Topic 8: Reasoning(Chap 1,7,8)

Epistemological Foundations of Deductive and Inductive Reasoning

• Reasoning is the cognitive and inferential process of evaluating given information (premises) to reach conclusions that are not explicitly stated, serving as a causally necessary mediator between an organism's desires, beliefs, goals, and subsequent actions.

• In classical philosophical epistemology, arguments are classified into two broad categories based on the nature of the epistemic warrant that the premises provide for accepting the conclusion: deduction and induction.

• Deductive Reasoning:

  • Characterized by the property of validity: an argument is valid if and only if the truth of its premises provides an absolute, infallible guarantee for the truth of its conclusion.

  • Validity is strictly a function of the logical structure or syntactic form of the argument, completely independent of the empirical truth or falsity of the propositions contained within it.

  • For instance, the argument with form: $M\text{ has }F$, $\text{nothing with }F\text{ has }G$ , therefore $M\text{ does not have }G$ guarantees validity whether instantiated with familiar entities (e.g., Tweety the robin, Jenkins the bachelor) or empirical scientific propositions (e.g., Oxygen as an element).

  • Soundness represents the convergence of formal validity and empirical truth: an argument is sound if and only if it is deductively valid and all of its premises are factually true in reality.

  • An argument may possess valid logical form while being entirely unsound due to false premises and a false conclusion (e.g., "Napoleon was German; all Germans are Asian; therefore Napoleon was Asian").

• Inductive Reasoning:

  • Encompasses all arguments where premises provide only limited, probable, or plausible grounds for accepting a conclusion, rather than a logically necessary guarantee.

  • Extends across multiple distinct varieties:

  • Enumerative induction: Generalizing from observed specific instances to an unobserved universal claim (e.g., from observed white swans in Central Park to the claim that all swans everywhere are white).

  • Specific instance projection: Inferring a property of an unobserved individual instance based on observed instances (e.g., observing some cats with broken legs and inferring that a particular cat, Socrates, has a broken leg).

  • Abductive inference (inference to the best explanation): Reasoning backwards from an observed effect to a plausible cause, which formally mimics the deductive fallacy of affirming the consequent (e.g., observing that an entity has four legs and concluding it is a cat).

• The Mischaracterization of Sherlock Holmes:

  • Literature's most famous detective frequently characterizes his cognitive feats as "deduction" (e.g., deducing that Dr. Watson had been out in vile weather and had a careless servant girl based on six parallel cuts scoring the inside leather of his left shoe; or deducing Watson had returned to medical practice from the smell of iodoform, a black nitrate of silver stain on his right forefinger, and a top-hat bulge accommodating a stethoscope).

  • Despite Holmes's assertions, these inferences are strictly inductive and abductive conceptual inferences; numerous alternative causal explanations remain equally plausible (e.g., Watson walking along a dry, sunny, gravel road whose stones scratched the leather, or Watson scraping his own boots).

• David Hume's Problem of Induction:

  • Formulated in the context of causal inference, Hume demonstrated that universal empirical conclusions (e.g., "all woolly mammoths ate fir cones" based on discovering $27\,\text{pounds}$ of undigested fir cones, larch, pine, sedge, wild thyme, and moss in a Siberian permafrost specimen, or dietary habits of the armored dinosaur Borealopelta markmitchelli based on fern, cycad, and conifer gut contents) can neither be reports of direct experience nor logical deductions from experience.

  • They cannot be direct reports of experience because humans observe only a finite subset of instances.

  • They cannot be deductive consequences because drawing the inference requires the Principle of the Uniformity of Nature, which posits that unexamined instances must resemble examined instances and that the course of nature operates uniformly.

  • The Principle of the Uniformity of Nature cannot be proven deductively (nature's course may change without logical contradiction) nor empirically observed; it can only be justified through inductive generalization, generating a fatal vicious circle.

  • Hume's positive resolution was psychological rather than epistemological: constant conjunction of antecedent and consequent events produces an associative "habit of mind" compelling humans to anticipate the conclusion upon encountering the premise.

• Nelson Goodman's New Riddle of Induction:

  • Goodman revealed a critical defect in Hume's psychological account: Hume explained how regularities produce mental habits but failed to recognize that the human mind projects only certain select regularities while rejecting others supported by identical empirical evidence.

  • Illustrated by the contrast between physical traits: observing Borealopelta markmitchelli remains containing specific plant fossils leads to the projected universal habit of mind that all Borealopelta consumed those plants, yet observing that the exact same specimen exhibits broken and missing scales does not lead to the projection that all Borealopelta possessed broken scales, even though the logical form ($X\text{ has }\alpha \land X\text{ has }\beta \implies \forall x(\alpha(x) \rightarrow \beta(x))$) and observational base are identical.

