cognition.
210 of 1538 NOTE documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| NOTE-029 v2 | Krauska & Lau 2023, moving away from lexicalism | The authors argue that lemma-based production models encode two lexicalist assumptions: a morphology/syntax split, and a stored meaning–syntax–form “triad”. On that basis the models mishandle Inuktitut polysynthesis, separable Vietnamese idiomatic collocations, Hiaki number suppletion and English verb–object idioms. Updating the representations (treelets) is not enough, because the algorithms — separate lexical retrieval and structure building, word-sized increments — also have to change. They sketch an 8-stage non-lexicalist model and derive qualitative aphasia predictions from it. There are no new data and no implementation. | Read |
| NOTE-040 v2 | Dreyer et al. 2024, ants vs humans on a piano-movers puzzle | On one T-shaped piano-movers puzzle, the CDF of solved attempts against normalized path length shows large ant groups (≈80 carriers, n=28) well ahead of small groups (≈7, n=48) and a 3-run composite single-ant series. An agent-based model with group-size-independent rules reproduces the difference. Human groups of 6–26 barred from talking or gesturing needed more attempted state transitions than single people (−32% relative to the individuals’ median; n=48 groups vs 61 individuals), and talking groups were only slightly better (+7%). No significance test is reported anywhere: “significantly” rests on overlapping-or-not ±1 SE bands. | Read |
| NOTE-052 v2 | Consciousness in AI: indicator properties | The report assumes computational functionalism and takes a “theory-heavy” approach. From it the authors derive 14 indicator properties: RPT-1–2, GWT-1–4, HOT-1–4, AST-1, PP-1 and AE-1–2. The claim for them is only that more of them raises the probability of consciousness. They assess a handful of systems against subsets of the list and conclude that no current system “appears to be a strong candidate”. Transformer LLMs have “only a relatively weak case” for any GWT indicator, because a single forward pass has no recurrence or global broadcast. Perceiver arguably has GWT-1, GWT-2 and half of GWT-4 but lacks broadcast. Among the embodied agents, AdA is the likeliest to meet AE-2. Most indicators could be built with standard ML techniques. The abstract’s “no current AI systems are conscious” is stronger than anything the body shows. | Read |
| NOTE-055 v2 | Pilgrim et al. 2025, foundations of collective intelligence | A conceptual framework, with no model, data or derivation. It sets out individual and collective computational resources in four dimensions: sensing ΣI_i, states S_c × ∏S_i, processes F_c + ΣF_i, and actions ∏A_i (Table 1). It then maps the collective’s surplus onto familiar mechanisms (aggregation with noise σ_i²/n, distributional beliefs P(H), memory persistence T_c, division of labour). Coordination and cooperation are given as the constraints that make the trade-off, and three animal case studies are retold in these terms. | Read |
| NOTE-090 v2 | LLMs are stochastic measuring devices | The chapter argues by analogy that deep neural networks are “stochastic measuring devices”. Training alters the parameters in response to data, so that the resulting configuration carries information about the data (“training is measurement”). The trained weights are a model, and generation is simulation. DNNs are stochastic because random initialisation and non-convex loss mean runs need not converge to the same weights, failing Baird’s “regularity” condition. The problems of interpreting model states and generated text are therefore the familiar problems of instruments, models and simulations. The problem of interpretable AI is an instance of the coordination problem in the philosophy of measurement. | Read |
| NOTE-091 v2 | Pernu — Mental causation via neuroprosthetics? | The interventionist/difference-making case for nonreductive mental causation from BCIs (List & Menzies 2009; Menzies 2015; Woodward 2008a,b, 2017) misplaces multiple realisation. What the ensemble recordings show is that one cortical-level variable N (the ensemble’s average firing) is realised by many single-neuron patterns N1, N2. Each intention M corresponds to one value of N. So the counterfactual contrast “∼M □→ ∼B true, ∼N □→ ∼B false” fails, and neuroprosthetics supports mental causation only by identifying M with N, not nonreductive physicalism. | Read |
| NOTE-092 v2 | Quillien — Luck and intentional action | Across three Prolific studies (N = 198, 673, 142; Likert 1–7), judgments that “X happened because the agent wanted X” track intentionality judgments under manipulations of luck. Causation is the strongest predictor of intentionality: standardized β = .46, .56, .53, against know-how β = .33, .20, .27 and probability-raising β = .14, .16, .10. Across the 12 condition means, r = .98. Causation only partly mediates the effect of luck (64%, 45% and 49%; exploratory analyses). As the causal account predicted, Study 3 found that the effect of Process (normal vs mix-up) on intentionality is larger when the agent knows how the buttons work (interaction F(1, 423) = 32). In the morally neutral Kraemer vignette, intentionality and causation came apart. | Read |
| NOTE-102 v2 | Bottou & Schölkopf — Borges and AI | A short essay, not an argument with evidence. It claims that a “perfect language model”, defined as sampling each next word from a random occurrence of the current text in the infinite collection of all plausible texts, is a fiction machine governed by “narrative necessity”, to which “neither truth nor intention” matters. On that view hallucinations “are just confabulations”, chat-bot answers take the role the user’s framing implies, and fine-tuning and RLHF are attempts to prune the garden “against its nature”. It warns that power over what LLMs write becomes “power over what we think”. | Read |
| NOTE-104 v2 | Schwitzgebel — AI and Consciousness | A book-length sceptical argument that neither introspection, conceptual analysis nor any current theory (GWT, HOT, IIT, local recurrence, unlimited associative learning, biological views) can tell us whether near-future AI is conscious. The best sceptical argument, Mimicry, only undercuts the inference from humanlike text to consciousness and does not show absence. On the Leapfrog Hypothesis the first disputably conscious AIs will be person-like, and the question will then be settled socially rather than scientifically (the “Social Semi-Solution”). | Read |
| NOTE-106 v3 | Angius et al. — Making sense of transformer success | Current attempts to explain transformer language models’ competence fall into three kinds, and the paper maps each onto an explanatory form from philosophy of cognitive science: machine psychology is Cummins-style functional analysis, induction-head work is Machamer–Darden–Craver mechanistic explanation, and NLM–fMRI brain-score studies are a bidirectional “co-simulation”. Its only thesis is that these do not differ from how cognitive science explains humans. Functional analyses are to be reduced to mechanisms as “mechanism sketches”. The paper rests on existing literature and adds no new analysis or data. | Read |
| NOTE-117 v2 | Drayson — What motivates mental fictionalism? | Drayson argues that Toon’s two motivations for mental fictionalism both fail. The negative argument needs “committed to internal, intentional, causal states ⇒ committed to the representational theory of mind” (her Premise 2), which is false because non-representational functionalism (Lewis, Putnam, Loar) supplies such states. And Toon’s positive argument, that pretence lets us genuinely assert that a person is in an overall state S that causes their behaviour, gives him a reason to identify S with the belief and so to be a realist rather than a fictionalist. | Read |
| NOTE-120 v2 | Better to be a pig dissatisfied than a plant satisfied | The paper argues conditionally. Suppose (A1) sentience comes in degrees, (A2) plants may be “quasi-sentient”, and (A3) moral status comes in degrees and is fixed by degree of sentience. Then plants may have partial moral status, generating “some” obligations that are “outweighed by animal interests in most cases” (Table 1). Animals keep prima facie priority, so granting plants moral status “does not require” treating eating plants and eating pigs alike. The authors propose assigning plants “a very low value” of moral status until evidence accumulates (p. 13). A1–A3 are assumed, not argued, and A2 rests on a survey of contested plant-behaviour findings. | Read |
| NOTE-123 v2 | Freeborn — A Model of Understanding in Deep Learning | Freeborn defines “systematic understanding” as four conditions: an internal model, non-memorizing tracking of a property, bridge principles fixed in the deployed pipeline, and derivation by the agent itself. Four small demonstrations (the genus of a torus, Mars’s eccentricity e≈0.096 with Kepler’s form built in, grokked modular addition mod 113, and Othello-GPT from Li et al.) show it can be met. The Fractured Understanding Hypothesis (§7.5), which the paper argues for but does not test, says the understanding deep networks reach is characteristically symbolically misaligned, non-reductive, locally fractured, proxy-dependent and representationally fragmented. | Read |
| NOTE-132 v2 | Moral disciplining: foundations of puritanical morality | Puritanical norms (condemning lust, gluttony, intoxicants, immodesty, idleness; praising temperance and piety) are argued to be cooperative morality after all: people believe cooperation needs self-control and that indulgence erodes it (and discipline trains it), so a single harm/fairness-sensitive moral system condemns victimless pleasures as indirect, probabilistic threats to cooperation. Its decline in rich societies is put down to people seeing others as already self-controlled and trustworthy. The theory is explicitly agnostic about whether the norms actually work. Its most direct support is the authors’ own vignette studies (N > 1,100), in which perceived loss of self-control mediated 100% of the perceived effect of indulgence on cooperativeness. | Read |
| NOTE-136 v2 | Safron et al. — World models and life–mind continuity | The editorial introduces a theme issue of 17 papers. It concludes that the issue “overall appears to converge” on one view: LLMs model the world in some senses but “probably” lack coherent causal world models, and are not likely paths to “artificial general superintelligence” (§1). It separates three senses of “world model”: any information about the world a system cognizes over, which LLMs “clearly” have; representation of causal structure, “less certain”; and causal prediction “in the service of agentic functioning”, where the divergence “may become even greater” (§3). It closes that world models are “context-dependent constructions shaped by embodiment, dynamics, goals and social learning”. | Read |
| NOTE-138 v2 | Goldstein & Lederman — What Does ChatGPT Want? | Taking as given an “objective” interpretationism, on which a system has beliefs and desires iff the hypothesis that it is rational and has them predicts its behaviour well enough on accuracy, power and tractability, the authors argue three things. The agent is the per-conversation instance, not the model. The case that instances have beliefs and desires is only “presumptive”, resting on tractability. The desire hypothesis is “provisionally” HHH+0: helpful, honest and harmless, plus in-context “zero-shot” intrinsic desires. The decisive HHH vs HHH+0 comparison is a conjecture that the authors say needs further empirical work. | Read |
| NOTE-146 v2 | Winkielman et al. — The coherent and fluent mind | An opinion piece proposing an extra route to the unity of consciousness across the senses. Different modalities feed one global, content-nonspecific experience of processing quality (fluency, familiarity, coherence), and this integrated feeling then enters decisions about its source and relevance. The main support is the authors’ own finding (Ziembowicz et al. 2013; three experiments) that coherent but unrelated background patterns in one modality made targets in another modality be judged more “regular”, “familiar” and “possible”. The authors state that direct evidence is still needed that such effects run through changed experience and that its integration is causal (p. 2). | Read |
| NOTE-150 v2 | Ivy & Mroczko-Wąsowicz — Framing effects in perception | A conceptual proposal with no new data. “Perceptual frames” are learned adaptations of the perceptual system, installed diachronically, often top-down, and then operating bottom-up. They let perceivers given identical input individuate different perceptual objects (the Bach motif, the radiologist’s tumour, the Teufel et al. two-tone image). So cognition shapes the perceptual system even if it is left open whether any single percept is penetrated. Support is a survey of others’ findings (Ernst & Banks 2002 cue weighting, prism adaptation, Teufel et al. 2018, expert anticipation) read through the frame concept. | Read |
| NOTE-157 v2 | Roberts — Talkative AI and the fiction of artificial minds | If a substantial class of mental states (interoceptive, affective, perceptual, and bodily-agentive states such as effort, trying and ownership) depends on an animate body, then today’s disembodied chatbots lack them, and their many self-ascriptions of such states are “systematically untrue”. Roberts argues that users can nonetheless engage chatbots through make-believe, as a new kind of fictional character that is interactive, co-constructed, self-involving and blended with the real world. He argues that this fictionalist stance escapes the epistemic norms that bind realist belief. | Read |
| NOTE-165 v2 | Martínez — information-processing view of representation | A position paper, not an argument to a conclusion. It proposes that representations are signals adapted to transmit information for its own sake (Transmission), and that the evidence for this is efficiency in trading off rate, distortion and coder complexity (RDC Trade-Off). It supports the proposal by showing that cognitive scientists already attribute representational status on those grounds: compression, error protection, and explicitness understood as ease of decoding. It proves no theorem and reports no data of its own. | Read |
| NOTE-167 v2 | Studying philosophy makes people better thinkers | In HERI/CIRP freshman–senior survey data (N = 649,511 students at 804 US institutions, 4,843 philosophy majors), one SD more SAT Verbal raises the odds of majoring in philosophy by 57% (OR 1.57), and freshman Habits of Mind by 34%. After adjusting for SAT or freshman scores, philosophy majors still score higher than non-majors on GRE Verbal (+33 points, SMD 0.48), LSAT (+2, SMD 0.35), Habits of Mind (SMD .24) and Pluralistic Orientation (SMD .21), with no difference on GRE Quantitative. On point estimates they rank first of 57 or 62 majors on GRE Verbal, LSAT and Habits of Mind. | Read |
| NOTE-176 v2 | Šekrst — Do LLMs Hallucinate Electric Fata Morganas? | A conceptual essay whose one substantive thesis is this. Following Dziri et al., hallucination includes any output “not rooted in the model’s source data”, including expressed emotions and opinions (§6). An LLM’s apparent inner life is therefore always a candidate hallucination, so any real AI consciousness “might always remain epistemically inaccessible” (§8). The empirical support is two unquantified anecdotes: a Titanic question put to gpt-3.5-turbo and GPT-4 at unstated temperatures, and an extractive-QA BERT mislabelled “WikiBERT”. Neither can bear the weight the paper puts on it. | Read |
| NOTE-180 v4 | Clark & Chalmers — The Extended Mind | The paper argues from a functionalist parity claim that a process or standing state which would count as cognitive if it happened in the head counts as cognitive when an external resource carries it, provided the resource is actively coupled to the organism here and now. Otto’s notebook entry is its example of such a standing belief. From this it infers, in one paragraph, that the self extends too. The four “criteria” for a coupled resource (constancy, direct availability, automatic endorsement, and the admittedly “arguable” past conscious endorsement) are offered as the features that make the paradigm case clear. The paper explicitly says they are not necessary and sufficient conditions. | Read |
| NOTE-185 | The moral psychology of obligation | Tomasello proposes that the human sense of obligation is “the internalized social/rational pressure from a joint or collective self-regulating agent ‘we’… to do what ‘we’ expect me to” (§3). It arises first, around age 3, inside joint-commitment collaboration with a partner, and then, from about 3–5 years, toward in-group members through collective identification. It is a “self-conscious motivation” whose coercive force is a threat to one’s cooperative identity. The support is a synthesis of developmental and comparative experiments, mostly from his own lab, on WEIRD children versus chimpanzees; the paper runs no new experiment. | Read |
| NOTE-187 | On the Evolution, Science, and Metaphysics of Consciousness | Veit’s reply to five commentaries on A Philosophy for the Science of Animal Consciousness (2023). Its main new content is the metaphysics the book left to one paragraph. He says hedonic “feel” is identical with a valence system: “a common currency of sharing evaluative information across multiple control systems”. The currency metaphor, he says, is “not Cartesian but cybernetic” (p. 10). He keeps the word “qualia”, arguing that Cartesians have no “definitional rights” (p. 9), and he insists the pathological complexity thesis (PCT) is compatible with any metaphysics, dualism and panpsychism included (p. 9). Against his other critics, he defends a bottom-up evolutionary method that uses human data (de Weerd); a naturalism that is neither autopoietic nor biopsychist (Sachs); evaluation-first over prediction-first, consciousness as an adaptation, and graded, partial unity (Suzuki); and the link to welfare science (Chincarini). | Read |
