{"path":"research/careful-and-broad-minds.md","content":"# Careful and broad at once: what the research says about the two minds Deliberus wants\n\n*2026-10-20. Prompted by his sentence of 19 October, in [causes-of-belief-and-the-three-minds.md](causes-of-belief-and-the-three-minds.md):\n\"The mind of Deliberus can and should be as capable of autistic, careful, meticulous deliberation and\nlogical scaffolding as it is of sly, savvy, spiritual vibe-checks and balances.\" Four research reports\nwere written for it the same day and are kept verbatim in the private notes\n(`.private/research/careful-and-broad-minds-2026-10-20/`: autistic cognition on the detail and gist\naxes; hemispheres, analogy and McGilchrist; the popular images of the word; machines that combine\nexact reasoning with analogy). This note keeps what bears on the design. The word itself, and the\ncandidate phrasings for it, are discussed in [three-minds-and-the-leap.md](three-minds-and-the-leap.md)\n(AM3, *the word autistic*); this note does not repeat that.*\n\n## 1. The word does two jobs in his record\n\nOn 18 October a virtue: a mind *\"more honest and autistic than most of us\"*, one that keeps every\ndistinction. On 30 August a vice: daemons that reason about the graph *\"in a way that isn't autistic\nin some sense, too detail-focused you know, seeing the broad strokes and intuiting what belongs or\ndoesnt\"* ([the-analogy-daemon.md](the-analogy-daemon.md) § 1). The first sense is exactness without\nrounding; the second is the folk picture of detail at the expense of the whole. The research treats\nthe two very differently.\n\n## 2. What the research supports\n\n**The folk picture does not hold as a deficit.** The best meta-analysis (Van der Hallen et al. 2015,\n*Psychological Bulletin*, 56 articles, about 1,000 autistic participants) found no overall local\nadvantage and no overall global deficit. What it did find is that **global processing takes longer**.\nKoldewyn et al. (2013) call it *\"not a disability, but a disinclination\"*: children who reported the\nwhole less when free to choose processed it normally when asked. The field had already moved from\n*deficit* to *processing bias* in 2006 (Happé & Frith). A weaker version survives: some tasks built to\nseparate local from global still show reduced integration (Booth & Happé 2018).\n\n**The virtue is partly supported.** Several lines describe a bottom-up pattern-finding from precise\ndetail (Mottron's veridical mapping among them), close to his *exactness plus seeing what belongs\ntogether*. Most claims of a general rationality advantage failed to replicate in large studies\n(Taylor et al. 2022; Morsanyi & Hamilton 2023; Bastan et al. 2025). **The best-replicated exception\nis a smaller framing effect**: the same choice described as a loss or a gain moves autistic adults\nless (De Martino et al. 2008; it held with alexithymia controlled, Shah et al. 2016). That is the\nprecise sense in which *more honest* has support: less moved by how a question is worded.\n\n**\"Autistic cognition\" as one style is itself contested**, and the popular images (the logical\nandroid, the lack of empathy) carry more than the research: the empathy findings concern cognitive\nempathy, with affective empathy reported intact, and the *double empathy problem* (Milton 2012) reads\nthe communication gap as mutual rather than one-sided.\n\n**The hemispheres do not split this way.** McGilchrist ties autism to the right hemisphere (if you\nwanted to create autism, lesions *\"preponderantly\"* on the right, with *\"a loss of the sense of the\nwhole\"*). The evidence does not support *left is logic, right is analogy*: analogy runs in the\n**left** frontopolar cortex, rising with the semantic distance between the analogy's terms (Green et\nal. 2010), and damage there specifically impairs it (Urbanski et al. 2016); metaphor shows no general\nright-hemisphere advantage (Bohrn et al. 2012, a meta-analysis). What survives as a design idea is\nnot the anatomy but the order: **a wide pass, a focused pass, and a return to the wider picture**,\nand coarse association feeding a separate relational check.\n\n## 3. What machines that do both look like\n\nCognitive science has long modelled the pairing as two stages: **a cheap retrieval that is biased\ntoward surface likeness, then a structural check** (MAC/FAC, Forbus, Gentner & Law 1995). Philosophy's\ntest for a good analogy is the same shape: the known likeness must be associated with the one\nprojected, and the target must show **no critical disanalogy** (Bartha, *Stanford Encyclopedia*).\nMeasured on machines: a 2026 system that matched papers across fields by stripping their vocabulary\nto mechanism \"fingerprints\" more than doubled precision, and still **22 of its 30 top-ranked pairs\nwere not genuine** when judged blind (Kulikowski, JCDL 2026). Broad matching proposes; most of what it\nproposes fails a careful check. No study was found of this pairing applied to argument maps.\n\n## 4. What follows for Deliberus\n\n- **The two minds are passes that check each other, not two characters to balance.** The broad pass\n  (the analogy daemon, kinship across the graph, the *vibe-check*) proposes; the careful pass checks\n  structure before anything is linked. This is the coupling of [coupling-the-poles.md](coupling-the-poles.md),\n  with its return edge, and the research gives it a mechanism: retrieval by surface, acceptance by\n  structure.\n- **Every analogy the daemon proposes carries its mechanism check**: what is projected, what in the\n  source is associated with it, and whether the target has a critical disanalogy. A proposal that\n  cannot answer the third is an organising metaphor, and is labelled one, the distinction of\n  [convictions-across-scales.md](convictions-across-scales.md) (check 2, mechanism identity: *the same mechanism, or only the same shape?*). The base rate to expect, from the one measurement there is: most proposals fail.\n- **Time is part of the careful pass.** The finding that global integration is slower, not absent,\n  suggests the whole is reached by a later pass, not skipped: the order *wide, focused, wide again*.\n- **In public text, describe the mind by what it does.** The research supports *less moved by how a\n  question is worded* and *keeps every distinction*; the label carries the folk picture the evidence\n  does not support. His own words stay his, quoted as his.\n- **Fractal priors set the expectation for kinship across scales**:\n  [fractal-priors-for-convergence.md](fractal-priors-for-convergence.md) holds that structure converges\n  across scales while weights diverge, which is exactly what a structural check can test and a\n  surface match cannot.\n"}