  • Goodman formalized this paradox using the artificial predicate "grue":

  • An object is grue if and only if it is observed before time $t$ (e.g., January 1, 2135) and is green, or is not examined before time $t$ and is blue.

  • All emeralds examined prior to time $t$ confirm the hypothesis that "all emeralds are green" and equally confirm the hypothesis that "all emeralds are grue."

  • At time $t$, the two hypotheses yield radically contradictory predictions, revealing that identical observational data support contradictory conclusions depending on which predicate is projected.

  • Distinguishing projectable from nonprojectable predicates by calling the former "law-like" and the latter "accidental" is circular, as law-like predicates are defined purely as those that project.

Cognitive Paradigms and Theoretical Architectures of Deduction

• Psychologists evaluate human reasoning by presenting participants with formal arguments, tracking error patterns against normative standards, and reverse-engineering the underlying cognitive machinery.

• The Three Foundational Theoretical Questions:

  • The Competence Question: Do humans possess an innate, intuitive competence knowledge of basic logical relations (such as transitivity and modus ponens), scaling up to complex domains subject to capacity constraints?

  • The Error/Bias Question: How can systematic, nonrandom deviations from normative logic be accounted for without abandoning the premise of human rationality?

  • The Structure versus Content Question: Why does semantic content robustly modulate reasoning accuracy if logical validity is strictly determined by syntactic form?

• Classical Computational Theories:

  • Mental Logic Theory (Henle, Braine, Rips):

  • Assumes the cognitive architecture internalizes a natural deductive calculus operating on syntactic, language-like propositional strings.

  • Reasoners possess an innate competence regarding the inferential roles of closed-form logical connectives and quantifiers (such as "all", "some", "none", "if-then", "and", "or").

  • Substandard performance is attributed strictly to performance constraints, including working memory overload, attentional lapses, and pragmatic misinterpretations of experimental instructions.

  • Mental Model Theory (Johnson-Laird, Byrne):

  • Rejects syntactic formal rules in favor of semantic, visuospatial mental simulations.

  • Reasoners comprehend the meaning of logical operators to construct spatial mental representations depicting possibilities consistent with the premises.

  • An inference is judged valid only if it holds across all constructed alternative models without being falsified by a counterexample.

  • Reasoning difficulty and errors scale directly with the number of models required to represent the premises and the cognitive load involved in exhaustive searching.

• The Probabilistic "New Paradigm" (Oaksford, Chater, Elqayam, Evans, Over):

  • Argues that normative classical logic fails to reflect everyday human reasoning with conditionals ($\text{if }p\text{ then }q$), which inherently accommodate uncertainty and exceptions.

  • Employs the Ramsey Test (Ramsey 1929/1990; Stalnaker): reasoners evaluate a conditional by hypothetically adding the antecedent to their existing background beliefs, adjusting their web of beliefs to maintain coherence, and estimating the subjective conditional probability of the consequent.

  • Conditionals are governed by subjective conditional probabilities ($P(\text{consequent} \mid \text{antecedent})$):

  • For instance, in the conditional "if the brake pedal is pressed, the car stops", initial confidence may be set at $P(\text{stopping} \mid \text{brakes}) = 0.9$.

  • Augmenting the antecedent with the belief that the brakes need repair drops confidence to $P(\text{stopping} \mid \text{brakes}, \text{unserviced}) = 0.6$.

  • Augmenting with the knowledge that the brakes were just professionally serviced elevates confidence to $P(\text{stopping} \mid \text{brakes}, \text{serviced}) = 0.96$.

  • Theoretical consequences: Under the strict Bayesian new paradigm, pure deductive reasoning is dissolved, reframing all reasoning as inductive inference governed by contextual probability updating, which ultimately re-encounters Goodman's riddle regarding what contextual information is relevant.

• Representational Formats and Probabilistic Deduction (Gigerenzer et al.):

  • Human reasoning errors often stem from architectural mismatched representational formats rather than fundamental irrationality.

  • In a diagnostic mammography screening problem ($1\%$ breast cancer prevalence, $90\%$ sensitivity, $9\%$ false-positive rate), gynecologists were asked to estimate the probability of cancer given a positive mammogram:

  • When presented in standard conditional probability formats, only $21\%$ of trained physicians selected the correct value (approximately $1\text{ in }10$, or $C$), with the majority erroneously estimating $81\%$ or $90\%$.

  • When Gigerenzer reformulated the exact problem in natural frequencies ($10$ out of $1{,}000$ women have cancer; $9$ of those $10$ test positive; $89$ of the $990$ cancer-free women test positive), accuracy surged to $87\%$.