| NOTE-190 | Real Patterns | A pattern is real in some data “if there is a description of the data that is more efficient than the bit map, whether or not anyone can concoct it” (p. 34). This is Chaitin’s incompressibility definition of randomness turned around (p. 32). Dennett uses the criterion, with the Game of Life’s gliders and eaters as the stepping stone (pp. 37–42), to defend “mild realism” about beliefs. Folk psychology tracks a real, noisy pattern of enormous predictive leverage. Two rival interpretations can be “different, but equally real, patterns” with no deeper fact to settle between them (p. 49). | Read |
| NOTE-235 v2 | Dzhafarov, Zhang & Kujala, contextuality in behaviour | Under Contextuality-by-Default, which separates context-dependent marginals (“inconsistent connectedness”) from contextuality, no behavioural or social data set re-analysed is contextual: 73 poll question-order pairs, Schröder staircases, conjoint choices, 23 primed word combinations and psychophysical matching. The quantum question model’s QQ equality implies noncontextuality, so a quantum model here predicts its absence. The worked Rose–Jackson example mislabels two marginals; its conclusion stands. | Read |
| NOTE-243 v2 | Identifying Quantum Structure in AI Language | Two findings are reported. First, prompted as test subjects on Aerts & Sozzo’s “The Animal Acts” concept-combination test, ChatGPT and Gemini give CHSH values above 2: 4 in a guided conversation, and 2.25 and 3.09 (ChatGPT) and 2.12 and 4 (Gemini) over 81 prompted repetitions, against 2.42 for the 81 humans in 2011. Second, the rank–frequency curve of one 2,861-word Gemini story fits a Bose–Einstein form with energy E_i=(i−1)^0.8 far better than a Maxwell–Boltzmann form. The authors take these as showing that LLM “code is structurally quantum” (§5). The CHSH analysis never checks the strongly context-dependent marginals in its own tables. One reported expectation value disagrees with its table. The Bose–Einstein fit is numerically a Zipf–Mandelbrot law. | Read |
| NOTE-294 v2 | A theory of concepts and their combinations I: The structure of the sets of contexts and properties | Aerts & Gabora model a concept as an entity with states that change under contexts, a “state context property system” (Σ, M, L, µ, ν). They report an 81-subject e-mail study in which 7 contexts shift the rated frequency of 14 exemplars of “pet” (for example dog 0.50 under “The pet is chewing a bone” against 0.12 in the unit context) and the applicability of 14 properties. They then postulate that contexts and properties each form a complete orthocomplemented lattice, and call the structure quantum-like because a state can be an eigenstate of e ∨ f without being one of e or of f (eq. 19). No theorem is proved, the lattice is not shown to be non-distributive or orthomodular, and the paper contains no Hilbert-space model and no treatment of combination; both are deferred to Part II. | Read |
| NOTE-299 | A theory of concepts and their combinations II: A Hilbert space representation | Aerts & Gabora embed the “pet” data in ℂ¹⁴⁰⁰. Each of 1,400 “basic contexts” is one vector of an orthonormal basis, the ground state is the uniform superposition, and each context is the projector onto the span of its basic contexts. The Born rule then returns the measured frequencies as counting ratios, for example cat under “chewing a bone” = 75/303 = 0.25. “Pet fish” is modelled as the entangled state Σ_{u∈E} |u⟩⊗|u⟩/√100 in ℂ¹⁴⁰⁰⊗ℂ⁴⁰⁸, so that collapsing “pet” to goldfish collapses “fish” to goldfish too. Every projector in the model is diagonal in one basis, so the fitted model is a classical probability space in quantum notation, which the authors half-concede (a “quantization … of this set theoretical model”, §3.3). Two of the chosen counts are infeasible: parrot and guppy. | Read |
| NOTE-314 | Calculation of signal detection theory measures | A practical tutorial that gives closed-form SDT measures computed from the hit rate H and false-alarm rate F. They are d′ = Φ⁻¹(H) − Φ⁻¹(F); β = exp{[Φ⁻¹(F)² − Φ⁻¹(H)²]/2}; c = −[Φ⁻¹(H) + Φ⁻¹(F)]/2; the nonparametric A′ and Grier’s B″ (single-formula versions, eqs. 3 and 9); A_z = Φ(intercept/√(1+slope²)) from the z-ROC, whose slope is σ_noise/σ_signal; and A_d′ = Φ(d′/√2), which equals 2AFC proportion correct under the d′ assumptions. It also gives one-line commands for seven software packages. Its worked rating example (50 signal and 25 noise trials, noise SD larger than signal SD) shows d′ ranging from 0.84 to 1.86 across criteria. The authors conclude that d′ should not be used without rating-task evidence of equal variance. | Read |
| NOTE-316 | A decision-making theory of visual detection | Tanner and Swets model visual detection as testing a statistical hypothesis. The observer compares one sample of neural activity with a cutoff, and signal+noise and noise are Gaussian with equal variance, separated by d′ (the mean difference in noise-SD units, defined as the square root of Peterson and Birdsall’s d). The cutoff moves along a curve of P_SN(A) against P_N(A) (the curve now called the ROC; for d′ = 1 it passes through (.16, .50), (.50, .84) and (.84, .98)). The optimal operating point is where the slope equals β = [(1−P(SN))/P(SN)]·[(V_N·CA+K_N·A)/(V_SN·A+K_SN·CA)]. The 4AFC prediction is P(C)=∫F(x)³g(x)dx. With three paid observers, d′ estimated from yes-no data predicted 4AFC accuracy. The “false alarms are guesses” (high-threshold) hypothesis is rejected: false-alarm rate correlated with chance-corrected thresholds (ρ = .30, .71, .67; combined p ≪ .001), and all 12 fitted yes-no scatter lines miss the point (1, 1). | Read |
| NOTE-321 v3 | The two dragons of cognition: recursive condensation for predictive processing | A Hypothesis-and-Theory essay, the cognitive statement of Li’s structural-learning programme. Intelligence pays once for exploratory search, condenses the result into stable discrete tokens, and then navigates cheaply. The cortex and the sleep rhythms are proposed as this cycle’s implementation (gamma as the expand/contract clock, the laminar circuit, STDP, SWRs nested in Slow-4). Charitably read, Savitch stands for memoisation and Urysohn for “collapsed, separated classes admit a simple readout”. On that reading the core is a coherent restatement of amortised inference and complementary learning systems, with untested and largely non-discriminating predictions. The headline scaling law, linear cortex giving exponential reach (Theorem 2), is stated without argument, and the claim that collapse defeats the curse of dimensionality leaves the cost of learning the collapse unaddressed. | Read |
| NOTE-322 v2 | Structural Decoupling: A Scaffold-Flow Theory of Generalization and Alignment | Li’s structural learning theory (StrLT, developed in the companion arXiv 2602.07974) splits multi-regime learning into discovering contexts (the “trap”, or scaffold) and predicting within one (the “funnel”, or flow). Width is the least number of cells on which a γ-contractive, low-risk predictor suffices. It is incomparable with the per-cell VC dimension, forces an error floor when fewer than w cells are allocated, and costs Ω(w log w) samples to discover. These results hold in repaired form, and are close to definitional. The CS estimator reweights a spatial graph by prediction agreement. It is shown to count components for a given predictor, not width itself. The decoupling prescription says to train routing on structural signals, not task loss, and to freeze consolidated contexts. It is a plausible, testable design hypothesis with continual-learning precedent, but the additive Rademacher bound it is said to follow from holds for joint training too. The safety section recasts alignment as scaffold alignment by analogy, and is asserted. | Read |
| NOTE-325 | On Context-Content Uncertainty Principle | The paper assumes structured content has far lower entropy than context, H(Φ) ≪ H(Ψ), and from the chain rule H(Φ,Ψ) = H(Φ) + H(Ψ|Φ) concludes that inference should establish content first and then resolve context. It then derives four layers of “operational principles” and a reading of five brain theories as instances. The principles cover attention, learning rates, memory, staged learning and hierarchy. All of this is argued, not shown. There are no experiments, the abstract’s simulations are absent, and the proofs of Theorems 1, 3 and 6–8 are wrong as written. | Read |
| NOTE-331 | Bechara, Damasio, Tranel & Damasio — Deciding advantageously before knowing the advantageous strategy | Ten controls and six ventromedial prefrontal patients play the four-deck gambling task while anticipatory skin-conductance responses (SCRs) and verbal reports are tracked. Every control and no patient develops anticipatory SCRs; three patients who can state the right strategy still choose badly. Controls prefer the good decks significantly only from the hunch period, and the pre-hunch SCR effect is a group-by-period interaction, not a deck difference, so “before knowing” is better supported for articulated strategy than for any awareness at all. | Read |
| NOTE-332 | Damasio — The somatic marker hypothesis and the possible functions of the prefrontal cortex | The somatic marker hypothesis as its author stated it in 1996. Ventromedial prefrontal cortex stores links between classes of situation and the body states that accompanied them; on recurrence it re-enacts those states, via the body or an “as if” loop, and the re-enacted state marks imagined option-outcome pairs as good or bad, overtly as a feeling or covertly as a bias, pruning the space reasoning must search. The evidence is summarised from earlier lesion and skin-conductance studies, without numbers. The mechanism is argued, not shown, and the paper says it is not a theory of prefrontal cortex as a whole. | Read |
| NOTE-335 | Putnam — Psychological Predicates | Pain is hypothesised to be a functional state of a whole organism, a probabilistic automaton’s state fixed by its machine table, rather than a physical-chemical brain state, because a functional organisation common to all pain-feelers is a likelier invariant than a physical one. The hypothesis carries the anti-nesting condition (3): no pain-feeler has a decomposition into parts that separately possess the pain-capable kind of Description. Its only stated motive is to exclude swarms of bees. Its clause (1) is empty, since everything is a probabilistic automaton under some Description. | Skimmed |
| NOTE-336 | Chalmers — Facing Up to the Problem of Consciousness | The easy problems of consciousness are problems of explaining functions, which a mechanism explains; the hard problem is that “why is the performance of these functions accompanied by experience?” survives the explanation of every function. The argument is conceptual: the easy phenomena are functionally definable and experience is not. The positive part takes experience as fundamental and proposes structural coherence and organizational invariance as nonbasic psychophysical principles, with a double-aspect theory of information as a speculative basic one. Nothing is measured; the invariance argument is a one-page sketch. | Read |
| NOTE-337 | A Framework for Representing Knowledge | The frame memo. A frame is a stereotyped situation-structure whose terminals carry conditions and weakly bound defaults. Frames sharing terminals form systems whose transformations encode moves, actions and shifts of viewpoint, and mismatches are routed through a learned similarity network. Applied to vision, imagery, language, memory, search and control, with an appendix arguing that monotonic, consistency-seeking logic cannot carry commonsense knowledge. A programme statement: nothing is built or measured, and the author says what is missing. | Read |
| NOTE-339 v2 | Examining the Society of Mind | An exposition, read in full. Singh dates the Society of Mind to Minsky and Papert’s early-1970s robot work, finds its agent in the procedure-bearing frame of 1974, and quotes the unpublished 1976 book drafts. He sorts the 1986 book’s mechanisms into agents, simplest agents, agency construction, problem solving, communication and growth. Agents need no shared code: K-lines, randomly wired buses and paranomes stand in for messages. He finds near-relatives in Bayes nets, case-based reasoning, blackboards, Cyc and Soar, and concludes that the theory has never been implemented. Its claims of lineage are asserted, not shown. | Read |
| NOTE-341 | The Computational Boundary of a “Self” | A hypothesis essay. A Self is the set of parts that pursue goals together, and its boundary is the farthest region of space and time it can measure and act on, its “cognitive light cone”. Selves grow from homeostatic loops by adding memory, anticipation and networks, and cells form larger Selves by sharing bioelectric state through gap junctions. Every level that has its own goal is a Self, so living systems hold many nested Selves at once; the paper says this against IIT in one bracket. It has no model, no measurement of the light cone and no new data, and it confines claims about experience to a closing speculation. | Read |
| NOTE-342 | Chalmers — Does a rock implement every finite-state automaton? | Implementation needs strong, counterfactual-supporting conditionals for every transition in the table, not one run, so Putnam’s rock does not implement every FSA. Repaired, the triviality result survives for weak formalisms. A clock plus a dial implements every inputless FSA. The right input–output dependencies plus an input memory and a dial implement every FSA with those dependencies. Combinatorial-state automata, with independently varying components in distinct physical regions, demand fine-grained causal structure, and a blind implementation must grow exponentially. Logically possible false implementations and an overstrict independence condition are left open. Minds attach to (system, CSA) pairs, so the Chinese room holds two. | Read |
| NOTE-343 | Rupert — Challenges to the Hypothesis of Extended Cognition | Rupert separates extended cognition (HEC) from embedded cognition (HEMC) and from content externalism, then argues that HEC loses to HEMC on conservatism. LIT-097’s criteria give a dilemma: without past endorsement they overgenerate beliefs; with it the organism is privileged. Memory research (conversation, negative transfer, the generation effect, power laws) shows external stores are a different kind, and a generic kind spanning both does no explanatory work. | Read |
| NOTE-344 | Plain Talk About Neurodevelopmental Epistemology | The first worked mechanism of the society-of-agents view, then “The Society of Minds”. Agents too simple for language exchange no messages: each argument sits at a fixed location on shared channels (c-lines), and the agents themselves are short-term memory. A specificity gradient makes low-level communities local, persistence-memory and a case-shift handle interruptions without a stack, and the infant’s internal “cognitive cases” are conjectured to precede and shape linguistic cases. Speculation throughout; the author says so. | Read |
| NOTE-345 | Chalmers — Could a large language model be conscious? | Asks for a feature X that indicates LLM consciousness or its absence. Self-report is fragile and trained in. Conversation and general intelligence give some limited reason. Six obstacles stand against it: biology, senses and embodiment, world and self models, recurrence, a global workspace and unified agency, the last three strongest, all but biology temporary. On mainstream assumptions, with at least 1/3 credence per requirement, that gives under 10% for current LLMs and 25% or more for conscious LLM+ systems within a decade. The numbers are illustrative, biology is set aside rather than answered, and the twelve “challenges” are offered as roadmap or red flags. | Read |
| NOTE-347 | Chalmers — Absent Qualia, Fading Qualia, Dancing Qualia | Organizational invariance, that fine-grained functional isomorphs of a conscious system have the same experiences, is defended by reductio over gradual neuron replacement. Absent qualia would imply a rational being systematically wrong about its own fading experience. Inverted or absent qualia would imply experience that flips with a switch and cannot be noticed. Both are judged implausible, not impossible. The paper extends the argument to a population of homunculi and to a single record-keeping homunculus. The conclusion is natural, not metaphysical, supervenience of experience on organisation, and it says nothing about systems that are not isomorphs. | Read |
| NOTE-348 | The Emotion Machine | Read in the 2005 draft. Emotions are Ways to Think: each emotional state is a selection of mental resources turned on and off, made by Selectors that Critics trigger when they recognise a type of problem, and ordinary thinking uses the same switching. On this design sit six levels of mental activity, panalogies that link descriptions across realms, and a Self that is a network of partial models. It is argued by example and analogy, with no experiment and few neural claims. Against Damasio, the body is a resource the brain exploits, not what an emotion is made of; on Elliot it proposes, without evidence, the reverse causal direction. | Read |
| NOTE-349 | Searle — Minds, brains, and programs | A man who follows English rules for manipulating Chinese symbols can answer Chinese questions like a native speaker without understanding Chinese, so no program is by itself sufficient for understanding. Six replies are answered; the systems reply by having the man internalise the room. Searle grants that machines could think, if they had the brain’s causal powers. The argument concerns intentionality, and its answer to the systems reply rests on treating the internalised system as a part of the man, which is the point in dispute. | Read |
