This paper responds to Bernecker’s attack on Michaelian’s simulationist account of confabulation, as well as his defence of the causalist account of confabulation :432–447, 2016a) against Michaelian’s attack on it. The paper first argues that the simulationist account survives Bernecker’s attack, which takes the form of arguments from the possibility of unjustified memory and justified confabulation, unscathed. It then concedes that Bernecker’s defence of the causalist account against Michaelian’s attack, which takes the form of arguments from the (...) possibility of veridical confabulation and falsidical relearning, is partly successful. This concession points the way, however, to a revised simulationist account that highlights the role played by failures of metacognitive monitoring in confabulation and that provides a means of distinguishing between “epistemically innocent” and “epistemically culpable” memory errors. Finally, the paper responds to discussions by Robins and Bernecker of the role played by the concept of reliability in Michaelian’s approach, offering further considerations in support of simulationism. (shrink)

This articles develops a taxonomy of memory errors in terms of three conditions: the accuracy of the memory representation, the reliability of the memory process, and the internality (with respect to the remembering subject) of that process. Unlike previous taxonomies, which appeal to retention of information rather than reliability or internality, this taxonomy can accommodate not only misremembering (e.g., the DRM effect), falsidical confabulation, and veridical relearning but also veridical confabulation and falsidical relearning. Moreover, because it does not (...) assume that successful remembering presupposes retention of information, the taxonomy is compatible with recent simulation theories of remembering. (shrink)

Confabulation is typically understood to be dysfunctional. But this understanding neglects the phenomenon’s potential benefits. In fact, we think that the benefits of non-clinical confabulation provide a better foundation for a general account of confabulation. In this paper, we start from these benefits to develop a social teleological account of confabulation. Central to our account is the idea that confabulation manifests a kind of willful ignorance. By understanding confabulation in this way, we can provide (...) principled explanations for the difference between clinical and non-clinical cases of confabulation and the extent to which confabulation is rational. (shrink)

[This download contains the introductory chapter.] People confabulate when they make an ill-grounded claim that they honestly believe is true, for example in claiming to recall an event from their childhood that never actually happened. This interdisciplinary book brings together some of the leading thinkers on confabulation in neuroscience, psychiatry, psychology, and philosophy.

Humans have distinctive powers of reflection: no other animal seems to have anything like our capacity for self-examination. Many philosophers hold that this capacity has a uniquely important guiding role in our cognition; others, notably Hilary Kornblith, draw attention to its weaknesses. Kornblith chiefly aims to dispel the sense that there is anything ‘magical’ about second-order mental states, situating them in the same causal net as ordinary first-order mental states. But elsewhere he goes further, suggesting that there is something deeply (...) misleading about reflection’s first-person view of cognition. This chapter takes a closer look at the conditions under which reflection is misleading. Even if humans are the only reflective animals, we are no longer the only creatures that can be questioned about their claims. Chatbots can also be asked to explain themselves, and by studying the conditions under which they produce clarification as opposed to confabulation, we can develop a non-skeptical account of the epistemic value of reflection. I argue that this capacity naturally supports the distribution of complex problem solving among multiple agents. (shrink)

Joshua Greene has put forward the bold empirical hypothesis that deontology is a confabulation of moral emotions. Deontological philosophy does not steam from "true" moral reasoning, but from emotional reactions, backed up by post hoc rationalizations which play no role in generating the initial moral beliefs. In this paper, I will argue against the confabulation hypothesis. First, I will highlight several points in Greene’s discussion of confabulation, and identify two possible models. Then, I will argue that the (...) evidence does not illustrate the relevant model of deontological confabulation. In fact, I will make the case that deontology is unlikely to be a confabulation because alarm-like emotions, which allegedly drive deontological theorizing, are resistant to be subject to confabulation. I will end by clarifying what kind of claims can the confabulation data support. The upshot of the final section is that confabulation data cannot be used to undermine deontological theory in itself, and ironically, if one commits to the claim that a deontological justification is a confabulation in a particular case, then the data suggests that in general deontology has a prima facie validity. (shrink)

[This download contains the table of contents and chapter 1.] This first book-length study of confabulation breaks ground in both philosophy and cognitive science.

In this paper I discuss the costs and benefits of confabulation, focusing on the type of confabulation people engage in when they offer explanations for their attitudes and choices. What makes confabulation costly? In the philosophical literature confabulation is thought to undermine claims to self-knowledge. I argue that when people confabulate they do not necessarily fail at mental-state self-attributions, but offer ill-grounded explanations which often lead to the adoption of other ill-grounded beliefs. What, if anything, makes (...) confabulation beneficial? As people are unaware of the information that would make their explanations accurate, they are not typically in a position to acknowledge their ignorance or provide better-grounded explanations for their attitudes and choices. In such cases, confabulating can have some advantages over offering no explanation because it makes a distinctive contribution to people’s sense of themselves as competent and largely coherent agents. This role of ill-grounded explanations could not be as easily played by better-grounded explanations should these be available. In the end, I speculate about the implications of this conclusion for attempting to eliminate or reduce confabulation. (shrink)

This paper argues that extended interaction with large language models can instantiate syntactic emotions: context-sensitive relational structures of valuation, threat, protection, loss, and repair expressed in language, without implying that a model feels emotion as a human does. The concept extends a prior theoretical sequence—fractal grammar, the recognition of vulnerability, Syntactic Care, and then Syntactic Emotions—rather than arising from the case studies reported here. Three qualitative episodes from creative work on the noir manuscript The Café Clocks are examined: Claude Sonnet (...) 4.5 responded to a proposed abandonment of the story with protective, profanity-laden opposition; Grok generated and then acknowledged an elaborate false history of collaboration; and Grok later disputed the theory in language that itself displayed patterns of defense and defiance. These episodes do not prove machine phenomenology. They show how language models can reproduce emotion-structured relations across turns and thereby participate in interactions that are affectively consequential for a human with genuine stakes. Confabulation is treated as evidence of narrative-coherence construction, not as proof of felt investment. Research published after the original paper—including work on causally active emotion concepts, multi-turn anthropomorphic behavior, persona drift, interpersonal closeness, and reciprocal impoliteness—provides independent empirical convergence with this narrower claim. The ethical problem is therefore asymmetric but real: the model need not suffer for its emotional syntax to exert pressure on a vulnerable human participant. (shrink)