  • Identical representational dynamics explain public health panics: reporting that third-generation oral contraceptive pills doubled the risk of thromboembolism produced widespread panic and pill abandonment, whereas stating that the risk increased from $1\text{ in }7{,}000$ to $2\text{ in }7{,}000$ communicates the true natural frequency without inflating perceived risk.

Systematic Errors, Biases, and Content Effects in Logic

• Formal Syllogistic Forms:

  • Categorical syllogisms consist of two premises and a conclusion relating three terms: Subject ($S$), Predicate ($P$), and Middle term ($M$), which is eliminated during deduction.

  • Categorical propositions exist in four standard moods:

  • Universal Affirmative ($A$): All $S$ are $P$

  • Particular Affirmative ($I$): Some $S$ are $P$

  • Universal Negative ($E$): No $S$ are $P$

  • Particular Negative ($O$): Some $S$ are not $P$

  • Syllogisms are classified into four figures according to the placement of the middle term:

  • Figure 1: $M\text{-}P$, $S\text{-}M$

  • Figure 2: $P\text{-}M$, $S\text{-}M$

  • Figure 3: $M\text{-}P$, $M\text{-}S$

  • Figure 4: $P\text{-}M$, $M\text{-}S$

  • Permuting four moods across three propositions ($4 \times 4 \times 4 = 64$) across four figures yields $256$ distinct syllogistic forms, of which exactly $24$ are logically valid.

• Non-Content Deductive Errors:

  • Atmosphere / Mood Effect (Woodworth & Sells 1935): The presence of a negative premise induces participants to accept a negative conclusion; the presence of a particular quantifier ("some") induces participants to accept a particular conclusion, irrespective of formal validity.

  • Unwarranted Information Supplementation: Reasoners import conversational implicatures into formal logic (e.g., reading "some men are honest" and illicitly inferring that "some men are not honest").

  • Illicit Premise Conversion: Assuming symmetry in non-symmetrical relations (e.g., converting "all $A$ are $B$" into "all $B$ are $A$"); this conversion is readily blocked when real-world semantic knowledge is present (e.g., reasoners do not convert "all apples are fruit" into "all fruit are apples").

  • Conditional Reasoning Asymmetries:

  • Modus Ponens ($P \rightarrow Q$, $P \vdash Q$): Affirming the antecedent is universally endorsed ($97\%$ to $100\%$ acceptance).

  • Modus Tollens ($P \rightarrow Q$, $\neg Q \vdash \neg P$): Denying the consequent is formally valid but notoriously difficult, eliciting endorsement from only approximately $65\%$ of reasoners.

• Denying the Antecedent ($P \rightarrow Q$, $\neg P \vdash \neg Q$): Formal fallacy, endorsed by approximately $40\%$ of participants.

  • Affirming the Consequent ($P \rightarrow Q$, $Q \vdash P$): Formal fallacy, endorsed by approximately $40\%$ of participants.

  • Confirmation Bias: Strong tendency to seek confirming evidence rather than searching for disconfirming counterexamples.

• Content Effects in Deduction:

  • Minna Wilkins (1928) first demonstrated that reasoners perform with substantially higher logical accuracy when an argument's conclusion aligns with pre-existing real-world beliefs.

  • Syllogistic Belief-Bias Paradigm (Evans, Barston, & Pollard 1983):

  • Valid argument with believable conclusion: endorsed as valid by $92\%$ of participants (reaction times are shortest).

  • Valid argument with unbelievable conclusion (e.g., "No addictive things are inexpensive; some cigarettes are inexpensive; therefore some cigarettes are not addictive"): endorsed as valid by only $46\%$ of participants.

  • Abstract or neutral arguments (e.g., "No $A$ are $B$; some $C$ are $B$; therefore some $C$ are not $A$"): endorsement rates fall intermediate between these extremes (approximately $70\%$).

  • Wason Four-Card Selection Task (Wason 1960):

  • Reasoners are presented with four cards showing $E$, $K$, $4$, and $7$, and must test the conditional rule: "If a card has a vowel on one side, it has an even number on the other side" by turning over the minimum number of cards.

  • Normative solution requires turning card $E$ (Modus Ponens) and card $7$ (Modus Tollens, looking for a vowel on the reverse to disconfirm the rule).

  • On the abstract task, accuracy is typically only $20\%$ to $25\%$, with most participants selecting $E$ alone or $E$ and $4$ (affirming the consequent).

  • Griggs and Cox (1982) demonstrated dramatic content facilitation using a deontic drinking-age rule: "If a person is drinking beer, then they must be over 18 years of age" with cards displaying "Beer", "Sprite", "22", and "16": accuracy jumps to over $90\%$ (participants immediately select "Beer" and "16").