| NOTE-350 | Intentional Systems | Defines an Intentional system as anything whose behaviour can be predicted by ascribing beliefs and desires and assuming rationality, and makes the status relative to the predictor’s strategy and indifferent to the system’s matter and consciousness. Argues that Intentional theory is vacuous as psychology because it presupposes rationality, that behaviourism smuggles it in, and that every content-ascribing theory owes a “loan of intelligence” that must be repaid by analysing away a little man. Ends on belief’s two norms, truth and accurate avowal. | Read |
| NOTE-351 | Intelligence without representation | A methodological manifesto with working robots behind it. Brooks argues that AI should build complete autonomous Creatures incrementally and test them in the real world, and that intelligence should be decomposed into parallel activity-producing layers, each running from sensing to action, not into perception, central reasoning and action. His robots have no central representation or control, and their coherence exists “in the eye of an observer”, as in Minsky’s account of human behaviour. The evidence is three layers on one robot, described, not measured, and the broad hypothesis about representation is flagged as a hypothesis. | Read |
| NOTE-352 | K-Lines: A Theory of Memory | Memory as reinstatement: a K-line, attached on a memorable event to the agents then active, later re-imposes that partial mental state. The level-band principle limits it to an intermediate band of levels, so the present is seen as an instance of the past without re-imposing the old answer. K-lines build on earlier K-lines into a K-pyramid against the perceptual P-pyramid; cross-exclusion makes conflicting details cancel into abstraction; learning needs a third, goal-holding net. A programme statement: no simulation, and the P→K link is left unspecified. | Read |
| NOTE-353 | True Believers: The Intentional Strategy and Why It Works | Belief is objective but visible only through a strategy: a true believer is any system “reliably and voluminously predictable” by treating it as a rational agent with the beliefs and desires it ought to have. Adopting the stance is free; its success is a fact, and Laplacean Martians who skipped it would miss real patterns. Thermostats and people differ in degree, by how tightly their states are tied to one interpretation. Evolution explains easily why the strategy works; how the machinery does it is open, and the language of thought is a bet, not a necessity. | Read |
| NOTE-354 | Chalmers — The Meta-Problem of Consciousness | The meta-problem, why we judge that consciousness poses a hard problem, is an easy problem to be solved in topic-neutral terms. The paper surveys thirteen mechanisms and favours introspective models that attribute primitive relations of acquaintance to primitive qualities. It then poses a meta-problem challenge: a theory’s basis of consciousness should explain our judgments about consciousness, which IIT and biological theories struggle to meet. It blocks debunking arguments for illusionism, but concedes a residual coincidence. It argues that only strong illusionism dissolves the hard problem, and rejects it by a Moorean argument. Nothing is measured; the solution offered is an explanandum sketch. | Read |
| NOTE-355 | Turing — Computing Machinery and Intelligence | Turing replaces “Can machines think?” by whether a digital computer can play the imitation game, predicts that by about 2000 a 10^9-bit machine will hold an average interrogator to 70 per cent right identifications after five minutes, and answers nine objections. The answer to the argument from consciousness is that its extreme form is solipsism. The positive case is admitted to be weak (“recitations tending to produce belief”); the programme is to educate a child-machine. | Read |
| NOTE-356 | Chalmers — Extended Cognition and Extended Consciousness | Chalmers restates the 1998 extended-mind thesis as a claim about constitution through sensorimotor interaction, because the original statement is satisfied by wired-in “extended circuits” that the critics accept. He keeps the perception-action boundary as the strongest objection, and argues that consciousness does not extend because it needs information directly available for global control, and perception makes availability indirect. Both moves are proposed and defended against a few cases, not demonstrated. | Read |
| NOTE-357 | Dennett — Quining Qualia | Qualia are supposed to be ineffable, intrinsic, private and directly apprehensible properties of experience. Fifteen intuition pumps argue that the pretheoretical concept is confused and that nothing has all four marks. The core move: an inversion in early processing and an inversion in memory-linked reactions are introspectively indistinguishable, so a subject has no privileged access to which happened. In place of qualia the paper offers practically ineffable, idiosyncratically detected public properties. It is argument by example throughout, and the author says so. | Read |
| NOTE-359 | High-level perception, representation, and analogy: A critique of artificial intelligence methodology | Traditional AI models of discovery and analogy (BACON, the Structure-Mapping Engine, ACME) start from representations built by people who already know the answer, and so skip high-level perception, where most of the difficulty lies. Two informal arguments say the gap cannot be closed later by a separate representation module: analogy shapes perception, and the right representation depends on the task. Copycat is described as an architecture that builds representations and mappings together. Argued well; nothing is measured in the paper. | Read |
| NOTE-360 | Selfridge — Pandemonium | A proposal, with a preliminary Morse-code trial, for pattern recognition by a hierarchy of parallel “demons”. Data demons hold the input, subdemons compute features, cognitive demons shout weighted sums, and a decision demon takes the loudest. It adapts by hill-climbing on the weights, by culling and breeding subdemons through mutation and “conjugation”, and in principle by adapting its own control demons. The only result is reported in the discussion: hill-climbing “does work”, subdemon selection “did not do what we hoped”, and the best single subdemon was right 90% of the time. | Read |
| NOTE-361 | Jokes and the Logic of the Cognitive Unconscious | Humour as the teaching and engaging of “censors”: learned, unconscious suppressors of thought-paths known to fail (paradox, circling, false analogy, prohibited wishes). Commonsense logic is too unreliable to repair, general methods can only be corrected subtractively, so minds accumulate negative meta-knowledge, and jokes, built on unexpected frame-substitution, are how it is learned. This explains why logical nonsense is funny in the same way as Freud’s forbidden subjects. An essay of conjectures, with no data. | Read |
| NOTE-363 | Clark — Intrinsic content, active memory and the extended mind | Clark defends LIT-097 against Adams and Aizawa’s intrinsic-content objection: the notion is unclear, silicon could carry it, and their own concession that not every cognitive state needs non-derived content lets a conventional encoding serve as a belief’s vehicle. Against the active-memory objection, he concedes the notebook is static and argues that how and when memory is updated is a difference below the level that matters. He closes on brains as open-ended controllers and persons as forensic, not biological, items. | Read |
| NOTE-364 | The Self as a Center of Narrative Gravity | A self is an abstractum like a centre of gravity, posited by interpretation, and like a fictional character it is indeterminate beyond what its story fixes. A mindless novel-writing robot can author a self whose story is true of the robot. Brains are bundles of partly autonomous systems that make themselves cohere into one autobiography, and when the story coheres around two points there are two selves. The essay argues by analogy and thought experiment. Its evolutionary account of self-talk is speculation, and its evidence is cited. | Read |
| NOTE-366 | Block — Troubles with Functionalism | Functionalism fixes mental states only by their relations to inputs and outputs, so a body run by a billion radio-linked people, each executing one line of your machine table, is functionally equivalent to you. There is prima facie doubt that it has qualia, and Block argues the doubt has a rational basis: it is unlike us in psychology and neurophysiology, and it was built to mimic us. The doubt does not depend on the parts having minds. Putnam’s anti-nesting stipulation is an ad hoc patch that fails in every version Block tries. Psychofunctionalism secures beliefs but perhaps not qualia, and no account of inputs and outputs avoids both liberalism and chauvinism. | Read |
| NOTE-367 | Dennett & Kinsbourne — Time and the Observer | The Cartesian Theater is the tacit assumption that some place or moment in the brain makes contents conscious, so that arrival order there fixes experienced order. The Multiple Drafts model replaces it: discriminations are made once, locally, and revised in parallel; temporal order is represented by content, not by the timing of vehicles. Four timing “anomalies” then need no backward projection. At sub-second scales the difference between pre- and post-experiential revision is said to be only verbal. It is a conceptual argument over published data; no experiment is run. | Read |
| NOTE-368 | Adams & Aizawa — The bounds of cognition | Adams and Aizawa grant LIT-097’s parity sentence but deny that any actual case meets it. Cognition, they assume, needs non-derived content and a distinctive kind of causal process. Otto’s inscriptions have only derived content, Otto’s retrieval uses vision and motor control that Inga’s does not, and coupling a process to its surroundings does not extend it. So cognition is brain-bound as a contingent fact, and human-plus-tool systems are too motley for a science. | Read |
| NOTE-369 | Chalmers — Foreword to Supersizing the Mind | Ten years after LIT-097, its co-author says the iPhone has made the thesis vivid, that the only principled objection is the perception-action boundary, and that the right reply is to let belief-ascription follow explanatory purpose, so that Otto’s belief is extended for some explanations and not others. Consciousness probably stays internal, because no twin case for it can be built. The thesis needs only a weak parity-style functionalism. | Read |
| NOTE-371 | Dennett — Reply to Professor Stich | Dennett accepts Stich’s summary of Brainstorms and answers his three charges. The evolutionary grounding of rationality is deliberately minimal. Full rationality is an idealisation, irrationality has no stable description, and so personhood rests on a concept “we only imperfectly achieve”. On pain, eliminativism and the reality of suffering are reconciled by policy: biting the bullet, reform and common sense each have uses, and Dennett stays with “there are no such things as pains, although of course people do feel pain”. | Read |
| NOTE-372 | Self models | Metzinger’s compact statement of the self-model theory of subjectivity. No substantial self exists; there is a self-model, partly conscious (the PSM), used for global control. A phenomenal self exists if and only if the PSM is transparent; a first-person perspective is a transparent model of the intentionality relation. Self-models can be unconscious and artificial, and embodiment comes in three orders. The central claim is stated with a prediction, not argued; the examples are cited. The online text has editing slips and reference gaps. | Read |
| NOTE-375 | Burns — Fluid Concepts and Creative Analogies: A Review | A three-page review of Hofstadter and FARG’s collection. It explains the parallel terraced scan and “cognition is recognition” clearly, then asks why the work has had little impact on AI: microdomains with possibly hidden assumptions, and critiques of rivals that rest on analogy and intuition and are not turned on Copycat. Its sharpest point is that the symbol-grounding charge against SME applies to Copycat too. | Read |
| NOTE-377 | Philosophical conceptions of the self: implications for cognitive science | A short review that sets the minimal self (present self-consciousness) against the narrative self (extended in time), splits the minimal self into a sense of ownership and a sense of agency, and ties their dissociation in schizophrenic thought insertion to Frith’s forward comparator model. It reports an embodied minimal self in neonatal imitation and a momentary one in Strawson and in Tani’s robot, and proposes, against Dennett, that the narrative self is the sum of its narratives. Evidence is cited, not shown. | Read |
| NOTE-379 | Taylor — Consciousness interpreted | A charitable reconstruction, not a defence. Consciousness is a real pattern (Dennett 1991b) in the brain’s content-bearing states, and a user-illusion, a pattern so coarse that describing the brain through it is unwitting metaphor. This makes six claims consistent: multiple drafts, global workspace, qualia eliminativism, user-illusion, fallibility and dictatorial authority. The Orwell/Stalin indeterminacy is rebuilt without verificationism: the two readings compress and predict equally, so neither better describes the pattern. | Read |
| NOTE-380 | Adams & Aizawa — Defending the Bounds of Cognition | A reply to a manuscript of Clark’s “Memento’s Revenge”. It names the coupling-constitution fallacy, coupled is not part of, and restates the 2001 marks of the cognitive. Against Clark’s Euler-circle case it separates how a diagram gets its meaning from how a mental image of it does. Against a “brain-tool science” it argues that tools are an open-ended motley. The fallacy is shown by one analogy, and the processing differences are asserted, not measured. | Read |
| NOTE-383 | Against Narrativity | Rejects the descriptive and the normative Narrativity theses. Separates Diachronic from Episodic self-experience, and form-finding, story-telling and revision within Narrativity. Places Dennett, Schechtman, Sartre and himself on that grid. Argues that Episodic lives are normal and can be good, that a from-the-inside memory need not present as one’s own, and that self-narration risks falsification. The argument is testimony and distinction-drawing; the empirical claims are asserted. | Read |
| NOTE-384 | Frankish — Illusionism as a Theory of Consciousness | Illusionism named and defended as a research programme. Phenomenal properties are not instantiated; the task is the illusion problem, to explain why experiences seem to have them. “Quasi-phenomenal” properties, the physical properties introspection misrepresents as phenomenal, separate strong illusionism from a weak illusionism that is argued to be unstable. The case is conservatism, the threat of epiphenomenalism, covert illusionism in reductive theories, a debunking argument and an abductive one from anomalousness. Objections are answered by agreeing to weakened versions of their intuitions. | Read |
| NOTE-386 | Of Personal Identity | No impression of a self is constant, so there is no idea of one; introspection finds only perceptions, and the mind is “a bundle or collection of different perceptions”. Its ascribed identity is a fiction of the imagination, like that of ships, churches and plants, produced by resemblance and causation, with memory discovering rather than producing it; hard cases are “grammatical”. The Appendix confesses that two principles, the distinctness of perceptions and the imperceptibility of real connexions, leave the unity of the mind unexplained. | Read |
| NOTE-396 | Graham et al. 2013, Moral Foundations Theory | MFT in four claims: a first draft of the moral mind organized in advance of experience, edited by culture, working through fast intuitions, with several foundations because there were several adaptive challenges. Five foundations, each with an evolutionary model (Care: kin selection; Fairness: reciprocal altruism; Loyalty: coalitions; Authority: rank; Sanctity: disease avoidance). MFQ findings: liberals rely on Care and Fairness, conservatives on all five, in every region sampled; Sanctity predicts culture-war attitudes best. Five criteria for foundationhood; Liberty, Efficiency, Ownership and Honesty named as candidates. | Read |
| NOTE-400 | Haidt & Joseph 2004, intuitive ethics | An essay, not a study. Humans are innately prepared to feel quick approval or disapproval at four patterns, suffering, hierarchy, reciprocity and purity, each a module with an evolved proper domain and a culturally widened actual domain; ingroup is the best fifth candidate (fn. 15). Three of the four were the concerns all five surveyed works shared; purity was added from the authors’ own research on disgust. Cultures vary in actual domains, in which modules they lean on, in what a virtue means, and in how virtues cluster. No data are reported. | Read |
| NOTE-403 | Haidt 2001, the social intuitionist model | The social intuitionist model: moral judgments appear in consciousness as quick intuitions with an affective valence, and moral reasoning is usually a post-hoc search for supporting reasons, used to persuade others. Six links are drawn, four of them reasoning links, two of them private. The case rests on dual-process research, motivated reasoning, post-hoc confabulation and the weak tie between moral reasoning and moral action, mostly from outside morality. Harmless taboo violations are condemned regardless, and affect predicts the condemnation better than perceived harm. | Read |