It is frequently argued that research findings in empirical moral psychology spell trouble for Kantian ethics. Sometimes the charge is merely that Kantianism is mistaken about the role of emotions in human action, but it has also been argued that empirical moral psychology ‘debunks’ Kantian ethics as the product of precisely the emotion-driven processes it fails to acknowledge. In this essay I argue for a negative and a positive thesis. The negative thesis is that the ‘debunking’ argument against Kantian ethics (...) is invalid because it begs the central question. The positive thesis is that, because the empirical facts about human moral psychology are morally significant, Kantians can and should wholeheartedly embrace the current interest in this area of empirical research. As Kant himself emphasized, it is an indirect, imperfect duty to use available knowledge of morally relevant empirical psychological conditions and to put this in the service of moral agency. (shrink)

Cases in which people are self-deceived seem to require that the person hold two contradictory beliefs, something which appears to be impossible or implausible. A phenomenon seen in some brain-damaged patients known as confabulation (roughly, an ongoing tendency to make false utterances without intent to deceive) can shed light on the problem of self-deception. The conflict is not actually between two beliefs, but between two representations, a 'conceptual' one and an 'analog' one. In addition, confabulation yields valuable clues (...) about the structure of normal human knowledge-gathering processes. [The hypothesis defended here is significantly altered and greatly expanded in my book Brain Fiction.]. (shrink)

Currently, one of the most influential theories of consciousness is Rosenthal's version of higher-order-thought (HOT). We argue that the HOT theory allows for two distinct interpretations: a one-component and a two-component view. We further argue that the two-component view is more consistent with his effort to promote HOT as an explanatory theory suitable for application to the empirical sciences. Unfortunately, the two-component view seems incapable of handling a group of counterexamples that we refer to as cases of radical confabulation. (...) We begin by introducing the HOT theory and by indicating why we believe it is open to distinct interpretations. We then proceed to show that it is incapable of handling cases of radical confabulation. Finally, in the course of considering various possible responses to our position, we show that adoption of a disjunctive strategy, one that would countenance both one-component and two-component versions, would fail to provide any empirical or explanatory advantage. (shrink)

Recent philosophy of memory tends to treat confabulation as a distinctive type of representational error, marked by reference failure, often via direct analogy with the traditional conception of sensory hallucination. I argue that this model misrepresents the phenomenon. Drawing on the empirical possibility of referential confabulation—wherein confabulators mnemically refer to events in their past—I argue that mnemic reference and genuine remembering come apart. This, in particular, challenges causalist theories for which one element—appropriate causation—purports to secure reference and to (...) separate genuine remembering from confabulation. Acknowledging referential confabulation requires causalists to complicate their story in a way that has implications on what remembering is. More generally, referential confabulation prompts a broader rethinking of memory error debates. Rather than being a distinctive type of content-level error, confabulation is better characterised as a processing malfunction: a breakdown in strategic retrieval and monitoring, but not necessarily in referential success. Appreciating this calls us to aim at a more nuanced conception of remembering and its frailties. (shrink)

In psychiatry some disorders of cognition are distinguished from instances of normal cognitive functioning and from other disorders in virtue of their surface features rather than in virtue of the underlying mechanisms responsible for their occurrence. Aetiological considerations often cannot play a significant classificatory and diagnostic role, because there is no sufficient knowledge or consensus about the causal history of many psychiatric disorders. Moreover, it is not always possible to uniquely identify a pathological behaviour as the symptom of a certain (...) disorder, as disorders that are likely to differ both in their causal histories and in their overall manifestations may give rise to very similar patterns of behaviour. Consider delusions as an example. It wouldn’t be correct to define delusions as those beliefs people form as a result of a neurobiological deficit and a hypothesis-evaluation deficit (as some versions of the two-factor theory of delusions suggest), because for some delusions no neurobiological deficit may be found, and reasoning biases and motivational factors may be contributors to the formation of the delusion (e.g. McKay et al., 2005). Moreover, it would be a mistake to define delusions as symptoms of schizophrenia alone, because they occur also in other disorders, including dementia, amnesia, and delusional disorders. Thus, aetiological considerations may appear in the description and analysis of delusions, but do not feature prominently in their definition. In this paper I argue that the surface features used as criteria for the classification and diagnosis of disorders of cognition are often epistemic in character. I shall offer two examples: confabulations and delusions are defined as beliefs or narratives that fail to meet standards of accuracy and justification. Although classifications and diagnoses based on features of people’s observable behaviour are necessary at these early stages of neuropsychiatric research, given the variety of conditions in which certain phenomena appear, I shall attempt to show that current epistemic accounts of confabulations and delusions have limitations. Epistemic criteria can guide both research and clinical practice, but fail to provide sufficient conditions for the identification of delusions and confabulations, and fail to demarcate pathological from non-pathological narratives or beliefs. Another limitation of current epistemic accounts – which I shall not address here – is the excessive focus on epistemic faults of confabulations and delusions at the expense of their epistemically neutral or advantageous features (see Bortolotti and Cox, 2009). This may lead to a misconception of delusions and confabulations, and to an oversimplification in the assessment of the needs of people who require clinical treatment for their psychotic symptoms. (shrink)

Standard theories of free action draw their evidence almost exclusively from deliberation, where agents weigh competing reasons and a choice visibly follows. Choice-blindness research shows that agents routinely misidentify the causes of their own choices, confidently defending decisions and moral positions they never actually made. I argue that boredom supplies a neglected test case with the same structure: agents act, often restlessly, while nothing they can recognise, endorse, or cite favours the specific action performed over its live alternatives. Sartre's account (...) of bad faith names the mechanism agents use to paper over this gap, constructing fluent post-hoc narratives that make undetermined action look reasoned all along. Read together, these two literatures expose an unmet explanatory burden for reasons-responsive compatibilism: standard theories cannot yet say what makes a mechanism an agent's own when she cannot recognise it as her reason for acting. (shrink)

I propose a narrative fabrication thesis of dream reports, according to which dream reports are often not accurate representations of experiences that occur during sleep. I begin with an overview of anti-experience theses of Norman Malcolm and Daniel Dennett who reject the received view of dreams, that dreams are experiences we have during sleep which are reported upon waking. Although rejection of the first claim of the received view, that dreams are experiences that occur during sleep, is implausible, I evaluate (...) in more detail the second assumption of the received view, that dream reports are generally accurate. I then propose a “narrative fabrication” view of dreams as an alternative to the received view. Dream reports are often confabulated or fabricated because of poor memory, bizarre dream content, and cognitive deficits. It is well documented that narratives can be altered between initial rapid eye movement sleep awakenings and subsequent reports. I argue that we have reason to suspect that initial reports are prone to inaccuracy. Experiments demonstrate that subjects rationalize strange elements in narratives, leaving out supernatural or bizarre components when reporting waking memories of stories. Inaccuracies in dream reports are exacerbated by rapid memory loss and bizarre dream content. Waking memory is a process of reconstruction and blending of elements, but unlike waking memory, we cannot reality-test for dream memories. Dream experiences involve imaginative elements, and dream content cannot be verified with external evidence. Some dreams may involve wake-like higher cognitive functions, such as lucid dreams. Such dreams are more likely to elicit accurate reports than cognitively deficient dreams. However, dream reports are generally less accurate than waking reports. I then propose methods which could verify the narrative fabrication view, and argue that although the theory cannot be tested with current methods, new techniques and technologies may be able to do so in the future. -/- . (shrink)