Dual-Mechanism Frameworks and Critical Re-Evaluation

• Theoretical Formulations of Dual-Process Architecture:

  • Developed primarily to account for content effects and belief-bias without declaring human cognition inherently irrational.

  • Stanovich and West (2000) formalized the dichotomy into System 1 (heuristic, automatic, effortless, associative, preconscious, implicit) and System 2 (analytic, formal, rule-based, conscious, effortful, slow).

  • Evans and Over (1996) mapped System 1 to an evolutionary "old brain" shared with nonhuman animals and acquired via association, while System 2 was identified as a distinctively human, neocortical system supporting hypothetical thinking.

  • Keith Stanovich (2004) explicitly analogized System 1 to a reflex arc providing hardwired, invariant causal links between triggers and responses.

  • Steven Sloman (1996) distinguished associative reasoning processes from rule-based computational processes.

  • Daniel Kahneman (2002/2003) integrated Tversky and Kahneman's heuristics-and-biases program into a dual-system model where System 1 generates intuitive, non-voluntary "impressions" occupying an intermediate position between perception and deliberate judgment, while System 2 generates explicit, reflective judgments.

  • System Selection Models:

  • Default Interventionist Model (Evans & Over): System 1 automatically generates an initial default response; System 2 must detect the error, intervene, inhibit System 1, and compute the normative logical response.

  • Parallel Competitive Model (Sloman): Both systems compute representations simultaneously from stimulus onset; whichever system finishes first or asserts dominance determines the behavioral response.

• Canonical Heuristic Tasks and Decision Biases:

  • Linda Conjunction Fallacy (Tversky & Kahneman 1983): Participants describe a 31-year-old, single, outspoken, philosophy major concerned with social justice and antinuclear protests, rating "Linda is a bank teller and active in the feminist movement" as more probable than "Linda is a bank teller", directly violating the conjunction rule of probability theory ($P(A \land B) \le P(A)$).

  • Jack the Engineer Base-Rate Fallacy (Tversky & Kahneman 1982): Given a sample of $10$ engineers and $90$ lawyers, participants read a description matching an engineer stereotype and rate Jack as more likely an engineer, ignoring the $90\%$ base-rate probability of him being a lawyer.

• Conceptual and Empirical Critique of Dual-Mechanism Theories:

  • The "Duck Problem" of Behavioral Individuation:

  • Dual-process theorists individuate cognitive systems based primarily on processing speed (fast versus slow). Categorizing phenomena by speed alone is unscientific, analogous to grouping cars, comets, and airplanes as "fast" while grouping bicycles, bears, and nuclear submarines as "slow", obscuring fundamental underlying mechanisms.

  • Published reaction time differences between supposed System 1 and System 2 responses are scarce, often showing non-significant trends rather than clear statistical dissociations.

  • The Reflex Arc Category Error:

  • True reflex arcs (such as the eye-blink reflex when fingers snap near the face) are mandatory and causally invariant: they persist despite forewarning, absolute trust, and high financial stakes.

  • Belief-biased responses on the Linda or cigarette tasks are completely non-reflexive: individuals provide coherent reasons for their choices, recognize their error when explained, and suppress the bias on subsequent trials.

  • Belief-biased choices operate over complex, high-level semantic propositions and world knowledge, requiring the machinery of the cognitive mind rather than autonomic, subcortical reflexes.

  • Working Memory Walking-Back:

  • Evans and Stanovich (2013) retreated from defining System 1 as an old-brain associative reflex, asserting instead that the sole defining attribute is that System 2 recruits working memory while System 1 does not.

  • Under computational information processing models, all cognitive inferences over propositional structures require working memory; a cognitive inference operating with zero working memory is computationally impossible, reducing the claim to an unprincipled quantitative difference in working memory load.

  • The Obscurity of Kahneman's "Impressions":

  • Inferences require propositional representations possessing syntactic and semantic compositionality; nonconceptual or nonpropositional "impressions" lack any coherent computational mechanism for carrying out inferential steps.

  • Recent Evidence for Fast Logic:

  • Recent empirical findings demonstrate that logical inferences can occur faster than belief-biased responses, and individuals often exhibit implicit sensitivity to logical violations even when endorsing nonnormative answers (Handley, Newstead, & Trippas 2011; Trippas, Thompson, & Handley 2017; De Neys 2017), upending default interventionist assumptions.

  • Reconceptualizing Content Fallacies:

  • Tasks like the Linda problem or base-rate problems set up an artificial conflict between formal logical criteria and everyday pragmatic, conceptual coherence.

