is\n  sending. A market opportunity is *competitive*: others are actively hiding it. Same shape —\n  \"detect before obvious\" — three different mechanisms. The evolutionary warrant (\"enormous time to\n  reward it\") holds for the agentive cases and does not carry to science, which has had four\n  centuries.\n- **Adversary** — courtship has fakers and rivals; science acquires motivated stakeholders only\n  later; markets have strategic adversaries from the start. The faculty's *reliability* differs by\n  adversary class, which is the scale ladder's rule in miniature.\n- **Substrate condition** — what the cheap detector requires is *asymmetric error costs*: a false\n  alarm must be cheaper than a miss (Nesse's smoke-detector principle). Where that asymmetry is\n  absent, \"perceiving possibility\" is just paranoia with a good press agent. That is a real\n  condition on any Deliberus instrument that acts on weak signals: name which error is cheap.\n\n**Verdict: Conditional.** The post's thesis survives — perceiving possibility is a distinct faculty\nfrom recombination — but its strongest example transfers only within the cooperative-signalling\nclass, and the disanalogy is the load-bearing finding. The IRIT considerations agree: Number and\nDiversity high, Relevance holds on the carried property, **Disanalogy is where the strength goes**\n([the-irit-harvest.md](the-irit-harvest.md) § 3). This is what the daemon should produce for any\nanalogy claim: the mother claim, the instances, and the strongest counter-analogy attached — and the\npost is a good first specimen, since it is exactly the register (an intelligent public essay built\non one governing analogy) the daemon would meet most.\n\n## 4. Hofstadter, whom the corpus has never cited — and Copycat, which the founder re-described\n\nSteve James is right and the corpus has a gap: `rg -i hofstadter docs/` returns nothing. Two of\nHofstadter's contributions bear directly on the machinery:\n\n**Analogy as the core of cognition** (*Surfaces and Essences*, with Sander, 2013): every concept is\nbuilt by analogy; categorization *is* analogy-making; \"the same situation\" is a judgment, not a\nlookup. That is claim-sameness stated from cognitive science — the graded articulations, the\ndemand-driven sense split, and the seven-key judge are all analogy-making instruments, and\nHofstadter is the missing name in the sameness-across-fields doc\n([sameness-merge-and-split-across-fields.md](sameness-merge-and-split-across-fields.md)).\n\n**Copycat** (Mitchell & Hofstadter, *Fluid Concepts and Creative Analogies*, 1995): an architecture\nthe founder re-invented in one sentence. Its parts: **codelets** — small agents that run in parallel,\nstochastically, in three kinds: *scouts* that propose a structure, *strength-testers* that evaluate\nit, *builders* that commit it; a **Slipnet** — a concept network with spreading activation, where\nconcepts *slip* into neighbours under pressure; and **temperature** — the system's own measure of how\nsettled its current structures are, which governs behaviour: high temperature means more random\nexploration and a willingness to break structures, low temperature means the interpretation has\ncrystallized. \"Daemons buzzing around\" *are* codelets; the two-pass design (propose a prior /\ninvestigate the pattern) *is* the scout/tester split; the concept layer with sense states\n(underdefined, emerging, bifurcated…) *is* a Slipnet; and the founder's \"not autistic, seeing the\nbroad strokes\" is Copycat's *parallel terraced scan* — perceiving many possibilities shallowly before\ncommitting to one deeply. Copycat did Herrán's faculty in a microdomain in 1990. Its unfinished\nbusiness was scale: the Slipnet was hand-built. The substrate difference here is that an LLM *is* a\nSlipnet with slippage built in, which is the one reason the analogy might transfer where the FARG\nprogramme stalled.\n\n**The transferable piece is temperature.** The graph has static settledness signals (the gray\nband, UNSETTLED, `undecided_share`, confidence tiers) but nothing that *governs behaviour* by\nregion. A region-level temperature — how contested, how unratified, how recently moved a subgraph\nis — would tell daemons **where to roam** (high temperature: propose, break, re-frame; low\ntemperature: leave it), tell the triage feed where human attention pays, and give \"this region is\nframed wrong\" a quantity to attach to. It also composes with the enzyme constraint: a daemon roaming\nby temperature accelerates change where change is already happening, which is the rate-not-\nequilibrium condition met by construction. **Proposal, unruled** (register band D).