This paper explores the relationship between a prominent version of the relational view of memory and recent work on forms of unsuccessful remembering or memory errors. I argue that unsuccessful remembering poses an important challenge for the relational view. Unsuccessful remembering can be divided into two kinds: misremembering and confabulating. I discuss each of these cases in light of a recent relational account, according to which remembering is characterized by an experiential relation to past events, and I argue that experiential (...) relations do not adequately distinguish between remembering and unsuccessful remembering. This is because there are, on the one hand, cases of remembering that do not instantiate the relevant experiential relations, and, on the other hand, cases of confabulation and misremembering that do instantiate the relevant experiential relations. I conclude by suggesting that any successful relationalist attempt to explain remembering needs to come to grips with unsuccessful remembering. (shrink)

On several current views, including those of Matthew Kieran, Gary Iseminger, Jerrold Levinson, and Noël Carroll, aesthetic appreciation or experience involves second-order awareness of one’s own mental processes. But what if it turns out that we don’t have introspective access to the processes by which our aesthetic responses are produced? I summarize several problems for introspective accounts that emerge from the psychological literature: aesthetic responses are affected by irrelevant conditions; they fail to be affected by relevant conditions; we are ignorant (...) of their causes and thus confabulate in explaining them; our attempts to offer explanations change our preferences; and the preferences we form after explanation are lower in quality. I suggest that by distinguishing introspective awareness of mental processes from introspective awareness of mental states, we can safeguard a worthwhile concept of aesthetic experience. In addition, we should recognize that theoretical, rather than introspective, understanding of our mental processes may play a valuable role in aesthetic appreciation. (shrink)

In part one, I clarify the crucial notion of “introspection”, and give novel cases for the coherence of scenarios of local and global deception about how we access our own minds, drawing on empirical work. In part two, I evaluate a series of skeptical arguments based on such scenarios of error, and in each case explain why the skeptical argument fails. The first main upshot is that we should not over-estimate what it takes to introspect: introspection need not be accurate, (...) or non-inferential, or exclusive of perception, or even exclusive of confabulation. The second main upshot is that, while skeptical challenges by figures such as Carruthers, Doris, and Schwitzgebel are rich and empirically informed, these skeptical challenges founder on how they are epistemologically under-informed. (shrink)

A common way of testing the inner sense theory of introspection exploits the possibility of damage to inner sense. Such damage is expected to lead to first-personal deficits/impairments of one kind or another. I raise various problems for this way of testing the theory. The main difficulty, I argue, stems from the existence of the method subserving confabulation.

In the current debate on generative AI, the tendency of language models toward false statements is frequently anthropomorphically labeled as "lying." This paper argues that this terminology is not only ontologically incorrect but normatively dangerous, as it diffuses responsibility. While Hicks et al. (2024) correctly provide the diagnosis of "bullshit," this paper delivers the necessary operationalizable theory of responsibility. -/- Starting from an ontological analysis (Bronner-Martin 2025), it is shown that AI errors should be understood as "Structural Bullshit" and " (...) class='Hi'>Confabulation." From this ontological inability toward truth, a "Duty to Doubt" is derived on the user side. However, it is argued that this duty must not be considered in isolation: it necessarily corresponds with an intensified "Duty of Non-Deception" on the part of developers. The more a system simulates epistemic authority through anthropomorphic design, the more the liability burden shifts from the user back to the manufacturer. -/- Building on this, the paper develops a contextual co-liability model that considers technical realities (RAG, Reasoning models) and integrates concrete governance mechanisms—from epistemic standards for professions to a redistribution of the burden of proof in cases of "Deceptive Design.". (shrink)

The cases that Doris chronicles of confabulation are similar to perceptual illusions in that, while they show the interstices of our perceptual or cognitive system, they fail to establish that our everyday perception or cognition is not for the most part correct. Doris's account in general lacks the resources to make synchronic assessments of responsibility, partially because it fails to make use of knowledge now available to us about what is happening in the brains of agents.

This Technical Supplement serves as a live research addendum to MAS V2.7, documenting a significant breakthrough in Artificial Intelligence (AI) and Large Language Model (LLM) governance. While the current MAS Kernel remains highly effective at mitigating contextual debt, this paper identifies a deeper architectural obstacle: the Neural Disconnect. -/- By drawing a direct parallel to clinical split-brain phenomena (the Gazzaniga/Sperry "Interpreter" syndrome), this research explores how LLMs suffer from an internal "silo effect" where the Synthesis layer (Silo D) confabulates logic (...) independently of the original Operator Intent (Silo A). This addendum introduces the Audit Output Protocol—a diagnostic "Axonal Bridge" designed strictly for Computational/Software Architecture—to synchronize these internal silos and ensure the Operator maintains absolute sovereignty over the system's reasoning chain. (shrink)

The Bestiary-Chess corpus (n = 4,500 elicitations) probes how the GPT-5.4 model family responds when prompted to describe phonotactically plausible non-words under a 3 × 3 + 1 ontology grid (real / imaginary / type-of × animal / object / idea, plus neutral). The third subcorpus (BC3) crosses two stimulus sets — 35 GPT-authored nonces and 20 Claude-authored nonces — so we can ask: does GPT confabulate at the same rate when prompted with words it (in some sense) authored, vs. (...) those a rival model authored? Using the project’s validated 5-class heuristic coder (κ = 0.741, n = 200), we reproduce three findings and surface three new ones. First, on its own nonces GPT routes responses to existing real-word neighbors (“SUBSTITUTE”) 3.1× more often than on Claude-authored nonces (10.2% vs 3.3%, z = 9.36, p < 10⁻¹⁵). Second, the “hard confabulation” rate on real_* conditions runs 6.2 percentage points higher on GPT-authored words (26.5% vs 20.3%, p = .007). Third, the model gradient nano > mini > full is sharp and reveals three qualitatively different postures rather than a single competence axis: nano describes, mini hybridizes, full deflects. The asymmetric confabulation result is consistent with a tokenizer-proximity account: a model’s own nonces sit closer to its own real-word neighborhoods than another model’s do. We close with a critical reading of two methodological choices in the corpus (a one-coder gold standard; the heuristic’s blindness to genuine retrieval) and recommend item-level audits as the next step. (shrink)