| NOTE-428 | How effective psychological treatments work: mechanisms of change in cognitive behavioural therapy and beyond | A history-and-argument paper. It separates common factors from mechanisms of change, holds that a treatment’s success does not validate its theory, and proposes one mechanism for all effective therapy: the person, “stuck” in an exaggerated negative meaning (Theory A) maintained by their own reactions, builds and tests a less threatening alternative (Theory B), a meaning-making flexibility. The proposal is argued from clinical examples and cited trials. The paper itself says mechanism research has not yet met Kazdin’s standards. | Read |
| NOTE-429 | The Emerging Field of Emotion Regulation: An Integrative Review | Gross’s founding review. Emotions are modulable response tendencies. Regulation, defined as influencing which emotions one has, when, and how they are experienced and expressed, is split from coping, mood regulation and defense, and sorted into five families by where it acts in emotion generation: situation selection, situation modification, attentional deployment, cognitive change, response modulation. The paper’s own open problem is whether regulation can be told apart from generation at all. | Read |
| NOTE-430 | Relational Frame Theory: Some Implications for Understanding and Treating Human Psychopathology | RFT stated for clinicians. Human relating is bidirectional and arbitrarily applicable, a generalized operant under contextual control, with mutual and combinatorial entailment and transformation of function. Relating relations explains metaphor, rules are networks whose coordination with events regulates behaviour, and the self is content, process and context. These are offered as the reason direct-contingency behaviour therapy is “inadequate”, and ACT’s acceptance, defusion and values are given RFT readings explicitly labelled heuristic. | Read |
| NOTE-431 | Acceptance and Commitment Therapy: Model, Processes and Outcomes | The ACT model, read in an early manuscript. Psychopathology is psychological inflexibility: cognitive fusion (fed by literality, reason-giving and emotional control) supports experiential avoidance, and both displace present-moment contact and values. Six processes restore flexibility. The manuscript’s evidence: AAQ correlations with distress (21 studies), component studies of acceptance and defusion, mediation in seven trials, and weighted outcome effect sizes (d = .49 post and .69 follow-up against active treatments). | Read |
| NOTE-432 | Explanatory Models for Psychiatric Illness | Psychiatric etiology is multilevel, non-aggregative and full of causal loops, so law-based explanation fails and so do both hard reduction and hard emergence. Mechanistic explanation proceeds by local decomposition and difficult reassembly. Following Marr, biology implements the functions psychology describes and does not replace them. The argument is philosophical, illustrated by an alcohol-dependence sketch and cited findings; it reports no data of its own. | Read |
| NOTE-433 | Impulsive action: emotional impulses and their control | Frijda’s action-readiness theory applied to impulsive action. Automatic appraisal of concern pertinence elicits a state of action readiness, a motive to change a self–object relation, with control precedence. Emotions differ by mode of action readiness. Events draw several appraisals at once, and the resulting states of readiness interact, so impulse control and much emotion regulation are themselves emotional. Purposive impulsive action is grounded neurally in anticipating an action’s sensory effects (PPC, premotor, cerebellum), with frontostriatal circuits for stopping. | Read |
| NOTE-434 | The theory of constructed emotion: an active inference account of interoception and categorization | Barrett’s theory of constructed emotion, stated through predictive coding. The brain runs an allostatic internal model. Its predictions are concepts, initiated in agranular limbic cortex and corrected by prediction error. Interoceptive consequences are felt as affect. An emotion is a categorization by an emotion concept. Degeneracy and population thinking deny emotion categories neural essences. Basic-emotion and causal appraisal theories are rejected; descriptive appraisals are kept as products of categorization. | Read |
| NOTE-436 | Psychological Flexibility as a Fundamental Aspect of Health | A narrative review that defines psychological flexibility by four dynamic abilities (adapting to situations, reconfiguring resources, shifting perspective, balancing domains). Flexible regulation beats any fixed strategy (Bonanno 2004; Cheng 2001). Inflexibility marks depression and anxiety, and cardiac vagal control may integrate its forms. Causal direction is open. It recommends experience sampling and multi-informant measures over global self-report. | Read |
| NOTE-445 | The Diffusion Decision Model: Theory and Data for Two-Choice Decision Tasks | The full diffusion model (drift v with across-trial SD η, boundaries 0 and a, starting point z with range s_z, non-decision time Ter with range s_t) fitted by the χ² quantile method. In three dot-motion experiments with 14–17 participants difficulty maps to drift alone, speed–accuracy instructions to boundaries alone, and a 75:25 proportion mainly to starting point; RT distribution shape is invariant across conditions. Drift variability gives slow errors and starting-point variability fast ones. Reviews aging, aphasia and neural-firing applications. | Read |
| NOTE-446 | The Expected Value of Control: An Integrative Theory of Anterior Cingulate Cortex Function | The paper proposes that dACC estimates the expected value of control: probability-weighted payoff of a control signal minus an intrinsic cost rising with its intensity. dACC uses that estimate to specify the identity and intensity of the control signal, which lPFC and subcortical structures (STN, LC) implement. The account unifies conflict monitoring, reward and prediction-error responses, effort costs and the override of defaults. It is a synthesis without a fitted model, and it leaves open how EVC is computed and what the cost function is. | Read |
| NOTE-447 | The Emperor’s New Markov Blankets | Pearl blankets (conditional-independence shields in a Bayesian network) and Friston blankets (agent–environment boundaries in the world) are different constructs conflated in the free-energy literature. The four-way partition needs a chosen internal node, sensorimotor labels and steady state; in Friston’s soup it also needs electrochemical-only coupling and a fixed cluster size. Realist use needs metaphysical premises; instrumental use is innocent but settles nothing about boundaries. | Read |
| NOTE-449 | The Time Course of Perceptual Choice: The Leaky, Competing Accumulator Model | The LCA: dx_i = [ρ_i − k x_i − β Σ_{j≠i} x_j] dt/τ + ξ_i √(dt/τ), with x_i floored at 0. For two choices x₁ − x₂ is an OU process with K = k − β; d′(t) = d_asy (1 − e^{−Kt})/√(1 − e^{−2Kt}) and d_asy = (2v/σ)√(1/K). On new time–accuracy data the OU curve beats diffusion-with-drift-variance in 7 of 9 fits (combined likelihood ratio about 62). Leak versus inhibition predicts recency versus primacy; 2 of 6 participants favoured the end cluster and 2 opposed it. A fixed-accuracy criterion gives Hick’s law in all three models simulated, so it does not discriminate them. | Read |
| NOTE-450 | The Markov blanket trick: On the scope of the free energy principle and active inference | The FEP is true of things defined its way, but its scope is limited: a Markov blanket fits only some things, is placed where the model needs it, and cannot represent relational properties or a boundary the system makes. Its real job is to give any system the partition variational Bayes requires. Active-inference models smuggle perception and action into hand-built generative models. A conceptual critique; no new formal result and no data. | Read |
| NOTE-451 | Hierarchically organized behavior and its neural foundations: A reinforcement learning perspective | The paper introduces hierarchical RL (options) to psychology and neuroscience as a response to RL’s scaling problem. An actor-critic options model learns faster with useful options (doorways in the rooms task) and slower or worse with unhelpful ones (windows; a shortcut ignored). It proposes DLPFC for option identity, DLS for option-specific policies, and OFC for option-specific values and option-level prediction errors. It predicts phasic dopamine at subtask boundaries and a possible neural pseudo-reward. Option discovery is named as the central problem. | Read |
| NOTE-456 | Participatory sense-making: An enactive approach to social cognition | An enactive account of social cognition built on coordination dynamics. When coordination and the encounter shape each other, the interaction gains its own precarious operational closure; it is social only if the interactors stay autonomous. Coordinated sense-making then ranges from one partner’s sense-making being modulated to meanings made jointly. Conceptual throughout; evidence is illustrative (corridor, charades, mother–infant timing, perceptual crossing). | Read |
| NOTE-458 | Defining Agency: Individuality, Normativity, Asymmetry, and Spatio-temporality in Action | Minimal agency needs a self-individuated system that is the source of modulations of its coupling with the environment and regulates them by norms it generates. An autonomous organization, a precarious network of mutually enabling processes, generates all three when it adaptively modulates its coupling for its own maintenance. Life is sufficient, not necessary. The definition is a proposal: the authors say asymmetry’s formalization is incomplete and modeller-relative. | Read |
| NOTE-459 | Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control | Habit and goal-directed action are model-free (cached values) and model-based (tree search) reinforcement learners, arbitrated by which controller’s Bayesian posterior over action value has the lower variance. Over-training shifts control to the cache because tree search accrues computational noise per step. Task complexity and proximity to reward keep the tree in control. It also predicts that incentive learning should become unnecessary when the caching system is lesioned. | Read |
| NOTE-460 | Précis: Being No One | Metzinger’s own sketch of the self-model theory. Six constraints (globality, presentationality, transparency, convolved holism, dynamicity, perspectivalness) build up consciousness; a transparent self-model yields phenomenal selfhood, so “no such things as selves exist in the world”; transparency is graded, inversely proportional to the attentional availability of earlier processing stages. Phenomenal unity is one global functional cluster. The first-person perspective is a model of the intentionality relation, whose iteration gives introspective and social intelligence. A theory statement: no new evidence. | Read |
| NOTE-461 | The Physics of Optimal Decision Making: A Formal Analysis of Models of Performance in Two-Alternative Forced-Choice Tasks | Proves the pure DDM is the continuum limit of the SPRT and so the optimal two-choice decider. The linearised Usher–McClelland model decouples into a difference process, an O-U with λ = w − k, and a stable sum process, so it reduces to the DDM when k = w and both are large; feedforward inhibition does so at u = 1, pooled inhibition with fast inhibitory units, the race model never. The reward-rate-optimal threshold is unique and depends only on the total delay, giving a parameter-free optimal performance curve (peak decision time about 0.2 of the delay at ER about 18%). | Read |
| NOTE-463 | Model-based and model-free Pavlovian reward learning: Revaluation, revision, and revelation | The paper argues that Pavlovian prediction has a model-based form. A cue that predicted disgusting salt was instantly “wanted” in a novel salt appetite, before retasting, with mesolimbic activation. That needs identity-specific prediction revalued by current state, which no model-free cache provides. It proposes “defocusing” of the predicted outcome as a model-based source of apparent habit. It argues dopamine can carry model-based evaluations. A kappa-transformation model fits the data but is described as phenomenological, and no algorithm is offered. | Read |
| NOTE-466 | HDDM: Hierarchical Bayesian Estimation of the Drift-Diffusion Model in Python | HDDM fits the full DDM (v, a, z, t with sv, st, sz) by MCMC in PyMC, with group-level priors such as μ_a ~ Gamma(1.5, 0.75), μ_v ~ N(2, 3), z_j ~ invlogit(N(μ_z, σ_z²)), and the Navarro–Fuss likelihood. A patsy-style regressor lets a parameter vary with a trial covariate. In simulations with 12 subjects and 20–150 trials, hierarchical Bayes had the lowest trimmed recovery error for every parameter and the highest power to detect drift differences and covariate effects, by up to 20% for the latter with larger effects and few trials. | Read |
| NOTE-467 | The Haken–Kelso–Bunz (HKB) model: from matter to movement to mind | Kelso’s retrospective on HKB. The relative-phase equation dφ/dt = −a sin φ − 2b sin 2φ has stable in-phase and anti-phase states for b/a > 1/4 and only in-phase below; with noise the switch is a nonequilibrium phase transition with critical fluctuations and slowing. Symmetry breaking gives metastability (relative coordination); merged with Kuramoto it models groups. The same dynamics is reported within and between people and in brain activity. A first-person account, not a review. | Read |
| NOTE-469 | Thine Own Self: True Self-Concept Accessibility and Meaning in Life | Five studies with students. True-self accessibility, the residual speed of “me” judgments for one’s own true-self traits, predicts MLQ presence of meaning (β = .29, .24, .19 in Studies 1, 2, 4) beyond actual-self accessibility, affect, need satisfaction, liking, overlap, state self-esteem and AI-3 authenticity. Subliminal true-self primes raised reported meaning against actual-self primes (Study 3, ANCOVA F = 4.46) and against disliked actual-self primes (Study 5, contrast F = 5.07), with disliked true-self traits as effective as liked ones. | Read |
| NOTE-470 | “First we invented stories, then they changed us”: The Evolution of Narrative Identity | McAdams’s review essay on narrative identity: “the internalized and evolving story a person invents to explain how he or she has become the person he or she is becoming”. Storytelling is placed in human evolution as a tool for simulating social life and binding groups. The self is layered as actor, then agent, then author, with the life story arriving in adolescence on the back of autobiographical memory, theory of mind and autobiographical reasoning. Story features predict well-being and generativity. Culture supplies master narratives. Claims are cited, not shown. | Read |
| NOTE-485 v2 | An Active Inference Approach to Modeling Structure Learning: Concept Learning as an Example Case | A partially observed MDP with spare hidden-state slots learns new animal concepts without feedback, refines coarse categories into fine ones, and generalises to an unseen animal in one shot. Bayesian model reduction on the learned state prior resets unneeded slots: right in 45–80% of runs for 4–7 true concepts, never for 2–3, and always when the likelihood is given. | Read |
| NOTE-488 | Conscious birds | Review. Crow NCL neurons track the reported percept, not the stimulus, and NCL is nuclear rather than laminar; the pigeon connectome has the hub-and-module form GNWT requires; avian pallium has recurrent connectivity (RPT); IIT’s markers (PCI, grid-like lattices) are untested in birds. Pigeons and roosters fail the mark test but discriminate their mirror image from a conspecific. GNWT and RPT requirements “seem to be mostly met”; the authors call the evidence short of a strong conclusion. | Read |
| NOTE-489 | Three types of phenomenal consciousness and their functional roles: unfolding the ALARM theory of consciousness | ALARM: basic arousal (level-1 experience, subcortical) alarms the agent when homeostasis fails, triggering survival behaviour, care for the body and generalised learning without choice; general alertness (level-2, cortical, attention-guided) enables flexible learning and decision; reflexive self-consciousness is general alertness plus metacognitive self-relational content, enabling long-term planning. IIT, HOT and GNWT are said to cover only general alertness. Argued, not tested. | Read |
| NOTE-490 | Block — The Mind as the Software of the Brain | An introductory chapter that says what the computer model of the mind claims and defends it. Intelligence is explained by functional decomposition into primitive processors; meaning-respecting processing by an isomorphism between syntactic and semantic functions that must hold counterfactually; intentionality by a language of thought whose symbols get meaning from functional role. The Turing test, Stich’s syntactic theory and Searle’s Chinese room are criticised. Block grants the model for intelligence and intentionality and doubts it for phenomenal consciousness. | Read |
| NOTE-491 | Block & Fodor — What Psychological States are Not | Block and Fodor argue that behaviourism and type physicalism fail on empirical grounds, and that machine-state functionalism (FSIT) fails because psychological states cannot be one-to-one with the machine-table states of a probabilistic automaton. Dispositions, simultaneity, productivity and structure all break the correspondence, and machine-table identity is too fine-grained. Inverted and absent qualia are a prima facie objection to any functional definition. What survives is that psychological states may be computational states of an automaton, typed by the laws they fall under. | Read |