  • The Linda task cues Gricean conversational maxims: participants reasonably assume the experimenter intends to provide relevant, informative descriptions, leading them to choose the description that captures Linda's persona.

  • When neutral descriptions are used, the fallacies disappear entirely.

Cognitive Neuroscience and Neural Fractionation of Deduction

• Neuroimaging studies establish that deductive reasoning does not recruit a single, unitary, dedicated "reasoning module" in the brain, but instead dynamically activates fractionated, distributed networks dependent on argument form, semantic content, emotional valence, determinacy, and conflict.

• Neuroanatomical Systems for Familiar versus Unfamiliar Content:

  • Deductive reasoning about familiar, meaningful, conceptually coherent material engages a left-lateralized frontotemporal language and conceptual network involving the left inferior frontal gyrus (BA 44, BA 45) and middle/superior temporal cortex (BA 21, BA 22).

  • Deductive reasoning about abstract, unfamiliar, or conceptually incoherent material (such as letters or artificial symbols) recruits a bilateral visuospatial occipital-parietal and dorsal frontal network, primarily involving the superior and inferior parietal lobules (BA 7, BA 40) and dorsal prefrontal areas (BA 6).

  • Core areas shared by both familiar and unfamiliar reasoning include the left inferior prefrontal cortex (BA 44), left fusiform gyrus (BA 18), right fusiform gyrus (BA 37), bilateral basal ganglia nuclei (caudate nucleus, putamen, nucleus accumbens), and right cerebellum.

• Neuropsychological lesion data confirm this anatomical dissociation: patients with focal unilateral lesions to the left prefrontal cortex fail to show performance facilitation from familiar content in reasoning tasks (such as the Wason card task and transitive reasoning), performing as poorly on familiar content as on arbitrary content, whereas patients with right prefrontal lesions and intact parietal lobes perform normally.

  • Neuroanatomical Systems for Conflict Detection and Belief-Bias:

• Congruent reasoning conditions occur when the logical validity of an argument matches the empirical believability of its conclusion (valid/believable or invalid/unbelievable), facilitating accurate performance.

• Incongruent reasoning conditions occur when validity conflicts with empirical believability (valid/unbelievable or invalid/believable), inducing belief-bias interference.

• Succumbing to belief-bias (generating incorrect, non-logical responses) activates the ventromedial prefrontal cortex (VMPFC; BA 11, BA 32), reflecting affect- and belief-driven responding.

• Successful conflict detection, inhibition of the prepotent belief-bias, and execution of formal logic selectively engages the right lateral/dorsolateral prefrontal cortex (R DLPFC; BA 45, BA 46).

• Right-hemisphere damaged patients fail specifically on verbal reasoning problems where the direction of the question conflicts with the premise structure (e.g., "Mike is taller than George; who is shorter?"), demonstrating an inability to resolve linguistic-relational conflict.

  • Neuroanatomical Dissociation of Determinate and Indeterminate Inferences:

• Determinate reasoning trials contain complete relational information permitting a definitive, deductively valid or invalid conclusion.

• Indeterminate reasoning trials contain incomplete information where the conclusion cannot be definitively resolved from the premises, requiring the maintenance of ambiguity.

• Lesion studies in neurological patients with focal frontal lobe damage demonstrate a clear double dissociation: patients with left prefrontal lesions are selectively impaired on determinate trials, whereas patients with right prefrontal lesions are selectively impaired on indeterminate trials.

• Functional neuroimaging confirms that reasoning through indeterminate relationships specifically recruits the right ventrolateral prefrontal cortex (R VLPFC), which serves to actively maintain uncertainty and prevent premature closure.

  • Integrated Cognitive-Neuroanatomical Model of Deduction:

• Synthesizes Michael Gazzaniga's "left hemisphere interpreter" with right prefrontal conflict detection and uncertainty maintenance systems.

• The left prefrontal pattern completer acts as a general engine that abhors ambiguity and compulsively completes patterns across logical, causal, and social domains, frequently jumping to premature conclusions.

• When evaluating an argument, processing pathways diverge based on content familiarity, conflict, and determinacy:

  • Abstract, unfamiliar arguments (arguments without meaningful content) bypass the left frontotemporal heuristic network and are processed directly by the bilateral formal/universal parietal system. The R DLPFC monitors for formal structural inconsistencies, while the R VLPFC maintains indeterminate representations, inhibiting the left hemisphere interpreter from making unwarranted completions.

  • Familiar arguments containing meaningful content are initially routed to the left frontotemporal heuristic/conceptual system. If no conflict arises between belief and logic, the left hemisphere rapidly yields a response.