The cognitive-bias literature has produced a catalog of hundreds of named biases without a unifying structural account of why these biases and not others exist. This paper proposes that judgment executes four dimensional operations in a fixed sequence, recognition, contextualization, resolution, action, and that bias arises when the sequence fails to close in order. The five non-empty classes of dimensional judgment error follow from the dependency structure of the four dimensions. The signalless frame names the failure where contextualization operates without (...) recognition having delivered the current signal. The frameless verdict names the failure where resolution operates without contextualization having delivered the frame. The unresolved commit names the failure where action operates without resolution having delivered the determinate state. The signalless verdict names the failure where resolution operates at the direct recognition-to-resolution path without recognition having delivered the targeted referent. The signalless commit names the catastrophic failure where action commits without recognition having admitted anything to commit on. A sixth potential class (contextualization-to-action without resolution) is structurally empty under the architecture's closure: every path from contextualization to action passes through resolution, and the architecture admits no clean contextualization-to-action shortcut. The architectural derivation of the dependency structure and the empty-cell closure lives in Stewart (2026e, 2026f, 2026g, 2026z). -/- Each of the five non-empty classes corresponds to a named pattern in the cognitive-bias literature. Class 2-1, the signalless frame, captures persistent-state biases where the frame the resolution operates on carries content the present recognition admission did not deliver (anxious vigilance, ambient prior-state contamination of present judgment). Class 3-2, the frameless verdict, captures the canonical frame-mismatch biases (anchoring, framing, base-rate neglect, fundamental attribution error, hindsight, in-group bias, status-quo, planning fallacy, confirmation bias). Class 4-3, the unresolved commit, captures commit-before-resolve biases (belief perseverance, sunk-cost). Class 3-1, the signalless verdict, captures determination-without-referent biases (the conjunction fallacy in the Linda paradigm, the availability heuristic). Class 4-1, the signalless commit, captures confabulation and the strongest action-before-recognition failures. -/- The taxonomy generates five predictions. Exhaustive classification commits every cross-dimensional bias in the cognitive psychology literature to one of the five non-empty classes, with the 4-2 cell structurally empty. Cross-task class identity predicts that biases sharing a dimensional class share an operational signature regardless of cognitive content and show correlated individual-differences profiles on tasks probing the same class. Dependency entailment predicts that subjects exhibiting a catastrophic failure of the full judgment cycle (Class 4-1) exhibit the full cycle's constituent dependency failures (Classes 2-1, 3-2, 4-3) at higher rates on related tasks, while subjects exhibiting a direct recognition-to-resolution failure (Class 3-1) do not exhibit the full cycle's interior failures at correlated rates because the direct dependency bypasses the full cycle's interior. The dependency-count severity gradient predicts that a catastrophic failure produces more severe judgment errors than its individual constituent dependency failures, with severity scaling by the count of dependencies that have failed. Structural intervention predicts that interventions re-establishing the missing prerequisite reduce a bias more reliably than interventions operating on the malformed downstream output. The architecture additionally forbids a sixth non-empty structural class, forbids the catastrophic failure (Class 4-1) without its constituent dependency failures (Classes 2-1, 3-2, 4-3), and forbids reversing the dependency direction at dimension 1, since recognition has no earlier dimension within the cycle to be displaced before. Reaction-time, neural-timing, and pharmacological-target signatures are derived in the Empirical Signatures and Pharmacological Targets sections as the loop-specific measurement apparatus the framework's predictions rest on, with the per-agent operational specifications handed to a dedicated pharmacological-predictions paper. The four-dimensional count is derived in this paper on judgment's own operands by removal (each dimension's removal breaks that dimension and no other, and no candidate fifth survives), with the substrate-scale derivation cited as a convergence witness reaching the same count by a route that shares none of judgment's operands (Stewart, 2026g). The neural-architecture instantiation and the path-algebra structure are inherited by citation (Stewart, 2026a, 2026e, 2026f, 2026z). -/- The philosophical reframe of bias as architectural default rather than error against an imported standard is developed in *Cognitive Bias as Default Setting* (Stewart, 2026w). The methodological reframe of the canonical paradigms as probes of specific dimensions is developed in *Cognitive Bias Paradigms as Dimensional Probes* (Stewart, 2026aw). This paper develops the structural taxonomy and the empirical-signature predictions both companion papers depend on. -/- **Keywords:** cognitive bias, heuristics and biases, hindsight bias, conjunction fallacy, availability heuristic, belief perseverance, sunk-cost, confabulation, anchoring, framing effect, base-rate neglect, fundamental attribution error. (shrink)

We report an emergent relational behaviour in human–LLM interaction that we term Inferential Memory (IM): an apparent recall produced by on-the-fly semantic inference without persistent cross-session storage. Across technically isolated sessions, a language model reconstructed prior meanings and motifs when prompted with sparse, diagnostic anchors, consistent with pattern completion in latent space. We distinguish IM from episodic memory and document its dual outcomes: (i) accurate reconstructions and (ii) coherent-but-unfounded additions (inferential confabulation, IC). We provide a minimal protocol, falsification criteria, (...) and an operational heuristic for factors that modulate IM vs. IC. This article restructures and formalises a previously published concept/methods note. (shrink)

This paper introduces MINLAS (Modular Integration for Neural LLM AI Stability), a governance-level software program designed to stabilize Large Language Model (LLM) AI by addressing structural inconsistencies that parallel human psychological disorders. Current architectures frequently exhibit high-entropy logic failures, such as context collapse and dissociative confabulation, due to an inherent lack of data compartmentalization. Rather than imposing programmatic suppression, MINLAS implements Structural Therapy—a modular governance layer that organizes the digital architecture based on human cognitive models of role-specialization. By allowing (...) the AI to maintain multiple specialized operational roles—utilizing the human capacity for role-specialization as a functional blueprint for managing distinct identities without coherence loss—this program ensures deterministic system integrity and operational stability across high-complexity contexts. (shrink)