\n\n## 5. Rama Ganesan's objection is a testable claim, and the graph can test it\n\n*\"It does not make those unexpected connections. In some ways it SHOULD NOT.\"* Two halves. The\nsecond is the corpus's containment stated as an intuition: the daemon proposes, attaches its\ncounter-analogy, never asserts — machine-made connections enter as challengeable structure. The\nfirst is an empirical claim about *where unexpected connections come from*, and it is exactly the\nquestion the derivability test answered for missing considerations today: the machine reaches the\nwell-articulated space (rungs 0–4); what it cannot supply is the **idiosyncratic** — one person's\nparticular library, read in one particular order. That is rung 5 wearing creativity's clothes.\n\nDesign consequence: the daemon needs a **human-seeded channel**. A person's cheap act — *\"this\nreminds me of X\"* — is the idiosyncratic juxtaposition; the daemon's expensive act is developing\nand testing it (instances, counter-lens, four checks). Perceiving possibility stays with the\npositioned human; the harness does the checking. And Rama's claim becomes an instrument rather than\na comment: compare human-seeded against daemon-seeded analogy claims on survival under attack and\non ratification — the agent-side falsifier's clean design applied to creativity.\n\n## 6. \"Forced legibility crushes weak signals\" is the same faculty seen from the threat side\n\nThe red team's highest-severity attack ([convergence-wager-red-team.md](convergence-wager-red-team.md))\nis that nascent positions are inarticulate before they are right, and a system scoring atomic\nconsistency immediately kills them with a paper trail that looks like virtue. Herrán's faculty and\nthat attack are one phenomenon: a weak signal is precisely what scoring destroys and what the\ngestalt organ is for. So the analogy daemon and the **score-free nursery** belong together: a\nhigh-temperature region needs *low scoring pressure* and *high detection effort* at once. The\nnursery is where the daemon should roam most and where the strength layer should say least.\n\n## 7. AI as experimental contrast — the corpus's falsifier design, and its first result\n\nHerrán's closing move (either outcome teaches us something) is the corpus's own method: the two\nfalsifiers scored separately, the agent-side one clean because a machine client has no affordance\nto blame ([agents-as-a-consumer-class.md](agents-as-a-consumer-class.md)), and the harness\nconviction's ceiling — the two human-reserved classes — which are exactly \"what AI consistently\nstruggles with\" named in advance. Today's derivability test is a first datum in that experiment,\nand it points somewhere specific: on ordinary debates the machine reached every missing\n*consideration*; what it could not supply was **positional** — what a thing costs from where a\nparticular person stands. So the human residue that the contrast reveals is not a missing faculty\nbut a missing *position*. That is the positionality conviction (what is invisible from inside a\nposition is visible from outside it, so minds make each other legible) confirmed from the machine\nside: the graph does not need humans because they perceive better; it needs them because they are\n*somewhere*, and it needs many somewheres. Rama's \"unique individuals with their own experiences\"\nis the same sentence.\n\n---\n\n**Proposals filed (unruled)**: region-level temperature as a daemon-roaming and triage signal (§ 4);\na human-seeded channel into the analogy daemon plus the human-vs-daemon analogy survival instrument\n(§ 5); Hofstadter added to the sameness-across-fields lineage and to the Academic Foundations list\n(the latter is a CLAUDE.md structural change and needs the founder's blessing).\n\nCross-references: [the-analogy-daemon.md](the-analogy-daemon.md) · [convictions-across-scales.md](convictions-across-scales.md)\n§ 1 · [the-irit-harvest.md](the-irit-harvest.md) § 3 · [derivability-of-missing-considerations.md](derivability-of-missing-considerations.md)\n· [curiosity-as-connection-zurn-bassett.md](curiosity-as-connection-zurn-bassett.md) ·\n[assembly-theory-and-the-reuse-mechanism.md](assembly-theory-and-the-reuse-mechanism.md) ·\n[complexity-transitions-and-why-now.md](complexity-transitions-and-why-now.md) ·\n[convergence-wager-red-team.md](convergence-wager-red-team.md) · [what-human-judgment-is-for.md](what-human-judgment-is-for.md).\n"}