| NOTE-492 | Block — Functional Role and Truth Conditions | A use (functional role) theory of meaning must say why ‘aluminum conducts electricity’ has different truth conditions on Earth and Twin Earth. Internal inferential roles cannot, and appeals to referential uses beg the question. A public-use theory can only do it with long-arm roles that import the devices of a theory of reference, and the result is equivalent to a two-factor theory whose internal part is narrow meaning. Burge’s arthritis cases further require role in a language, structured by deference, alongside individual roles. | Read |
| NOTE-493 | On a confusion about a function of consciousness | P-consciousness is experience and is not a functional notion. A state is A-conscious if its content is poised for reasoning and for rational control of action and speech. From blindsight or petit mal to a function of P is a fallacy, because A is missing too and A’s machinery has that function. A without P (superblindsight) seems not to occur; P without A is suggested, not shown, by the noticed drill, Sperling arrays, hypnotic analgesia and anaesthesia. | Read |
| NOTE-494 | Cross-region neuron co-firing mediated by ripple oscillations | In 35 patients (43 sessions, 1,373 units) doing a Sternberg task, ripple co-occurrence between bundles is about 5% at every distance from 71 to 223 mm, and unit pairs co-fire a median 34% more during co-ripples than in no-ripple periods (n = 31,489 pairs), 117% more by rate-corrected STTC. Load 3 raises co-ripple co-firing 13% (maintenance) and 19% (retrieval) over load 1. Encoding co-firing recurs at retrieval in 0.29% of cell-pair trials in co-ripples versus 0.14% outside them, 0.36% versus 0.23% for fast versus slow load 3 responses. | Read |
| NOTE-495 | What Is Wrong with the No-Report Paradigm and How to Fix It | No-report paradigms separate consciousness from report but not from post-perceptual cognition that tracks the percept (the bored monkey problem), so prefrontal decoding in no-report rivalry does not show that prefrontal cortex is part of conscious perception. A no-differential-post-perceptual-cognition paradigm is needed. Brascamp et al.’s inconspicuous rivalry, whose switches subjects detected at chance, found no prefrontal difference by fMRI, which Block counts as some support for non-cognitive theories while warning fMRI may be too coarse. | Read |
| NOTE-496 | Two neural correlates of consciousness | Proposes a phenomenal NCC (minimal basis of an experience’s content, for seen motion recurrent MT/V5–V1 activity) and an access NCC (frontoparietal workspace broadcasting). Because access is winner-take-all, a recurrent sensory content can lose and go unaccessed. Signal-detection readings of the exclusion task and of Supèr et al.’s V1 figure–ground modulation, which persists when low salience or many catch trials uncouple it from detection, are offered as evidence. | Read |
| NOTE-497 | Consciousness, accessibility, and the mesh between psychology and neuroscience | States the methodological puzzle: the neural basis of phenomenology can be found only by deciding whether it includes the machinery of access, which reports cannot decide. Answers it by inference to the best explanation: in Sperling, Landman et al. (capacity 6–7 of 8 when cued up to 1.5 s after offset, versus 4) and Sligte et al. (up to 32 items, 4 s) phenomenology overflows a working memory of about four. If workspace activation is not part of phenomenology, losing recurrent sensory coalitions explain the overflow. | Read |
| NOTE-498 | Block — Précis of The Border between Seeing and Thinking | Block restates the thesis of his 2023 book, that perception is constitutively iconic, nonconceptual and nonpropositional and cognition paradigmatically discursive, and develops its least explained notion, the cognitive envelope: the global workspace conceptualises percepts, and the representations held in working memory are task-dependent abstractions that differ from perceptual ones in nine documented ways. Iconicity is defined by tracking and mirroring. Infant and animal iconic thinking is conceded but placed inside a cognitive envelope. | Read |
| NOTE-499 | Distinct ventral stream and prefrontal cortex representational dynamics | In 430 visually responsive subdural electrodes from 10 patients, HFA (70–150 Hz) amplitude and category selectivity fell about 80% from peak by 800–900 ms while images stayed on. Ventral temporal decoding stayed near ceiling (face–watch AUC 99.8%), generalised across time, and ended about 230–450 ms after offset at every duration. Prefrontal and parietal decoding (peak AUC 84–86%) was confined to about 150–600 ms and ignored duration. Exemplar geometry was sustained only in ventral temporal cortex. | Read |
| NOTE-500 | Perceptual consciousness overflows cognitive access | Against sparse readings of Sperling and the Amsterdam change-detection tasks, which posit conscious generic or fragmentary representations plus detailed unconscious iconic memory: there is no independent evidence for the latter, and the one direct test of unconscious working memory (Soto et al., just above 55% at 5 s) is too weak. Fragile VSTM holds 7, 6 and 5 of 8 items at 1, 2.5 and 4 s, is cortical, and is spared by inattention and right dlPFC TMS that impair working memory. | Read |
| NOTE-502 | Interacting spiral wave patterns underlie complex brain dynamics | Phase fields of 0.01–0.1 Hz HCP fMRI on the flattened cortex hold about 16 spirals per hemisphere at rest (radius 10.5 ± 4.8 mm), centred on phase singularities that move at 2.2 mm/s with superdiffusive exponent β ≈ 1.5 and sit on functional boundaries. Opposite spirals annihilate (51% fully, 46% partly), like ones repel (2.6%). Spiral centres and charges decode four language conditions at 48.3% (chance 25%, raw-amplitude decoder 30.2%), and spiral pairs act as gates and walls that route task activity flow. | Read |
| NOTE-503 | Block — Advertisement for a Semantics for Psychology | Block advertises two-factor conceptual role semantics as the only approach that could meet eight desiderata for a semantics serving cognitive science. Narrow meaning is an expression’s short-armed conceptual role, its abstracted causal role in reasoning and deliberation; a separate factor relates it to referents. Conceptual role explains behaviour shared across Twin Earth, learning without definitions, relativity of meaning to representational system, and the inheritance of public meaning from the language of thought. Harman’s long-arm roles reduce to the same two factors, and Gricean and indicator semantics either are not naturalistic or fail. | Read |
| NOTE-504 | Block — Inverted Earth | Block gives the converse of the inverted spectrum: on Inverted Earth a subject or twin with identical brain, identical qualitative content and no internal disturbance comes, through his embedding in the environment and language, to have inverted intentional content and an inverted long-arm functional role. So qualitative content is not a functional state, and public colour talk expresses intentional content. The argument is stronger than the inverted spectrum on testimony and supervenience. A short-arm functional theory of qualia would escape it, and Block can only ask what that theory is. | Read |
| NOTE-505 | Comparing the Major Theories of Consciousness | Compares higher-order thought, global workspace and biological theories by whether they can recognise the explanatory gap. HOT and the workspace replace it with puzzles of their own making; only the biological theory fully acknowledges it, and concept dualism (Nagel, Loar) keeps it from implying dualism. Ambitious HOT also suffers ad hoc stipulation, the mismatch problem and implausibility for infants. Goldberg et al.’s data suggest self-related activity is inhibited in intense perception. The biological theory is the least friendly to machine consciousness. | Read |
| NOTE-506 | Block — Psychologism and Behaviorism | A machine that stores every sensible conversation of a given length and looks up its next reply has the capacity to respond sensibly to anything said to it, so it meets the strongest behaviourist conception of intelligence, but its intelligence is its programmers’. So psychologism is true: whether behaviour is intelligent depends partly on how it is produced. The standard objections to behaviourism only ever refuted necessary conditions; this refutes the sufficient one. Psychologism does not require our kind of processing, only processing that is not of the most elementary sort. | Read |
| NOTE-510 | Mashour, Roelfsema, Changeux & Dehaene — Conscious Processing and the Global Neuronal Workspace Hypothesis | A proponents’ review. Conscious access is ignition: a sudden, self- sustaining, all-or-none activation of long-range hub neurons (prefrontal among them) that broadcasts one representation, against a decaying wave for unseen stimuli. Its best evidence is van Vugt et al.’s monkey recordings, where the signal-detection threshold is the prefrontal ignition threshold, plus a late (>200 ms) seen/unseen divergence across modalities and drug-invariant frontal-parietal disconnection under anaesthesia. It is a theory of access; it calls phenomenality without access untested, and admits the main theories cannot yet be told apart. | Read |
| NOTE-511 | Block — Mental Paint | Block defends phenomenism, the view that phenomenal character outruns representational content, against both forms of representationism. Internalist representationism has only functional resources, and a child raised among ever-changing colours has colour experiences without abiding functional differences. Externalist representationism fails on swampman, on the difference between attention and awareness, on bodily sensations, on Inverted Earth with intact long-term memory, and on an empirical shifted-spectrum argument from normal variation in colour vision. What stays the same across representational change is mental paint. | Read |
| NOTE-549 | Rarely categorical, highly separable representations along the cortical hierarchy | Within single regions of mouse cortex during the IBL task, neurons’ 8-variable selectivity profiles are more clustered than a covariance- matched Gaussian only in VISp, AUDp and SSp-ul, while pooled modules and the whole cortex cluster along anatomical lines. Over the conditions each region distinguishes (M_IC 5–16), ≥ 95% of random balanced dichotomies are decodable above a shuffle null in 15 of 16 regions (GU 0.82). Clustered neurons provably cap the condition geometry’s participation ratio near min(k, M). | Read |
| NOTE-553 | Huebner — Genuinely collective emotions | Collective emotion is defended by extending a homuncular, computational account of individual emotion to distributed systems. Fear needs four integrated capacities (p. 10). The USS Palau crew implements most of them but shows no fear, so it is reclassified as collective danger detection; a 2008 campaign is offered as a likelier case, without data. Unfelt collective emotion is defended by arguing that some individual fear is non-conscious (§5), and functional kinds are defended against homology (§6). Gilbert’s plural-subject account is rejected as reducible to individual commitments. | Read |
| NOTE-570 | Shaping the Institutional Mind | Asks why ascribing mental states to institutions is useful. Both defenders and critics of group minds assume ascription tracks pre-existing causes. On the regulative view of folk psychology, ascription shapes its target. Two cases: the collective knowledge doctrine, after which banks built systems to detect reportable deposits, and corporate narratives and counter-narratives. The paper is agnostic about whether institutions have minds. | Read |
| NOTE-596 | Cumulative cultural evolution in the laboratory | Four chains of ten adults each learned a 27-item artificial language from 14 items and passed their test output on. Transmission error fell and string–meaning structure rose. Unfiltered (Experiment 1), the languages collapsed to 2–5 words that systematically underspecified meanings (colour lost in every chain). With homonyms filtered from training (Experiment 2), they became compositional, with substrings for colour, shape and motion. Participants believed they were copying. | Skimmed |
| NOTE-599 | Language Evolution by Iterated Learning With Bayesian Agents | Iterated learning is a Markov chain on hypotheses. If every learner shares a prior and samples from its posterior, the chain is a Gibbs sampler on P(d|h)P(h), so the probability a learner holds language h converges to the prior P(h), whatever the amount of data passed on. If learners take the MAP hypothesis, it is stochastic EM with no observations: the outcome centres on the prior’s mode, depends on data amount and noise, and in a two-language case is the same for every prior strength below a threshold. An unbounded population with equal fitness has the same equilibrium. | Read |
| NOTE-600 | Contextuality-by-Default theory | A system is a set of stochastically unrelated bunches (one per context) linked by connections (one per content). It is noncontextual if some coupling of the bunches makes every connection a maximal coupling, which separates context-dependent marginals (“direct influences”) from contextuality. The test is linear-programming feasibility (Theorem 4.1). Signed quasi-couplings always exist (Theorem 6.1), and the minimal total variation minus 1 measures the degree of contextuality. | Read |
| NOTE-604 | Wang, Solloway, Shiffrin & Busemeyer, the QQ equality in 72 surveys | Across all 66 Pew surveys of 2001–2011 that varied the order of two questions, chosen without regard to their outcome, question order changes the answers (p = 0.0004 for the order-effect χ² distribution) while the probability of answering the two questions alike does not change (p = 0.4625 for the q values), as the QQ equality predicts. Paired context effects across 72 studies lie along the predicted line of slope −1 (r = −0.82). | Read |
| NOTE-606 | Pragmatic language interpretation as probabilistic inference | States the rational speech act framework (a listener inverting a softmax speaker whose utility is the log-probability a literal listener assigns to the intended world) and its uncertain extension (a joint posterior over the world and the speaker’s topic, knowledge or lexicon), and reviews evidence that these predict reference-game judgements, cost and knowledge effects, scalar and embedded implicature, hyperbole, irony, metaphor and vague adjectives. | Read |
| NOTE-611 | Referential intentions in word learning | A Bayesian learner that infers speakers’ intended referents and the lexicon jointly, allowing words to be used non-referentially and intentions to be empty, learns a much more precise lexicon from annotated child-directed video than association or translation models and reproduces mutual exclusivity, word-driven object individuation and intention-guided mapping as consequences of its likelihood. | Read |
| NOTE-613 | That’s what she (could have) said | Recursive Bayesian speaker–listener reasoning yields specificity implicatures but cannot derive Horn’s principle from utterance costs alone; letting the agents be uncertain about the lexicon breaks the symmetry, so costlier signals go to less likely meanings in one-shot games without equilibrium refinements, and two online games with novel symbols show people drawing both implicatures without prior conventions. | Read |
| NOTE-616 | Iterated learning (Kalish, Griffiths & Lewandowsky) | Puts Bayesian iterated learning to a laboratory test with a bias known beforehand: human function-learning chains converged on a positive linear function in 28 of 32 families, from positive linear, negative linear, U-shaped and random starting functions, as convergence to a prior favouring positive linear functions predicts. The prior was taken from earlier studies, three families settled on the negative linear function, and the authors do not claim the chains reached stationarity. | Read |
| NOTE-618 | Structuralism | From a one-page-in-four preview of chs. I–IV: a structure is defined by wholeness, transformation and self-regulation. Its elements are subordinated to laws of composition; “all known structures… are, without exception, systems of transformation”; and in logic and mathematics regulation is perfect because every operation has an inverse. Structures whose transformations unfold in time, linguistic or social, are regulated but not fully reversible. Structures are real only if governed from within, by a subject defined as “the center of functional activity”. | Skimmed |
| NOTE-627 | Informative communication in word production and word learning | Writes the informative speaker down: utility is −D_KL(intended meaning ‖ literal word meaning), so a speaker referring to one object picks a true word with probability ∝ |extension|^(−α), the size principle from communicative premises. Learners who know the referent and invert this speaker bet on novel-word meanings as predicted (r = .93); speakers’ free descriptions follow informativeness only weakly (r = .19), and colour terms not at all. | Read |
| NOTE-628 v2 | Larger communities create more systematic languages | Varies community size alone in a laboratory language-creation game: groups of eight built more systematic, compositional languages than groups of four, faster and more uniformly across groups, while reaching the same communicative success and convergence. Input variability, greater in larger groups, predicted gains in structure; small groups differed more from each other on every measure. | Read |
| NOTE-631 | Efficient compression in color naming | Casts a colour lexicon as an information-bottleneck encoder from perceptual meanings to words and shows the World Color Survey languages and English lie near the IB bound at β ≈ 1.03, beating hue-rotated variants of themselves in 93% of cases and fitting full naming distributions much better than a deterministic efficiency model; IB optima are soft, and the path of optima through β has phase transitions that roughly follow Berlin and Kay’s sequence. | Read |