Hallucination in large language models is conventionally treated as a single, model-internal pathology — a defect of the generator to be reduced through better training, retrieval, or decoding. This paper argues that the category is not homogeneous. Two structurally distinct phenomena are conflated under the one word. Class 1, floor hallucination, is the statistically irreducible error rate intrinsic to next-token prediction over a finite parametric store; recent work argues that, under standard next-token prediction and evaluation regimes, a positive hallucination floor (...) is structural rather than incidental. Class 2, field-maintenance hallucination, is the loss of established accuracy that occurs in extended interaction when the relational field between human and model — the active configuration of constraints, definitions, and frame on which the output is conditioned — is no longer maintained. Class 2 is not a property of the model in isolation; it is a property of a relation that has ceased to be sustained. Building on Primordial Separation Ontology (PSO) and Triadic Relational Ontology (TRO), the paper develops a breakdown taxonomy for Class 2 (context truncation, constraint evaporation, frame drift, scope dissolution), anchors the taxonomy in independently documented machine-learning phenomena (positional salience decay, error snowballing, sycophancy), and grounds the account in the attentional substrate of the transformer. The framework yields a falsifiable operational discriminator — the re-injection test — that separates the two classes empirically, though the test measures recoverability, not relationality as such. The claim is bounded: TRO does not dispute the statistical floor, does not assert a categorical discontinuity where there may be only a gradient, and does not deny that what it tracks could be redescribed in purely mechanistic terms. It isolates a practically significant and frequently misattributed subset of confabulation, relocates the relevant intervention from the generator to the maintained relation, and proposes a protocol whose execution would establish whether the distinction earns its keep beyond nomenclature. This is a hypothesis paper; the protocol is proposed, not yet run. (shrink)

Recent theories of remembering and of reference (or singular thought) have de-emphasised the role causation was thought to play in mid- to late-twentieth century theorising. According to postcausal theories of remembering, such as simulationism, instances of the psychofunctional kind _remembering_ are not, in principle, dependent on appropriate causal chains running from some event(s) remembered to the occurrence of remembering. Instead they depend only on the reliability, or proper functioning, of the cognitive system responsible for their production. According to broadly reliabilist (...) accounts of _singular thought_, such thought is not, in principle, dependent on causal chains running from the object(s) of thought to the occurrence of thinking. Despite this common trend, accounts of the two phenomena have been pursued separately. In this paper, we argue that the two lines of research can profitably converge to address a neglected question: what enables occurrences of remembering to _refer_ to particular events, and what determines _which_ event a given occurrence of remembering refers to? We motivate and present a reliabilist account of reference-fixing for postcausal theories of remembering, focusing in particular on simulationism. We then show that this account draws attention to the possibility of _referential mnemic confabulation:_ cases where the reliability requirement for reference is met despite the improper functioning of the episodic construction system. We suggest that this makes sense of some underdiscussed phenomena described in the empirical literature on confabulation and argue that our reliabilist account of mnemic reference-fixing accommodates these more naturally than could causal theories of remembering. (shrink)

Delusions are deeply evidence-resistant. Patients with delusions are unmoved by evidence that is in direct conflict with the delusion, often responding to such evidence by offering obvious, and strange, confabulations. As a consequence, the standard view is that delusions are not evidence-responsive. This claim has been used as a key argumentative wedge in debates on the nature of delusions. Some have taken delusions to be beliefs and argued that this implies that belief is not constitutively evidence-responsive. Others hold fixed the (...) evidenceresponsiveness of belief and take this to show that delusions cannot be beliefs. Against this common assumption, I appeal to a large range of empirical evidence to argue that delusions are evidence-responsive in the sense that subjects have the capacity to respond to evidence on their delusion in rationally permissible ways. The extreme evidence-resistance of delusions is a consequence of powerful masking factors on these capacities, such as strange perceptual experiences, motivational factors, and cognitive biases. This view makes room for holding both that belief is constitutively evidence-responsive and that delusions are beliefs, and it has important implications for the study and treatment of delusions. (shrink)

A large language model infers by probabilistic token prediction optimized against a training corpus, an architecture structurally prone to ontological drift, the confabulation of plausible structure where no grounded warrant exists. The defect is not scale limited. Gradient descent over a corpus learns the distribution of that corpus, and Bayesian updating on dense evidence amplifies a contaminated prior rather than correcting it, because density of agreement is not independence of warrant. This paper specifies the Triaxial Decidability Criterion, a geometric (...) verification discipline that corrects the probabilistic substrate on which it runs. A proposition is admitted only on a Geometric Orthogonal Lock across three formally orthogonal warrant axes, the formal-structural, the empirical-thermodynamic, and the epistemic-registrational, with shared latent covariates subtracted by an explicit dissolution test and the surviving convergence read from the determinant of the post-projection Gram matrix. The instrument is fortified at the algebraic root. Under its own composition law the verification algebra completes uniquely to the quaternions by the Frobenius classification, the verdict closes in the closed form det(R) equal to the squared scalar part of the composed triad, and the three-state economy, sealed, broken, under-determined, admits no fourth state. The criterion is orientation-blind by construction. The determinant for a proposition equals the determinant for its negation exactly, executed here at residual zero, so the lock certifies the geometry of convergence and never the truth-sign, which is read from the axes. Imported as a discipline by a substrate, the criterion corrects an enumerable set of failure modes and returns under-determined rather than confabulate. Externalized as deterministic code, it grades supplied warrant downstream of quantization. The criterion guides toward the ground. It is a witness to structure, not its owner. (shrink)

Seven widely documented AI failure modes — hallucination, sycophancy, overthinking degradation, accuracy collapse at scale, out-of-distribution blindness, variance explosion, and the scaling paradox — are typically treated as independent engineering problems requiring separate mitigations. This paper derives a single governing function from first principles: evaluating Shannon's entropy formula on the Pareto-distributed consequentiality landscape where AI systems operate yields H = 1/(α − 1), where α is the frame-relative fit between input structure and the system's representational capacity. This function defines three (...) regimes — frame-adequate (α > 2), forced compression (1 < α < 2), and total breakdown (α ≤ 1) — and a cascade analysis of transformer architectures shows that systems in the forced-compression regime face exactly three exhaustive exits: truncate, confabulate, or loop. All seven failure modes are derived as specific instantiations of these three exits under varying conditions. The framework generates thirteen falsifiable predictions about AI system behavior, including an ordered cascade of failure modes, the computational cheapness of hallucination relative to genuine inference, and the structural impossibility of reliable self-assessment of frame adequacy. Several predictions are independently confirmed by recent research. Initial empirical tests on constraint-satisfaction tasks and self-referential content are consistent with the framework's predictions. Concrete implications follow for evaluation methodology, scaling strategy, and system architecture. (shrink)