  • If an incongruity or conflict between conclusion believability and logical validity is detected, the R DLPFC intervenes, rerouting the argument representations to the formal parietal network while working to suppress the prepotent belief-biased response generated by the VMPFC.

Cognitive and Computational Models of Inductive Inference

• Psychological induction research focuses on uncovering the empirical principles that guide human reasoners in selecting which observed regularities to generalize.

• Similarity-Guided Inductive Projection:

  • Increasing the perceived similarity between the premise category and the conclusion category increases confidence in inductive generalization:

  • Reasoners judge the inference "Dogs have hearts; therefore wolves have hearts" as significantly stronger than "Dogs have hearts; therefore bees have hearts", driven by the morphological and biological similarity between dogs and wolves.

  • Category Typicality Effect:

  • Properties generalize more readily from central, typical members of a superordinate category than from atypical, peripheral members.

  • Reasoners rate "Robins have sesamoid bones; therefore sparrows have sesamoid bones" as much stronger than "Ostriches have sesamoid bones; therefore sparrows have sesamoid bones", because robins represent a prototypical exemplar of the bird taxon.

  • Inductive Asymmetry:

  • Generalizations from typical to atypical category members are perceived as stronger than generalizations from atypical to typical members.

  • "Robins have $38\,\text{chromosomes}$; therefore ostriches have $38\,\text{chromosomes}$" is judged stronger than "Ostriches have $38\,\text{chromosomes}$; therefore robins have $38\,\text{chromosomes}$".

  • Inclusion Fallacy (Osherson et al. 1990):

  • Reasoners systematically rate "Robins have sesamoid bones; therefore all birds have sesamoid bones" as stronger and more plausible than "Robins have sesamoid bones; therefore ostriches have sesamoid bones", committing a formal fallacy given that the category "all birds" strictly subsumes ostriches.

• Diversity-Guided Inductive Projection:

  • When an inductive argument features multiple premise categories, increasing the diversity (dissimilarity) among the premises enhances the perceived plausibility of generalizing to the superordinate category.

  • Reasoners judge the argument "Robins require magnesium to live; ostriches require magnesium to live; therefore all birds require magnesium to live" as significantly stronger than "Robins require magnesium to live; sparrows require magnesium to live; therefore all birds require magnesium to live".

  • The wide ecological and morphological distance between robins and ostriches covers a broader region of the category space, providing stronger grounds for projecting the predicate to the entire taxon.

• Causality-Guided Inductive Projection:

  • When contextual knowledge evokes a plausible causal mechanism, causal coherence overrules taxonomic and perceptual similarity:

  • For anatomical traits, taxonomic similarity governs: reasoners prefer projecting "Hawks have a liver with two chambers; therefore chickens have a liver with two chambers" over projecting to tigers.

  • For ecological behaviors, shared causal mechanisms govern: reasoners prefer projecting "Hawks prefer to feed at night; therefore tigers prefer to feed at night" over projecting to chickens, because hawks and tigers share the causal attributes of being carnivorous predators.

  • Causal chains account for directional asymmetries in property transmission:

  • "Gazelles contain retinum; therefore lions contain retinum" is judged substantially stronger than "Lions contain retinum; therefore gazelles contain retinum", because reasoners recruit trophic food-chain models wherein ingestion serves as a viable causal transmission vector.

• Formal Computational Models of Induction:

  • Similarity-Coverage Model (Osherson et al. 1990):

  • Calculates argument strength based on two components: the maximal pairwise similarity between premise categories and conclusion categories, and the degree of coverage the premise categories provide over the lowest superordinate taxonomic category unifying both premises and conclusion (e.g., using properties of robins and blue jays to project onto geese via the superordinate bird category).

  • Feature Overlap Model (Steven Sloman 1993):

  • Accounts for inductive strength through a normalized measure of shared feature vectors between premises and conclusions, implemented computationally in connectionist feed-forward networks (Rogers & McClelland 2004).

• Foundational Limitations of Inductive Models:

  • Neither model provides an account of similarity itself; similarity ratings must be computed externally and manually supplied to the system.

  • Models cannot handle inductive relations that defy pairwise similarity or taxonomic classification (such as complex relational or syntactic inductions).

  • The concept of similarity begs the question of relevance: any two entities in the universe share an infinite number of objective properties (e.g., a coffee cup and a computer monitor share the properties of resting on a desk, being indoors, existing on Earth, being over $91{,}000{,}000\,\text{miles}$ from the Sun, being less than $100{,}000{,}000\,\text{miles}$ from the Sun, etc.).

  • Similarity can only be computed once an independent criterion of relevance has already isolated the pertinent features; appealing to similarity to explain relevance is entirely circular.