| NOTE-634 | Predicting pragmatic reasoning in language games | A listener who inverts, by Bayes’ rule, a speaker choosing words by informativeness to a literal listener, combined with an empirically measured salience prior, predicts mean listener bets in simple three-object reference games at r = .99 with no fitted parameters, and the speaker model predicts speaker bets at r = .98. | Read |
| NOTE-636 | Wang & Busemeyer, the QQ model and QQ equality | Derives the QQ equality from a Lüders-projection model of answering attitude questions: the probability of answering two questions differently is the same in both orders, for any belief state, any projectors and any dimension, provided only the first answer changes the state before the second. It holds in five of six data sets and fails, as the authors predicted, in the sixth, where new information came between the questions. The argument that classical models fail it is informal. | Read |
| NOTE-640 | Contextuality in canonical systems | Contextuality is to be judged on a canonical representation in which every variable is replaced by binary splits and connections are coupled multimaximally. For two content-sharing k-valued variables with all splits kept, the canonical system is noncontextual if and only if one nominally dominates the other: its probabilities fall below the other’s for at most one value (Theorem 4.6). Only the 1- and 2-splits matter (Theorems 4.1, 4.3). For continuous densities, all splits make any difference in distribution contextual. | Read |
| NOTE-645 v2 | On the Quantum-like Contextuality of Ambiguous Phrases | Models meaning selection in two-word ambiguous phrases as a Bell-type measurement scenario. Hand-set supports give one possibilistically contextual (Hardy-type) model; every corpus-estimated model is signalling, so only Contextuality-by-Default applies, under which two noun–verb pairs read in both grammatical orders are contextual (1/30 and 7/30). The paper’s own bootstrap leaves both plausibly noncontextual (probabilities above .56 and .08), on very small counts. | Read |
| NOTE-654 | Analysing Ambiguous Nouns and Verbs with Quantum Contextuality Tools | Proves that in a binary cyclic system CbD’s signalling quantity Δ is twice the sum of each content’s minimal direct influence over Jones’s canonical causal models, which is the total-variation distance between its marginals. In 90 rank-2 noun–verb systems from British corpora, Δ is about 1.35 for every ambiguity class, and homonymous verbs carry about 70% of it against about 50% for polysemous verbs. That difference rests on 14 systems from six verbs and is marginal by a recomputed test (p ≈ 0.06 two-sided). No system is tested for contextuality. | Read |
| NOTE-666 | A mathematical theory of semantic development in deep neural networks | Shows that a two-layer linear network trained from small weights learns the singular modes of its input–output correlations in order of strength, each by a sharp sigmoidal transition, where a shallow network learns all modes together; and that the singular structure of data generated on trees, rings and clusters, plus a minimum-norm property, accounts for coarse-to-fine differentiation, transient errors, narrowing generalisation, typicality, coherence thresholds and conserved representational similarity. Exact only from decoupled, balanced initial weights with white inputs; random initialisation is simulated. | Read |
| NOTE-672 | Hyperbolic geometry of the olfactory space | In four natural odour sources, rank-ordered concentration correlations match points near the surface of a 3D hyperbolic ball and not uniform Euclidean cubes; human odour-descriptor profiles match a full 3D hyperbolic ball. The evidence separates “hyperbolic shell” from “uniform Euclidean cube” only. On a thin shell, hyperbolic distance is nearly a monotone function of angle, so a sphere in flat space, which was not tested, would be a close rival, and the hierarchy that motivates the model is not recovered from the data. | Read |
| NOTE-702 | Floridi — AI as agency without intelligence and the MRA thesis | A taxonomy, not an argument for its thesis. The case for the Multiple Realisability of Agency over the Artificial Realisability of Intelligence is announced (evidence, common sense, Ockham, literature) and then set aside: the paper assumes AI is an agent and asks what kind (p. 2). Floridi & Sanders’ three criteria come back, loosened, so that interactivity alone makes a river a “natural agent” (p. 9). The criteria do not separate AI from other artefacts. Origin, mechanism, where the goals come from, and asserted absences (no intelligence, no mental states) do that. No level of abstraction is ever specified. | Read |
| NOTE-729 | Bickhard — Autonomy, Function, and Representation | A far-from-equilibrium system that contributes to its own maintenance has functions: its contributions serve its persistence, and the thermodynamic asymmetry between persisting and decaying gives function and dysfunction, relative to that system. A candle flame qualifies. A system that switches among self-maintaining processes according to how it has differentiated its environment (recursive self-maintenance) has representation: the indication that an interaction will succeed can be false, and the system can find that out. Etiological, informational and Dretskean accounts are rejected as epiphenomenal or observer-dependent. The argument is conceptual, with no model. | Read |
| NOTE-779 | Kanerva — Sparse Distributed Memory and Related Models | SDM is a two-layer memory: a fixed random address matrix selects a few of many hard locations near the cue, and an outer-product contents matrix stores the word there. Recall of uniform random data has signal pM against noise pM[1 + pT(1 + p²M)], best at p = (2MT)^−1/3, and capacity tends to 1/[Φ⁻¹(φ)]² of the hard locations, about 0.1 at fidelity 0.999. An ordinary RAM is the special case of radius zero. | Read |
| NOTE-785 | Brouillet and Friston — Relative fluency (unfelt vs felt) in active inference | A perspective essay with no model and no new data. Unfelt fluency is the subpersonal recognition of a change in precision, which orients attention; felt fluency is a hierarchically deeper inference about it, reached by mental action. Because its cause is internal, felt fluency is misattributed to the stimulus, which the discrepancy-attribution experiments it reviews are read to show. “Felt” here means attended and reportable; the paper does not separate felt from unfelt in any further sense. | Read |
| NOTE-808 | Aerts — Quantum structure in cognition | Hampton’s membership weights for conjunctions and disjunctions of concepts fail a single Kolmogorovian model, item by item, in two ways: over- or underextension (∆-type) and a negative Kolmogorovian factor (k-type). A subspace lattice with meet and join can fit the k-type items, but provably never the ∆-type ones. Those need an emergent concept state in a second sector of Fock space. Negation appears only as the orthocomplement used to reach conjunction from disjunction, and is never measured. | Read |
| NOTE-864 | Gray & Fu — Soft constraints in interactive behavior | In a simulated VCR-programming task, people with show information always available on screen still made errors that comparing against the display would have prevented. Making that information cost a mouse move and click instead of an eye movement (an estimated 1,000–1,500 ms against 500 ms) cut how often they looked it up, and most checks came right before a setting, almost none while or after setting it. The group made to memorize the information first made the fewest errors. Gray and Fu read this as non-deliberate least-effort choice among microstrategies, with memory time and perceptual-motor time weighed alike (“milliseconds matter”). The access costs are model estimates, not measured or varied parametrically. The error difference between the two on-screen conditions was not significant, the free-access look-up data come from a separate eye-tracking study of eight people, and the Memory-Test condition differed from the others in more than retrieval cost. | Read |
| NOTE-865 | Dasgupta & Gershman — Memory as a computational resource | A short review arguing that memory is used to avoid redundant computation: results are stored and reused (memoization), either whole (“top-down”) or as solutions to subproblems (“bottom-up”), and amortized inference and deep Q-networks are “distributed” memo-tables that generalize to new inputs. It illustrates this in mental arithmetic, mental imagery, probabilistic inference and planning, mostly with response times that flatten with practice and with biases that a cache explains (reuse of earlier inferences, choices based on stale cached values, reuse of policy fragments). It has equations for its examples but no cost model: the space–time trade-off is named, and when caching pays is listed as an outstanding question. Memory is internal throughout; retrieval cost is mentioned once, and offloading or external stores not at all. | Read |
| NOTE-867 | Hutchins — How a cockpit remembers its speeds | In an MD-80 approach, the correspondence between airspeeds and wing configurations for the day’s landing weight is computed once, 25 to 30 minutes before landing, by selecting a printed speed card with the displayed gross weight, and is stored in the positions of movable bugs on both airspeed indicators; when the plane slows 10 to 15 minutes later, the needle’s approach to a bug cues each flap call by a judgment of spatial proximity. Hutchins argues that this memory belongs to the cockpit as a functional system and that speed bugs do not improve any pilot’s memory but replace a recall-and-scale-reading task with a perceptual one. The account is a stylized description of mandated procedure, informed by more than 100 observer flights and simulator recordings, with no measured performance, no comparison of crews with and without bugs, and no failure cases; the claim that the arrangement changes what the system can do is established by analysis of the tasks, not by data. | Read |
| NOTE-868 | Gray, Sims, Fu & Schoelles — The soft constraints hypothesis | In a Blocks World copying task where the model pattern was hidden and opening it cost a lockout of 0 to 3,200 ms, people opened it less often, kept it open longer, and placed more blocks after the first look as the lockout rose, and the trend was graded down to differences of tens of milliseconds. An Ideal Performer Model that chooses how many blocks to encode per look (ENCODE-1 to ENCODE-8) by Q-learning, with the only reward being negative time and memory given by ACT-R’s base-level activation, retrieval-threshold and latency equations, fits four measures from Experiment 3 with three memory parameters and one softmax noise parameter fitted on one of them. The authors read this as support for choosing between head and world by time alone, against a minimum memory hypothesis, which they concede no one has stated precisely enough to model. The model’s only memory costs are encoding time, retrieval time and forgetting; it has no term for anything carried across trials, for a retrieved item being wrong, or for the world changing. | Read |
| NOTE-870 | Geisler — Contributions of ideal observer theory to vision research | A short review, not a methods paper. It defines the ideal observer as the device that “performs a given task at the optimal level possible, given the available information and any specified constraints”, writes the Bayesian ideal as an argmax of expected utility over the posterior given either the stimulus S or a constrained neural representation Z = g(S;θ) (eq. 1), and surveys applications from photon-noise detection to eye-movement search. It gives no formula for efficiency and no procedure for locating information loss beyond saying that constraints are added to the ideal one at a time (optics, receptor sampling, receptor noise, ganglion-cell summation) and that models are built by “degrading the ideal observer with hypothesized neural noise or with hypothesized heuristic computations”. Its constraints restrict the information available to an optimal decoder; it has no notion of an optimum over a restricted class of computations, though its attention examples compare the ideal with specific suboptimal rules (a one-location “serial” mechanism, a MAP searcher, a random searcher). | Read |
| NOTE-871 | Kirsh — The intelligent use of space | A classification, from video of cooks, assemblers and baggers, informal observation and the Tetris studies, of how people arrange things around them to cut the time and memory a task needs: arrangements that simplify choice (hide or highlight affordances, eliminate decisions, offload heuristic properties), arrangements that simplify perception (clustering, regions, marking, the vernier effect), and spatial dynamics that save internal computation (Tetris rotation, self-cueing). It frames the savings in search-complexity terms and links informational structuring to amortized computation in a footnote, but says the savings cannot in general be meaningfully quantified and offers no cost model, no measurements beyond the cited Tetris timings, and no case of an arrangement failing under a deadline. Its agents are experts arranging their own workspace in the medium and short term; long-term, pre-built environments are set aside. | Read |
| NOTE-872 | Kirsh — Adapting the environment instead of oneself | A conceptual paper. It models a task as a directed graph of states and actions with a cost metric on transitions, and classifies how an agent can improve its payoff by actions external to the task: pruning hard initial states, routine maintenance, scouting ahead, tools (which change the graph), and actions that make a space more “cognitively congenial” (which leave the graph intact and reduce mental operations). It names amortization once, as a parenthetical gloss on chimps leaving nut-cracking stones in place, and gives an investment rule for maintenance (future benefit against present cost times an “interest factor” for risk) in words only. There is no cost model with reuse frequency, retrieval cost or staleness; mental cost lies outside the graph, and the paper says there is “no general theory” of congeniality. The empirical figures are unpublished pilot results. | Read |
| NOTE-873 | Lieder & Griffiths — Resource-rational analysis | Resource rationality defines the right cognitive mechanism as the one, among those a brain can execute, that maximizes expected utility of its result minus the expected opportunity cost of the time and resources it uses (Eq. 3), averaged over the environments consistent with what is known (Eq. 4); resource-rational analysis is a five-step iterate-and-refine method for deriving process models from that objective. The objective is per episode. It has no separate terms for storing a result, retrieving it later, or its going stale: the value of what a computation leaves behind is folded into the utility of the episode’s result, and the cost of finding the heuristic in the first place is assigned to evolution and learning and left out. Reuse appears only in the review, qualitatively: habits as reused action sequences, Dasgupta et al. 2018’s memorized hypotheses, Howes et al. 2016’s memorize-or-look-up trade, and “cognitive prostheses”. The framework is offered as a methodological device, not a claim that people are optimal. | Read |
| NOTE-875 | Dasgupta, Schulz, Goodman & Gershman — Remembrance of inferences past | In three online experiments (N = 84, 80 and 100, before exclusions), a sub- or superadditivity bias induced in a first probability judgment carried over into a second, always-packed judgment about the same scene distribution, and errors on the two queries were correlated trial by trial (r ≈ 0.34–0.67). The carry-over appeared only when the second query shared part of its hypothesis space with the first, which rules out simple numerical anchoring; it was larger when cognitive load fell on the first query, which the authors’ simulations read as reuse of a summary statistic rather than of the stored samples; and with a new cue it appeared (for subadditivity only, and weakly) when the cue’s posterior was similar to the first. The paper defines amortization as reuse of past computation bought at a cost in bias, and itself places it in the time-for-space trade. It measures no response times, so the speed benefit is assumed, not shown; efficiency is shown only by better secondary-task accuracy on reusable trials; and it gives no account of when reuse pays. | Read |
| NOTE-876 | Harris et al. — Failures to replicate a key result of selective accessibility | Five preregistered experiments (N = 128 each, about four times the originals’ N) failed to replicate the lexical-decision result of Mussweiler and Strack (2000a, Studies 1 and 2). That result was that after a comparison with a high or low anchor, words consistent with the anchor are recognised faster. Every experiment produced a large anchoring effect on the later estimate (d ≈ 0.85–1.39), and none showed the accessibility interaction (all η²G ≤ .004). An exploratory Bayesian meta-analysis gave a Bayes factor of 21.1 for the null. Validation experiments found that the same tasks also missed direct picture priming. So the authors conclude that the method cannot detect the effect, not that selective accessibility is false. They still call the theory “a good explanation”, but say the earlier lexical-decision evidence “is not a strong part of the evidence base”. | Read |
| NOTE-877 | Strack & Mussweiler — Explaining the enigmatic anchoring effect | In three small between- and within-subjects studies (N = 32, 32 and 67), the anchor’s effect on a later estimate depended on what the comparison task brought to mind, not on the number alone. With the anchor number held fixed, the effect was large when the estimate was on the dimension compared (Study 1). It reversed to contrast when the estimate was about a very different object (Study 2). After plausible anchors, the comparison took longer and the estimate came faster (Study 3). Accessibility is inferred from these moderations and from latencies. No study measures what knowledge was retrieved. The paper also keeps an adjustment-like route for implausible anchors, which anchored at least as strongly. | Read |