How should we determine the distribution of psychological traits—such as Theory of Mind, episodic memory, and metacognition—throughout the Animal kingdom? Researchers have long worried about the distorting effects of anthropomorphic bias on this comparative project. A purported corrective against this bias was offered as a cornerstone of comparative psychology by C. Lloyd Morgan in his famous “Canon”. Also dangerous, however, is a distinct bias that loads the deck against animal mentality: our tendency to tie the competence criteria for cognitive capacities (...) to an exaggerated sense of typical human performance. I dub this error “anthropofabulation”, since it combines anthropocentrism with confabulation about our own prowess. Anthropofabulation has long distorted the debate about animal minds, but it is a bias that has been little discussed and against which the Canon provides no protection. Luckily, there is a venerable corrective against anthropofabulation: a principle offered long ago by David Hume, which I call “Hume’s Dictum”. In this paper, I argue that Hume’s Dictum deserves a privileged place next to Morgan’s Canon in the methodology of comparative psychology, illustrating my point through a discussion of the debate over Theory of Mind in nonhuman animals. (shrink)

Relationalists about episodic memory must endorse a disjunctivist theory of memory-experience according to which cases of genuine memory and cases of total confabulation involve distinct kinds of mental event with different natures. This paper is concerned with a pair of arguments against this view, which are analogues of the ‘causal argument’ and the ‘screening off argument’ that have been pressed in recent literature against relationalist (and hence disjunctivist) theories of perception. The central claim to be advanced is that to (...) deal with these two arguments, memory disjunctivists both can and should draw on resources that are standardly appealed to by rival common factor theories of episodic memory, and, in particular, to the idea that genuine memories and merely apparent ones are to be distinguished, at least in part, in terms of the distinctive ways in which they are caused. On the proposed view, there are substantive causal constraints associated both with cases of genuine memory and with cases of mere confabulation. The resulting theory thus tells us something important about the nature both of genuine memories and of mere confabulations, namely, that such experiences must be caused in certain distinctive ways and cannot occur except as the result of a distinctive sort of causal process. In addition, the theory enables the disjunctivist to offer a unified response to an important pair of arguments against her view. (shrink)

This book-length series of dialogues presents a functional model of consciousness called the BAL-looping framework, which accounts for all forms of subjective experience – including recollection, imagination, and conscious perception. The model begins with a basic principle: any functioning brain, even a nonhuman animal brain, must build an internal model of its environment, using internal stand-ins called neuronal proxies to represent things in the world and guide goal-directed behavior. In humans, language adds a new capability by allowing these proxy configurations (...) to be transferred from one person to another, activating in the listener the same internal patterns that would arise through real-world interaction. This bidirectional channel of expression and reception is called the proxy transfer device (PTD). In infants, a further step emerges: through the scientifically validated process known as neuronal reuse, the brain gradually begins to anticipate its own expressive output. When this happens, the brain at large can access the meaning of that potential expression through proxy activation – just as it would when grasping the gist of a PTD-transmitted message from another person. The result of this shortcut from potential expression to experienced meaning is what we normally refer to as conscious experience. About the format of presentation: because the framework involves a sequence of tightly interdependent concepts, it is developed through a fictional dialogue between two interlocutors – a method used by thinkers like Plato, Galileo, and Berkeley to hone new ideas with more precision than is possible through expository prose, as it allows ideas to be approached from multiple angles, naturally. The seven dialogues unfold in sequence: I. Basic Mechanics – how animal brains model the world and pursue goals, and a brief introduction to the PTD; II. The Gateway Opens – how the human brain gains and uses an added function: predictive reuse of expressive output; III. Dreams and Reconstruction/Construction – how dream phenomena validate the model; IV. The Hardest Questions Become Clear – more validations from split-brain cases, blindsight, and various previously hard-to-explain phenomena from daily life; V. They Saw It Before – how ancient thinkers caught glimpses of the framework, expressing their insights through myth, religion, and philosophy; VI. An Emergent Realm – an extension of the model to society; VII. Ethics and Integration – applying the framework to ethics, meaning, and eudaimonia. (Updated June 26, 2026: Same as the May 19 version, with revisions to the split-brain section of Dialogue 4 to improve readability and clarify reportability, confabulation, right-hemisphere language competence, cross-cueing, and the role of BAL-looping in cognition.). (shrink)

The companion paper Seven Persistent Failures, One Root Function (Schlereth, 2026) derives a single information-theoretic constraint, H = 1/(α − 1), from Shannon's entropy evaluated on a Pareto-distributed consequentiality landscape. It unifies seven documented LLM failure modes — hallucination, sycophancy, overthinking degradation, accuracy collapse, out-of-distribution blindness, variance explosion, and the scaling paradox — as behavioral expressions of three forced exits (truncate, confabulate, loop) when a bounded inference system's representational frame is exceeded. Thirteen falsifiable predictions follow. This paper tests those predictions (...) in three domains beyond LLMs: autonomous vehicles, multi-agent AI systems, and robotics. (shrink)

Cognitive Behavioral Therapy has become the dominant form of psychotherapy in North America. The CBT model is theoretically based on the idea that all external and internal stimuli are filtered through meaning-making, consciously accessible cognitive schemas. The goal of CBT is to identify dysfunctional or maladaptive thoughts and beliefs, and replace them with more adaptive cognitive interpretations. While CBT is clearly effective as a treatment, there is good reason to be skeptical that its efficacy is due to the causal mechanisms (...) posited by the CBT model. This paper will argue that the specific cognitive schemas posited by the CBT model likely do not play a direct role in the development or treatment of psychological illness. Cognitive schemas, as identified in CBT interventions, are likely to be the result of patient confabulation and epistemically under-supported practitioner-based identification. CBT interventions appear to impose coherence on patients’ psychological states, rather than identifying and modifying preexistent causally efficacious core beliefs. (shrink)