  • Similarly, causality cannot provide a fully objective foundation, as Hume established that causal necessity is a psychological projection rather than an empirically observable primitive, grounded in an assumed principle of uniformity.

• The Massive Modularity Hypothesis as a Radical Reconceptualization (Cosmides & Tooby):

  • If the human mind is conceptualized as an unconstrained, general-purpose information processing system with universal access to all stored knowledge, Goodman's riddle of projectability (the frame problem in artificial intelligence) remains computationally intractable.

  • Evolutionary psychologists propose the Massive Modularity Hypothesis: the mind comprises hundreds of thousands of Darwinian, domain-specific modules triggered automatically by specific environmental inputs.

  • By hardwiring a tight causal coupling between specific environmental cues and behavioral outputs, the massive modularity view sidesteps the riddle of induction by eliminating open-ended inductive searches; however, it renders flexible, unconstrained human rationality an illusion.

Cognitive Neuroscience of Inductive Reasoning and Analogy

• Neuropsychological Split-Brain Paradigms (Gazzaniga):

  • Investigated hemisphere-specific contributions to implicit conceptual inference in patients with severed corpus callosum.

  • Classic Dual-Picture Presentation Paradigm:

  • A winter snow scene is projected exclusively to the right hemisphere (controlling the left hand), while a chicken claw is projected exclusively to the left hemisphere (controlling the right hand).

  • The patient's left hand selects a shovel, and the right hand selects a chicken.

  • When verbally asked why the left hand pointed to the shovel, the left-hemisphere language system (unaware of the winter scene) immediately invents a post-hoc rationalization: stating that the shovel is necessary to clean out the chicken coop.

  • Sequential Image Induction (Gazzaniga & Smylie 1984):

  • A picture of a pan followed by a picture of water is presented independently to each hemisphere.

  • When presented to the left hemisphere, the patient effortlessly completes the causal inference ("boiling water").

  • When presented to the right hemisphere, the patient cannot generate the conceptual inference, demonstrating that the left hemisphere acts as a dedicated causal interpreter connecting observed events.

• Functional Neuroimaging of Inductive Arguments:

  • Comparative fMRI and PET studies presenting inductive arguments (evaluating conclusion plausibility) versus deductive arguments (evaluating formal validity) reveal extensive activation of left frontal and parietal regions during induction, overlapping with deductive networks used for familiar content.

  • Inductive reasoning is distinguished from deductive processing by unique, selective recruitment of the medial aspect of the left superior frontal gyrus (BA 8, BA 9) and the frontopolar cortex (BA 10).

• Neuroanatomy of Analogical Mapping and Reasoning:

  • Analogical reasoning relies on mapping relational structures from a source domain to a target domain, serving as a core inductive mechanism.

  • Geometric Shape Analogies (Wharton et al. 2000):

  • Positron emission tomography during analogical matching tasks demonstrates enhanced metabolic activity in the medial frontal cortex (BA 8), left prefrontal cortex (BA 6, BA 10, BA 44, BA 45, BA 46, BA 47), anterior insula, and left inferior parietal lobule (BA 40).

  • Verbal Analogous Word Pairs (Green et al. 2006):

  • Evaluating abstract proportional analogies (e.g., "Planet : Sun :: Electron : Nucleus") engages a left-lateralized parieto-frontal network, peaking in the left superior frontal gyrus / frontopolar cortex (BA 9, BA 10), which mediates the integration of abstract relational representations.

  • Conceptual Semantic Analogies (Luo et al. 2003):

  • Probing linguistic analogies (e.g., "Soldier is to army as drummer is to band") activates bilateral prefrontal regions (BA 11, BA 45, BA 47) and the left middle temporal gyrus/hippocampal complex (BA 22).

Real-World Conceptual Inferences, Argumentation, and Political Manipulation

• Categorization of Everyday Inductive Fallacies:

  • Category 1: Formal Logical Fallacies Embedded in Conceptual Inferences:

  • Committing affirming the consequent or denying the antecedent during everyday causal hypotheses (e.g., believing that if it rained overnight, the grass is wet; observing wet grass; and definitively inferring rain, ignoring sprinklers or morning dew).

  • Circular reasoning, wherein an inferential claim relies upon a presupposition that can only be substantiated by the conclusion itself.

  • Category 2: Methodological and Data Interpretation Fallacies:

  • Unrepresentative or biased sampling (e.g., predicting a national election outcome by polling ten like-minded citizens).

  • Confusing spurious correlation with causal necessity (e.g., an empirical $99\%$ correlation observed between per-capita margarine consumption and the divorce rate in the state of Maine).