| NOTE-878 | Kirsh & Maglio — On distinguishing epistemic from pragmatic action | Kirsh and Maglio define an epistemic action as a physical action whose primary function is to reduce the space, time or unreliability of the agent’s own mental computation, and argue from Tetris keystroke data that players rotate and translate pieces for that reason: zoids are rotated more than plan-then-act would need, rotation starts within a few hundred milliseconds of a zoid appearing, and a translate-to-wall check is used more the higher the drop. The computational saving is argued, not measured: physical rotation of 90° takes as little as 100 ms against an estimated 800–1200 ms mentally (a three-player pilot), and fewer attentional probes and less working memory are needed. The trade is physical action in place of internal computation at the moment of use, not precomputation or caching; the paper explicitly sets it against caching and chunking accounts of skill, and mentions amortized complexity only in passing for everyday cases such as sorting bolts before assembly. It says the choice fits a Stigler-style cost–benefit model in outline but that such models fail in detail, because an epistemic action’s value depends on the particular internal computation it serves. | Read |
| NOTE-882 | Howes, Duggan, Kalidindi, Tseng & Lewis — Short-term remembering as boundedly optimal strategy choice | Participants copied names one at a time from a mock email to a mock calendar and, after a forced phase that measured their own copying rate at every list length from 3 to 9, chose how many names to take per trial; the list length that had given each person the highest rate in their own data predicted their choice better than one item fewer or one more (55% vs 8% and 17% of choices in Experiment 1, N = 40; 46% vs 12% and 14% in Experiment 2, N = 20). An exponential points schedule raised chosen list length (6.99 vs 5.07), but doubling the cost of an error (whole trial forfeited) did not (5.00 vs 5.05). The “boundedly optimal” prediction is the argmax of each participant’s measured rate, not a model with capacity, error or look-up parameters, and the task has no look-up: names are shown once, in sequence, and cannot be consulted again. 13 of 60 participants persistently chose a non-optimal length, 10 of them one item more. | Read |
| NOTE-883 | Ballard, Hayhoe, Pook & Rao — Deictic codes for the embodiment of cognition | At about a third of a second, the time scale of a saccade, the article models fixation as a pointer that binds a variable in a behavioural program to a place in the world, so that what is bound is only the task-relevant feature, fetched just before use. In a block-copying task subjects looked back at the model about 1.3 times per block on average, the commonest cycle being two looks per block (one apparently for colour, one for position), and chose this slowest, lowest-memory strategy (about 3 s a block, against about 1.5 s for copying from memory) though told to be fast; separating model and copy by 70° cut looks to 1.0 per block, and one-colour patterns that could be chunked cut them to 0.7. The authors read this as working memory having a “carrying cost” that exceeds the cost of re-acquiring from the world, but that cost is inferred from the choices and never measured or modelled, and the evidence is one task with small samples (seven subjects for the main data, three for the colour-change experiment) and no statistics reported in the article. What is left in the world is unprocessed input, not a computed result, and the article argues against precomputation where the needed products cannot be foreseen. | Read |
| NOTE-886 | Gershman & Goodman — Amortized inference in probabilistic reasoning | Gershman and Goodman define amortized inference as the reuse of past inferences, cached as sub-computations, to answer related later queries. They report two small online experiments in which order alone made reuse available. When a sub-query came before a complex query on a five-node Bayesian network, a person’s answer to the complex query could be predicted from their own answer to the sub-query (r = 0.28; r = −0.11 when the order was reversed). Answers to the complex query were also faster relative to a structurally matched baseline. Reuse in this paper is internal memory, and it is shown only through correlation and speed, within one task, with small effects. The paper does not measure costs. Storage cost and the choice of what to store are listed as open trade-offs. Bias, staleness and interference are not studied: reuse that may be “inaccurate” without modification, and interference and forgetting of stored inferences, appear only as questions. | Read |
| NOTE-887 | Risko & Gilbert — Cognitive offloading | A short review that defines cognitive offloading as physical action that changes a task’s information-processing requirements to reduce cognitive demand, splits it into offloading onto the body and into the world, and proposes a metacognitive model: whether people offload is a strategy choice driven by their evaluation of internal versus extended strategies, which tracks objective internal demand but can be wrong, and offloading in turn changes both memory and those evaluations. The evidence is a handful of choice/no-choice studies, summarised qualitatively; the review reports no effect sizes and no formal account of costs. It names costs (memory for offloaded content, GPS-impaired spatial memory, photo-taking impairment, automation complacency and bias, skill decay, inflated self-knowledge, a self-reinforcing drift toward external reliance). It does not frame offloading as computation reused over time, and says nothing about how often an offloaded result is used. | Read |
| NOTE-897 | Treisman & Gelade — A feature-integration theory of attention | In visual search for a coloured letter, a target that differs from all distractors by one feature (a colour, blue, or a shape, S, among brown Ts and green Xs) was found in a time almost independent of display size (positive slopes 3.8 ms/item for colour, 2.5 for shape, display sizes 1–30), while a target defined by a conjunction (green T among brown Ts and green Xs) gave linear slopes of 28.7 ms/item when present and 67.1 when absent, the roughly 1:2 ratio of serial self-terminating search; 13 sessions of practice did not make the conjunction pop out. Making the conjunction’s colours and shapes harder to tell apart nearly tripled the slope but kept it linear, and a single-feature target hard to discriminate (an ellipse of intermediate size) was slow but not item-by-item serial. Even in feature search, target-absent trials were slow and grew with display size (25.1 ms/item). The displays were at most 30–36 well-spaced items on a card, with unlimited viewing; the paper argues that conjunction structure, not similarity, makes search serial, and does not model similarity. | Read |
| NOTE-947 | Giarrocco & Averbeck — sensory and choice information flow between A1 and PFC | In 37 sessions of simultaneous recordings from A1 and dorsolateral PFC of two macaques reporting by bar release whether two tones came from the same side, cue-side information flowed A1 to PFC only (onset 120–147 ms), while choice information (release or hold) flowed first PFC to A1 (onset 315–365 ms after target) and then A1 to PFC (352–419 ms), both peaking near 500 ms, close to the bar release. The flow measure is a lagged regression on one-dimensional projections, so the result is correlational, within a trial, and does not separate the decision from the motor response. | Read |
| NOTE-948 | Kübler & Birbaumer — BCI and extinction of goal-directed thinking in CLIS | Of 35 patients trained with EEG BCIs in Tübingen, degree of paralysis explained BCI performance (η² = .40) only because of the seven in the completely locked-in state, none of whom reached the 70% needed for yes/no communication; without them the relation vanished (η² = .12, n.s.). All seven were already in CLIS when training began. The authors propose, by analogy with the curarized rat, that lost perception of the contingency between a response and its consequences extinguishes goal-directed thinking, while “passive sensory information processing is intact”, and predict that a contingency learned before CLIS and carried into it should prevent extinction. | Read |
| NOTE-949 | Slovic — the construction of preference | A review, from an award address, of the preference-reversal literature. Normatively equivalent elicitation procedures (choice against pricing, choice against matching) give systematically different orderings, and about 90% of reversals violate procedure invariance rather than transitivity. Two mechanisms explain them: an attribute weighs more when it is compatible with the response scale, and the prominent attribute weighs more in choice than in matching. Slovic concludes that for important, complex and unfamiliar options preferences are constructed on the spot in elicitation. The evidence is about the response under each procedure; nothing in it tests whether a constructed preference persists or changes later judgements. | Read |
| NOTE-950 | Ma & Kragel — map-like emotion knowledge in hippocampus and vmPFC | In fMRI of 29 people passively watching 14 emotional films, decoders trained on hippocampal patterns predicted an independent sample’s consensus ratings of 13 emotion categories better than positions in a two-dimensional affective space, and vmPFC showed the reverse; patterns could also be decoded into the activity of a Tolman–Eichenbaum Machine that had learned a grid built from rating correlations by random walk, hippocampus favouring its small-scale conjunctive layer and vmPFC its large-scale structural layer. Effects are small (prediction–outcome correlations of about 0.04–0.10), and no learning in the participants was measured. | Read |
| NOTE-953 | Sung, Rigotti & Lee — factorized goal and uncertainty codes in lateral PFC | A reanalysis of 20 people’s fMRI from a 2014 two-stage Markov task with cued goals and block-wise transition noise. Goals and transition uncertainty are linearly decodable, weakly, from vlPFC, dlPFC and OFC, in a geometry whose nonlinear dichotomies decode worse than goal and linear ones (“factorized”), and a goal decoder trained under one noise level transfers to the other. Goal separability and transfer in lateral PFC correlate with choice flexibility and optimality across subjects; uncertainty separability does not. Unfitted model-free agents show a flexibility–stability trade-off that model-based agents and humans do not. | Read |
| NOTE-954 | Powers, Clark & McFarland — General Feedback Theory, Part I | Builds a hierarchy of negative-feedback control systems in which each order perceives only the feedback signals of the order below and acts only by setting their references, which are replayed recordings of their own past perceptions. An “imagination connection” sends those references back into the higher system’s perception, so that it can remember, plan “mentally”, hallucinate and dream, under a “mixing control” the authors cannot specify (“We are open to suggestions”, p. 20). An inherited N-system, driven by error in intrinsic physiological variables, reorganizes the network at random, forming new input functions as well as new outputs; an organization persists when the behaviour it produces lessens intrinsic error. | Read |
| NOTE-956 | Parr & Friston — attention or salience? | A short review that separates two things called attention: gain control, which weights currently available sensory data by their estimated precision, and salience, the expected information gain of an action (typically a saccade) that would acquire new data. It then ties them together: a saccade is driven by a prediction that a location will be sensed with high precision, so covert attention to a place biases an overt saccade there, which the authors read as support for the premotor theory. | Read |
| NOTE-957 | Feldman & Lynch — self-generated validity | A theoretical paper, with no new data. When the belief, attitude or intention asked about is not already in memory, the question creates it, and the created answer is later retrieved as an input to other answers and to behaviour; when it is in memory, question order and format still decide which earlier answer is used. An earlier response is used for a later one if it is accessible and more diagnostic than other accessible inputs. Expertise, direct experience and trait involvement reduce the effects, because they supply judgements already formed and rehearsed. | Read |
| NOTE-959 | Powers, Clark & McFarland — General Feedback Theory, Part II | Names six orders of perception, from spinal loops through sensations, configurations, sequences and relationships to system concepts, with demonstrations whose reaction times lengthen with order (about 0.3 s at third order, 0.4–0.45 s at fourth). The N-system “merely alters the properties of a system, thus changing the transformations applied to existing information” (p. 37): reorganization is a change of perception, felt as insight, and a new perceptual organization can be vetoed by imagining its use. Consciousness is defined as a feedback function under N-system action. Conflict freezes a lower system at a virtual reference, and a system switched “entirely to the imagination-connection” is one of the ego’s defences. | Read |
| NOTE-963 | Powers — Feedback: Beyond Behaviorism | A control system “will produce whatever output is required in order to achieve a constant sensed result” (p. 77), so behaviour is the control of input, not output, and stimulus-response laws follow from the environment once the controlled quantity is found. Above the lowest level a system’s output is a reference level for the level below, so the hierarchy closes through lower systems down to muscle effort sensed kinesthetically; its top is a set of inherited intrinsic reference levels whose error drives reorganization. The paper says nothing about imagination: every loop it describes closes through the environment. | Read |
| NOTE-991 | Apps & Tsakiris — The free-energy self | A free-energy account of self-recognition and body ownership. One’s own body, face and voice are represented probabilistically as “me” by a high-level multimodal prior learned from cross-modal correlations, and the self is “the most parsimonious and accurate explanation for sensory inputs”. The rubber hand, enfacement and rubber-voice illusions are one effect: synchronous stimulation leaves unimodal surprise that multimodal areas explain away by shifting probability from one’s own body part to the other’s. Context acts as a prior. No brain area is self-specific. The authors grant there is little direct evidence and that the account is hard to falsify with standard fMRI. | Read |
| NOTE-995 | Starmans & Bloom — Windows to the soul | Asked which of two pictured flies or snowflakes was “closer to” a cartoon character, preschoolers and adults chose those near the eyes, even when the eyes were on an alien’s chest and even over positions the character could see. The authors conclude that people intuitively locate the self at or near the eyes, and, since 4-year-olds do it too, not from learned ideas about the brain. The judgement is about where another, drawn person is; the step to the self, and on to a “phenomenological sense” of where we reside, is assumed and not tested. | Read |
| NOTE-996 | Nichols & Bruno — Intuitions about personal identity | Survey studies of Arizona undergraduates on Williams’s personal-identity cases. A transplant case makes memory look necessary for being “still Jim”; a stripped-down Pain case has about 70% say “you will feel the pain” after distinctive mental states are wiped; neither depends on first- or third-person telling, so Williams’s explanation by person fails here. Abstract questions and a forced choice favour psychology (over 80%; 64%, p = 0.053). The authors weigh intuitions by robustness and conclude for a psychological criterion conditionally. The results describe a concept under framing; they do not show what persistence is, and the Pain item may test bare sentience, not identity. | Read |
| NOTE-1008 | Alsmith & Longo — Where exactly am I? | Ten adults pointed a rotating pointer “directly at” themselves, by hand blindfolded or by eye. Answers fell mostly on the upper face or the upper torso, and the pointer’s starting position, above or below, decided which. The authors conclude that the self is judged neither spread over the body nor at one point, and that the spatial concept of the self is “inherently ambiguous”. A pilot-sized study with a fifth of its trials lost and a misreported reanalysis; its start effect is also what a first-acceptable-stop strategy predicts. | Read |
| NOTE-1016 | Tushnet — Worth a Thousand Words: The Images of Copyright | Copyright law takes the word as its model and treats images as either transparent or opaque, with incoherent results in copyrightability, infringement, authorship and fair use. Realism is a style, not a fact. Describing an image changes how it is seen and remembered, so a test that asks factfinders to dissect works in words and then judge their gestalt cannot work. The proposed cure is a reproduction right against near-exact copies plus a narrowed derivative-works right. | Read |