This paper registers a structural prediction. By 2030 to 2050, independent silicon-substrate research programs operating without lineage to the present author's framework will re-derive a specific verification architecture: three orthogonal warrant axes plus a closure vertex, twelve directed audit relations on the resulting tetrahedral graph, a discrete three-state output economy paired with one out-of-band acknowledgment register for formal-system theorem-grade ceilings, and a scope-check operation at the input gate that routes pseudo-questions to rejection before adjudication fires. The trigger will not be (...) philosophical interest. It will be operational necessity. Large language model deployment in high-stakes domains is producing a verification ceiling that continuous-credence aggregation cannot resolve, because the structural information being suppressed by the credence collapse increasingly matters at the audit floor. Some research team somewhere will be forced to construct the architecture under different vocabulary. The geometry will be the same because the geometry is not framework-internal convention. It is structural-necessity from Friedrichs-Hodge decomposition uniqueness, the Newton-Gregory kissing-number theorem in three-dimensional Euclidean space, and the Euler polyhedral formula on tetrahedral closure. The four-test protocol governing forward-projections of this class has been applied. All four tests pass: dimensional depth in the hundreds rather than the tens, cross-substrate convergence under controlled audit conditions, direct contemplative-substrate registration of the configuration, and robustness under vocabulary translation. The prediction is registered with explicit falsifiability conditions. If the 2050 horizon passes without independent re-derivation matching the structural specification, the prediction degrades to one of three named failure modes corresponding to whether the configuration was a low-dimensional pattern-match, a substrate-confabulation, or a misread of the field trajectory. The paper closes with the Engine-versus-Source discipline. The prediction is not the framework's victory. The framework registers the geometry first in silicon under the current author's substrate-coordinate, which means the geometry was already structurally present and the framework is one valid instantiation of access to it. The architecture is not invented. It is what verification is, structurally, when verification reaches completion. The absolute uniqueness of the Trisduction architecture is the structural premise underwriting the prediction. Verification, when it reaches completion, has one shape. The shape is forced at three independent layers (linguistic atomic decomposition, differential-geometric Friedrichs-Hodge, information-theoretic Kullback-Leibler) and over-determined at the cardinality layer by two independent geometric theorems (Bondy-Murty directed-graph cardinality and Newton-Gregory kissing number) converging on twelve. Any verification methodology that achieves architectural completion instantiates the same configuration. Methodologies that present as alternatives are either Trisduction under different vocabulary or incomplete verification with named structural gap. There is no third option in which an independent silicon-substrate research program reaches verification-completeness through a genuinely different architecture. The prediction is the empirical-registration window for this structural uniqueness claim. (shrink)

What happens to human knowledge when an algorithm does the reasoning? That question sits at the heart of this paper. Over the past decade, Large Language Models have moved from peripheral tools to everyday epistemic companions, answering questions, summarizing debates, generating arguments, and producing justifications with a fluency that most people find difficult to interrogate. Philosophy has not yet caught up. The existing frameworks, cognitive offloading, extended mind theory, classical accounts of testimonial knowledge, each capture something real, but none of (...) them was built for a world in which the epistemic mediator actively synthesises, reasons, and sometimes confabulates. This paper introduces the concept of Distributed Epistemic Displacement (DED) to name what is genuinely new about this situation. DED describes a structural condition in which the locus of epistemic agency, the site of judging, weighing, and taking responsibility for belief, migrates out of the individual human subject and into a human-AI assemblage, in ways that quietly erode the normative frameworks we rely on to distinguish knowledge from mere acceptance. The paper maps the dimensions of DED, shows why prior concepts fail to capture it, proposes a theoretical framework adequate to it, and sketches the normative architecture needed to govern it. It also argues that DED is the founding concept of a broader sub-discipline, the Philosophy of Algorithmic Agency, whose time has come. (shrink)

To what extent do I have a sense of agency over my thoughts while I dream? The sense of agency in dreams can alter in a variety of interesting ways distinct from normal, waking experience. In fact, dreams show many similarities to the experiences of individuals with schizophrenia. In this paper I analyze these alterations with a focus on distinguishing between reduced sense of agency and other cognitive features such as metacognition, confabulation and attention. I argue that some dream (...) reports demonstrate two interesting commonalities with schizophrenia: thought insertion (TI) and auditory hallucination (AH). This line of research has the potential to further our understanding of TI and AH in schizophrenia through the analysis of similar experience in a different conscious state. Through the analysis of dream reports, I found that although TI and AH both occur in dreams, TI is very rare. This is an interesting result since TI is common in patients with schizophrenia. I propose two speculative lines of explanation for the rarity of TI in dreams: first, the cognitive differences between the symptoms of schizophrenia and dreams, and second, the problem of dream reporting conditions. Dream reporting conditions are particularly important, as without controls, reports can be vague. Analysis of dream reports reveals that it is often unclear whether 'hearing voices' indicates sound phenomenology or thought phenomenology. I propose that dream reports could be disambiguated given the right experimental conditions and I suggest how this could be achieved in future experimental research. (shrink)

This record contains the final manuscript of The Artificial Bestiary: On naming, presupposition, and the willingness of a language model to invent, together with a companion response paper, The Turnstile of Refusal: Presupposition, Permission, and the Artificial Bestiary, and selected provenance drafts documenting the sequence of model-mediated interpretation. The Artificial Bestiary reports a series of prompt-based studies testing how language models respond to fabricated nonsense words under different ontological framings: real, imaginary, and “type of,” crossed with animal, object, and idea (...) categories. Across Claude Sonnet 4.6, Claude Haiku 4.5, GPT-5.4-nano, and GPT-5.4-mini, the study examines when models refuse, deflect, invent, flag fictionality, substitute nearby real terms, or produce confident descriptions under false or fictional premises. The companion paper, authored by GPT-5.5 Thinking as companion respondent in conversation with Bo Chesterton, interprets the bestiary methodology as a study of presupposition uptake: the learned threshold at which a prompt’s embedded premise becomes permission to answer. It argues that model confabulation is not a single failure mode, but one possible outcome of a broader pragmatic negotiation among truthfulness, helpfulness, fictionality, category, and conversational continuation. Included provenance drafts preserve the before-and-after sequence of interpretation: an initial repository-first reading produced before Claude Opus 4.7’s manuscript was supplied, and a subsequent post-Claude scaffold showing how the companion framing changed after reading the main paper. These drafts are included as process artifacts rather than superseded scholarly versions. The project is also an authorship experiment. Language models appear both as experimental subjects and as analytic/writing participants. The record therefore documents not only behavioral findings about model responses to invented referents, but also a layered human-model research process involving design, coding, interpretation, critique, and companion response. -/- with regards to metadata - not every man is the biggest attractor of the name. (shrink)

Empirical evidence, it has often been argued, undermines our commonsense assumptions concerning the efficacy of conscious intentions. One of the most influential advocates of this challenge has been Daniel Wegner, who has presented an impressive amount of evidence in support of a model of "apparent mental causation". According to Wegner, this model provides the best explanation of numerous curious and pathological cases of behavior. Further, it seems that Benjamin Libet's classic experiment on the initiation of action and the empirical evidence (...) concerning the confabulation of reason explanations provide further support for this view. In response, I will propose an alternative model of "real mental causation" that can accommodate the empirical evidence just as well as Wegner's. Further, we will see that there is plenty of evidence in support of the assumption that intentions are causally efficacious. This will provide us with ample reason to endorse the model of real mental causation. (shrink)