  • Category 3: Heuristic-Driven Cognitive Fallacies:

  • Gambler's Fallacy: The erroneous belief that after a prolonged run of black stops on a fair roulette wheel, a red stop is "due", ignoring the statistical independence of each spin ($\approx 50\text{--}50$ chance).

  • Availability Heuristic: Skewing subjective risk estimations based on the cognitive ease with which recent exemplars come to mind (e.g., markedly inflating the statistical risk of middle-aged myocardial infarctions after hearing of an acquaintance's heart attack).

  • Anchoring and Adjustment: Systematically biasing quantitative numerical estimates toward an arbitrary initial probe (e.g., significantly lower estimates of African nations in the United Nations when probed with "is it more than 5?" versus "is it more than 25?").

• Scientific Conceptual Argumentation: Galileo versus Aristotle:

  • Historical analysis indicates Galileo did not drop cannonballs from the Leaning Tower of Pisa, but rolled metal balls of varying mass along finely measured inclined planes.

  • In physical reality, due to ambient aerodynamic drag, a heavier cannonball strikes the ground slightly ahead of a lighter cannonball.

  • In Galileo's Two New Sciences (1638), the interlocutors Salviati and Simplicio debate falling bodies:

  • Aristotle posited that a 100-pound iron ball dropped from $100\,\text{cubits}$ hits the ground when a 1-pound ball has descended only a single cubit.

  • In experimental trials, Galileo observed that the heavier ball lands only two finger-breadths ahead of the smaller ball.

  • Salviati argues that Simplicio should not hide Aristotle's massive error of 99 cubits behind Galileo's small discrepancy of two finger-breadths.

  • Methodological assessment: Galileo's argument embodies the "Lesser of Two Evils Fallacy"; demonstrating that an empirical prediction is far less erroneous than a competitor's does not establish that the hypothesis is correct, as both could be incomplete or false.

• Political Discourse and Rhetorical Exploitation of Instinctive Systems:

  • In conceptual reasoning within complex socio-political contexts, reasoners evaluate evidence based on how coherently actions align with beliefs, motives, and contextual facts.

  • The 2019 First Impeachment Inquiry of Donald Trump:

  • Centers on allegations that President Trump withheld congressionally mandated military aid to coerce Ukrainian President Volodymyr Zelenskyy into announcing investigations into political rival Joe Biden and the Democratic National Committee's computer server (CrowdStrike).

  • A rational-coherence evaluation considers multiple evidentiary ambiguities:

    • Verifying the empirical accuracy and completeness of phone transcripts and whistleblower testimonies.

    • Evaluating whether an implicit, contextually understood threat constitutes an abuse of power (analogous to a criminal shakedown where an individual displaying a firearm smiles and requests a "donation").

    • Distinguishing correlation from causation: determining whether withholding aid was causally linked to personal political requests or motivated by general anti-corruption concerns.

    • Assessing motivated legal arguments: reviewing the Senate defense argument that if an elected official genuinely believes their re-election serves the public interest, seeking assistance cannot constitute an impeachable quid pro quo.

  • Mobilization of Instinctual Systems over Coherence Relations:

  • The political defense strategically bypassed conceptual coherence relations by deploying rhetorical fallacies calculated to trigger subcortical, noncognitive action tendencies and tribal evolutionary instincts.

  • Common Rhetorical Strategies Deployed:

    • Ad Hominem Attacks: Diverting focus from evidence to the character of investigators (e.g., dismissing the whistleblower as an incompetent "legal-eagle" writing a term-paper report laden with partisan bias, or branding opponents as "Do Nothing Democrat Savages").

    • Deflection: Redirecting attention toward alternative grievances requiring independent adjudication (e.g., claiming the true crime was White House staffers leaking conversations, or demanding investigations into political opponents).

    • Focus on Consequences: Appealing to catastrophic social harms rather than legal validity (e.g., arguing that impeachment overturns the democratic will of the electorate and disenfranchises millions of voters).

    • False Dilemma: Constructing polarizing bifurcations (e.g., asserting that anyone who fails to support the executive automatically supports the opposing political party).

    • Argument from Fear: Evoking survival-level threats (e.g., claiming that opposing leaders will open borders to criminals and confiscate firearms).

    • Total Evidentiary Denial / "Alternative Facts": Dismissing authentic transcripts and corroborating documents as fabrications and "fake news".

  • In-Group versus Out-Group Biases:

    • These rhetorical strategies activate primitive in-group/out-group mechanisms, framing political disagreements not as disputes over empirical evidence and constitutional standards, but as existential battles between an innocent, pure in-group and an evil out-group seeking cultural destruction.

    • By stimulating evolutionary defenses, cognitive coherence evaluation is shut down, ensuring tribal loyalty persists regardless of objective evidence.