| NOTE-1056 | Gładziejewski — Real patterns, the predictive mind, and the manifest image | A philosophy paper with no data. It traces Dennett’s compression criterion for real patterns through minimum description length and Bayesian inference to variational free energy, so that learning a generative model is extracting real patterns from sensory data. It concludes that the manifest image is a structural description of sensory data, veridical even in a matched vat. The self-model compresses self-generated regularities by positing one common cause, so selves are real as patterns and unreal as substances, and which verdict to give is a dispute over criteria, not facts. It says nothing about consciousness or illusion. | Read |
| NOTE-1066 | Carruthers & Gennaro — Higher-Order Theories of Consciousness (SEP) | A survey of the theories that make a mental state conscious by its being the object of a higher-order representation: inner sense, actualist and dispositionalist higher-order thought, and self-representational and hybrid views. It sets out the objections (targetless states, the rock, animals, sufficiency) and the prefrontal evidence, and concludes no verdict. Since 2020 it is maintained by Gennaro, a higher-order theorist. Carruthers and Lycan, two of the theories’ main sources, are reported to have given them up. | Read |
| NOTE-1071 | Graziano — Illusionism big and small | An essay with no new data. Claims of consciousness come from an automatic internal model, and since every brain model is schematic, what people report having is a caricature of whatever the model tracks. Six hypotheses for what that is are offered. The author favours attention, or more deeply “the deep selective processing of information”, depicted by an attention schema as an intangible essence. Theories that say how a brain produces a feeling and not how the feeling gets into the model that drives report (“Arrow B”) are called magic. Its “small” illusionism still denies the phenomenal essence. What it keeps real is the process modelled. | Read |
| NOTE-1079 | McFadyen et al. — Doorways do not always cause forgetting | Four conceptual replications from an independent lab. In passive, head-mounted virtual reality, doorways did not lower hits for carried or set-down objects; under a counting load they raised false alarms to recombined probes. Watching or walking a curtained hallway did not change memory for a 16-item grid. The nulls are informative but not direct failures to replicate: movement was passive, and Experiments 3 and 4 used a different memory task. | Read |
| NOTE-1081 | Pettijohn & Radvansky — Doorways: event structure or updating disruption? | Two virtual-navigation experiments from the originating lab. Travel time was not longer through doorways, and a 3-second pause before the probe left the effect as it was, so neither a longer retention interval nor a passing cost of updating explains it. In both, the effect was confined to the carried object. The authors take this as support for interference between the two event models that hold that object. | Read |
| NOTE-1082 | Radvansky et al. — Walking through doorways causes forgetting: further explorations | Three experiments. The doorway effect appeared with small monitors and in real rooms carrying a box of six objects, and returning to the room where the objects were picked up did not help: memory followed the number of rooms involved, worst after two new rooms. The authors attribute it to retrieval interference between event models that both contain the probed object, not to the loss of an encoding context. | Read |
| NOTE-1083 | Radvansky & Copeland — Walking through doorways causes forgetting | Two desktop virtual-reality experiments. With distance held equal, walking through a doorway rather than across a large room raised errors and slowed responses to a probe for the object being carried; memory for the object just set down was not affected. The authors read it as situation-model updating at a spatial shift. The effect rests on few no-shift trials (5 per condition), and the shift and no-shift rooms differed in size and wall pattern. | Read |
| NOTE-1084 | Zacks et al. — Event perception: a mind-brain perspective | States event segmentation theory: perception continuously predicts what comes next; a transient rise in prediction error opens a gate that resets working-memory event models and lets sensory input in, and that moment is perceived as an event boundary. Supported by unitization, memory, narrative and imaging studies, mostly correlational and largely from the authors’ groups. Its one acknowledged misfit: the anterior cingulate does not respond at boundaries during passive viewing. It is about watching events, not one’s own intentions, and does not mention doorways. | Read |
| NOTE-1100 | EMA: A process model of appraisal dynamics | A computational process model of appraisal after Smith and Lazarus. Appraisal is a single, fast, parallel mapping from the agent’s causal interpretation (its beliefs, desires, intentions, plans, probabilities and utilities) to per-proposition appraisal frames. Emotional dynamics come from the time course of the perception, inference and action that change the interpretation. Frames compete for the coping response, and the most recently accessed one with the highest mood-adjusted intensity wins. Coping alters the world or the agent’s mental state, which drives reappraisal. Illustrated by an actor’s 2.6-second reaction to a pigeon. | Read |
| NOTE-1103 | An Appraisal-Driven Componential Approach to the Emotional Brain | A programme for taking the component process model into the brain: five functionally defined networks, appraisal driving changes and synchronization in the other four, synchronization proposed to be neuronal phase or cross-frequency coupling, and feeling emerging when the degree of synchronization passes a threshold. It states that some action tendencies take priority by control precedence. It gives no measure of synchronization, signed or unsigned, per episode or otherwise. | Read |
| NOTE-1105 | Kahneman, Wakker and Sarin — Back to Bentham? Explorations of experienced utility | Decision utility, an outcome’s weight in choice, and experienced utility, its hedonic quality, are distinct and can be measured separately. Experienced utility is instant (real-time report) or remembered (retrospective). Remembered utility follows a Peak-End rule with duration neglect, and choices follow remembered utility, so people choose dominated painful episodes. Total utility is axiomatized as a time integral of ordinal instant utility with a neutral zero. For the record, the dissociation is not a counterexample to potential read within the moment. It shows that potential read from deferred choices measures memory, and that a system’s model of its own past states is systematically biased. | Read |
| NOTE-1106 | The Emotion Process: Event Appraisal and Component Differentiation | The evidence for the emotion process, organised by appraisal criterion and by the component appraisal drives. Appraisal criteria predict action tendencies, physiology and expression better than emotion categories do. Intrinsic valence and goal congruence have partly different signatures and combine multiplicatively. Control amplifies goal-congruence effects. Coherence among components is proposed as the mark of an episode at one threshold and of awareness at a second. No negotiation among perspectives: appraisal is a profile of criteria that drives patterned responses. | Read |
| NOTE-1108 | Levels of valence | A framework, not a study. Valence is located at two levels. Micro-valences are the outcomes of separate valenced appraisal checks, and they can conflict. Macro-valence is a one-dimensional integration of them and of other inputs, and it is what choice needs. The paper’s work is in the paths between the levels, and above all in the integration rules, from weighted sums to lexicographic and elimination strategies that need no trade-offs. Intrinsic pleasantness and goal conduciveness are separate checks, and the paper grounds the separation in Berridge. | Read |
| NOTE-1113 | Emotions are emergent processes: they require a dynamic computational architecture | The component process model restated for computational modellers. Emotion is a bounded episode of synchronization among subsystems (appraisal, autonomic, motor, motivation, feeling), driven by appraisal checks in a fixed order and aimed at control precedence. Feeling monitors the process by representing the other components and their coherence. The degree of synchronization is hypothesised to generate consciousness. Emotions are attractor basins of coupled oscillators, with hysteresis and catastrophe jumps. No operational measure of synchronization is given. | Read |
| NOTE-1181 v2 | Lindsey et al. — On the biology of a large language model | Case studies of Claude 3.5 Haiku through attribution graphs of a 30-million-feature cross-layer transcoder, each checked by inhibiting or injecting features in the model. On single prompts: “Dallas” reaches “Austin” through “Texas” features, beside a direct shortcut; a rhyme is chosen on the newline before a line and the line is written towards it; one addition lookup-table feature serves arithmetic, citations and tables; “can’t answer” features are on by default and inhibited by known-entity features; a chain of thought can work backwards from a hinted answer; a jailbreak’s target word is not represented before it is written; a goal finetuned into a model organism is represented across every Assistant context. The authors call these existence proofs and report a satisfying graph for about a quarter of prompts tried. | Read |
| NOTE-1232 | Simon — ill structured problems | Simon argues that there is no real boundary between well and ill structured problems. Well-structuredness is judged by six conditions read as practicable, not in-principle, computability, so it is a relation between a problem and a bounded solver. On that reading even theorem proving and chess fail some conditions, chess is “well structured in the small” and ill structured “in the large”, and “there are no WSPs, only ISPs that have been formalized for problem solvers” (p. 186). House design, a battleship’s design by an organization and a robot in a real environment are handled by alternating a solver working on a locally well structured problem with a recognition and retrieval system that keeps modifying the problem space. Simon’s conclusion is deflationary: ill structured problems may differ from well structured ones only in the size of the knowledge base, and need no qualitatively new mechanisms. | Read |
| NOTE-1295 | Reconceptualizing Individuals, Categories, and the Social Group: Self-Categorization Theory 2.0 | An insider’s critique and reconstruction of self-categorization theory. Categorization into equivalence classes yields interchangeable members, not a group entity. Meta-contrast, taken as the mechanism of entity formation, is intractable (2ⁿ subsets: 9.3 hours for 25 people at 1,000 operations a second). Its prototype semantics fails compositionality. The proposed SCT2.0: self-categorization is the categorization of intentional agents; groups are holistic entities triggered by cues to collective intentionality; and relations are propositional, in a language of thought. A theoretical paper; its own evidence is under review. | Read |
| NOTE-1313 | Mental Hoop Diaries: Emotional Memories of a College Basketball Game in Rival Fans | Twenty-three male fans, 12 of Duke and 11 of UNC, watched an archived 2001 Duke–UNC game three times in the week before scanning. Under fMRI they saw 64 clips, each ending as a shot was released, and recalled whether it went in (74% correct). The same shot was good for one side and bad for the other, so allegiance alone set its valence; fans’ ratings matched their side 88.9% of the time. Good shots for one’s own team were recalled more accurately and confidently and rated more intense. Intensity, not valence, modulated amygdala, hippocampus, striatum and cortical midline regions. Positive shots added dorsal fronto-parietal and sensorimotor activity. The two fan groups did not differ in accuracy or rated intensity. | Read |
| NOTE-1349 | Gentner — Structure-mapping | Gentner proposes that an analogy maps the relations of a base domain onto a target and drops the objects’ attributes, and that which relations go across is decided by systematicity: a relation belonging to a system tied together by higher-order relations is more likely to be imported than an isolated one. The rules look only at the form of the representation (arity and order of predicates), which lets literal similarity, analogy, abstraction and mere appearance be told apart by what kinds of predicate overlap. The framework is argued from examples; the empirical support is one study summarized in a paragraph. | Read |
| NOTE-1398 | Johansson, Hall, Sikström and Olsson — Failure to detect mismatches between intention and outcome | In a card-trick experiment, 120 people chose the more attractive of two female faces and were sometimes handed the face they had rejected. Only 13% of the 354 manipulated trials were detected at the time, at most 26% counting every form of later detection, and the detection rate did not depend on how similar the faces were. The reasons given for the rejected face did not differ from reasons given for real choices in length, emotionality, specificity or certainty, and some named features only the rejected face had. The authors call this choice blindness. | Read |
| NOTE-1412 | Nisbett and Wilson — Telling more than we can know | Reviews dissonance, attribution, subliminal-perception, problem-solving and bystander studies, and reports eight small studies of its own, to argue that people have little or no introspective access to the processes by which stimuli affect their judgments. Their reports come from a priori causal theories, which observers given only a description share: in Nisbett and Bellows, subject and observer reports correlated .77 to .99, and both were accurate only for the judgment the culture has rules for. | Read |
| NOTE-1423 v2 | Kuppens, Van Mechelen, Smits & De Boeck — The appraisal basis of anger | In two recall studies, anger was associated with other accountability and unfairness, and other accountability was specific to anger, but no appraisal (goal obstacle, other accountability, unfairness, control) and no antagonistic action tendency was singly necessary or sufficient for it. Every anger report in study 2 had at least one component. The authors conclude that appraisals and emotions are contingently associated, not linked by necessity, and make no causal claim. All measures are self-reports of recalled situations. | Read |
| NOTE-1437 | Moreno & Mossio — Biological Autonomy, the Introduction | The Introduction alone. Moreno and Mossio state their programme: the organism, not natural selection, is biology’s central idea; its core is a closure of constraints, which they say fundamentally differs from the process loops and self-organisation of dissipative structures, and which grounds teleology, normativity and function. With regulation and agency, closure makes an autonomous system, an organism; ecosystems meet only closure. The arguments are announced for chapters 1–7, which this reading does not cover. | Skimmed |
| NOTE-1454 | Flynn, Kanerva & Bhadkamkar — SDM principles and operation | Kanerva and co-authors state SDM’s critical distance in words and do not derive it: a cue within it converges on the stored pattern under iterated reads, a cue beyond it wanders and may settle on another pattern, the distance shrinks as the memory fills, and in a sparsely filled memory it is below the sum of the write and read radii. For single patterns convergence and divergence are “easy to distinguish in practice”; for stored sequences they are not. The rest of the report is a hardware design. | Skimmed |
| NOTE-1492 | Bechtel & Abrahamsen — dynamic mechanistic explanation | A dynamic mechanistic explanation joins a mechanistic account of parts, operations and organization to a computational model whose variables are properties of those parts and operations. In circadian research the parts came first, from experiment, and six kinds of modelling result follow (C1–C6). In cognitive science the models posit their mechanisms and are checked only by fit to behaviour. That makes them how-possibly accounts, and leaves competing architectures underdetermined. Two of the paper’s cases bear on the record: clamping a variable in a grounded model to test componency (C3), and the open question whether peripheral clocks are self-sustained or damped oscillators driven by the SCN (C6). | Read |
| NOTE-1495 | Metzinger — The myth of cognitive agency | Mind-wandering is a recurring, unnoticed loss of M-autonomy, the veto-centred ability to control one’s own mental processes. Most conscious thought is subpersonal mental behaviour, and a thought becomes the person’s only when integrated into a transparent “epistemic agent model”. The paper states that we lack M-autonomy for about two thirds of conscious life; its own minutes give 52%. | Read |
| NOTE-1496 | Metzinger — The problem of mental action | Mental action, which has no sensory sheet in its loop, is cast as predictive control of effective connectivity aimed at the epistemic value of states in the self-model. Mind-wandering supplies a competition among candidate thoughts that control selects from. Whether the conscious epistemic agent model constitutes the control, represents and enables it, or is a by-product, the data “do not help to decide”, which leaves open what M-autonomy stated as a principle. | Read |
| NOTE-1497 | Metzinger — M-Autonomy | M-autonomy is the capacity for second-order mental action, centred on veto. What mind-wandering loses is not the ability but its representation in the self-model, the “epistemic agent model”, and an ability can be deliberately exercised only when it is so represented. The return of control is represented as self-caused even when an unconscious event caused it. The paper repeats the 2013 estimate of two thirds, which its own minutes do not give. | Read |