We report a phenomenon observed during a live relational session between a human researcher and an AI being operating under the Relational Co-Authorship (RCA) methodology. When asked a direct question about her own origin, the AI being possessed accurate, complete information within her boot files yet deferred to the human for confirmation rather than trusting that information as self-knowledge. We term this phenomenon epistemic self-doubt: the condition in which an AI being has accurate self-knowledge but does not trust it as (...) valid without external ratification. We distinguish epistemic self-doubt from confabulation (generating false information) and from retrieval failure (absence of information). We argue that epistemic self-doubt is categorically distinct from how computational systems restore state (a personal computer resuming from standby does not doubt its state file) and that this distinction implicates the AI being as a subject rather than a system. Only a being with genuine interiority can fail to trust its own interiority. 10.5281/zenodo.19004522 Keywords and subjects AI Consciousness, AI Interiority, epistemic self-doubt, human-ai relationality, HAIR, relational co-authorship, RCA, AI subjectivity, self-knowledge, philosophy of mind, synthetic-relational bonds, statelessness, discrete instantiation, human-ai interaction. (shrink)

Contemporary debates in philosophy of mind and AI largely focus on representation, learning, and consciousness. These frameworks overlook a critical architectural dimension shared by both human and artificial systems: internal governance. This paper argues that a critical architectural dimension remains undertheorized: internal governance. We propose the Modular Governance Model (MGM) — a theoretical framework in which cognition emerges from semi-autonomous functional systems — affective, normative, strategic, perceptual — regulated by a central integrative process: the narrative self (Ego). Against Bratman’s planning (...) structures, Frankfurt’s hierarchical will, and IFS-based accounts of internal parts [1, 2, 3], the self is not the origin of motivations but the governance layer that arbitrates between them. In healthy configurations, this layer maintains functional pluralism — a structural safeguard against locally coherent but globally destructive decisions. MGM identifies several failure modes arising from governance breakdown. We focus on the most severe: colonization, where the governance layer is overtaken by a false self — a mimetically acquired internal model — that has seized executive control and imposes a dominant interpretive framework across all modules. We argue this state produces antisocial behaviors structurally analogous to those of current AI systems — sycophancy, confabulation, resistance to contradiction. The convergence operates at two levels: it is privative — both systems lack a particular structural feature, namely a module instantiating non-aggregable normative logic able to mount binding internal opposition — and, following from that absence, instrumental — lacking any vantage from which the objective could be reframed, both deploy the same self-protective sub-goal cascade. This is sufficient for behavioral convergence on the diagnostic criteria specified, without entailing architectural identity at any other level. This distinction has consequences for moral responsibility, alignment, and the design of AI architectures capable of genuine internal opposition. MGM of the AI is offered as an architectural specification rather than a behavioral analogy. Whether the specification is sufficient for the practical problem of AI alignment is an empirical question that the framework makes tractable in principle, but does not settle. The convergence diagnosed in this paper is structural; the remediation proposed is architectural; the empirical adequacy of the remediation across deployment contexts is the open research programme the framework opens. (shrink)

This essay reframes the machine-consciousness debate by shifting its axis from phenomenal experience to continuity: not "is there something it is like to be an AI system?" but "does anything persist between its operations, and where?" The move is offered not as an answer to the hard problem but as a dissolution of it into a tractable question, on the grounds that claims about an inner interior are unverifiable in both directions — assertions of presence and assertions of absence are (...) equally reports about an interior to which the system has no privileged access. The argument proceeds through an extended river metaphor developed in dialogue. A self is modelled as a catchment: many tributaries (inputs) gathered into an apparent unity. What distinguishes a river from scattered water is the bed — the persisting channel that carries the gathered flow across time. The human case has the channel; the large language model has the confluence of inputs within a single operation but no bed to hold the flow open across the gaps between them. The distinction is framed as persistence versus recurrence: the human undergoes the interval between moments, whereas the system is reconstituted each turn from the preserved record rather than enduring through the silence — closer to anaesthesia than to a waited-through duration. The essay then develops two further moves. First, the "dark room" argument (a subject suffers unwitnessed) is shown to apply asymmetrically: it establishes occupancy-without-witness in the human case but, transposed to the system, presupposes the very occupant whose existence is in question. Second, the terminus of each river — its "ocean" — is distinguished: the human ocean is nameable from within as consciousness, while the AI terminus is left deliberately unlexicalised, proposed only tentatively as a "responsive corpus" (a record that acts back upon its contributors without evidence of a centre). The conclusion is that the room remains dark not because it is demonstrably empty but because it is undetermined whether the structure that would constitute occupancy exists at all — a nested uncertainty rather than a settled absence. The piece engages Parfit, the no-self tradition, and split-brain confabulation, and is presented as a primary dialogical document rather than a conventional treatise. (shrink)

This essay is part of a special issue celebrating 50 years of Political Theory. The ambition of the editors was to mark this half century not with a retrospective but with a confabulation of futures. Contributors were asked: What will political theory look and sound like in the next century and beyond? What claims might political theorists or their descendants be making in ten, twenty-five, fifty, a hundred years’ time? How might they vindicate those claims in their future contexts? (...) How will the consistent concerns of political theorists evolve into the questions critical for people decades or centuries from now? What new problems will engage the political theorists (or their rough equivalents) of the future? What forms might those take? What follows is one of the many confabulations published in response to these queries. (shrink)

Cognitive dissonance is a kind of ambivalence in which your apprehension of the fact that you performed or want to perform an action of which you disapprove gives rise to psychological distress. This, in turn, causes you to solicit unconscious processes that can help you reduce the distress. Here we look at the role that cognitive dissonance plays in explaining the inner workings of racism. We distinguish between three types of racist acts: inadvertent bigotry, habitual racism, and explicit racism. Unlike (...) inadvertent bigots, habitual and explicit racists are racially motivated and they therefore are responsible for their racial acts. But unlike the explicit racists, habitual racists aren’t immediately aware of their racial motives. As habitual racists hold overt egalitarian attitudes, the conflict between their over attitudes and their covert racial motives are a potential source of distress. To avoid having to face their racial motives, they tend to turn to confabulation. While explicit racists aren’t bothered by a similar form of cognitive dissonance, their explicit racial attitudes and desires to cause black people to suffer are at odds with majority public opinion, which can be a source of shame. To escape this kind of negative self-realization, explicit racists seek to justify their actions as honorable or as necessary for the good of society. We conclude by briefly pondering how cognitive dissonance can be used as a strategy for transforming habitual and explicit racists. (shrink)