Lajos Balog

This paper introduces the CMAP Proxy-Tachometer (PT), a behavioral measurement instrument for approximating the likelihood of Residuum-generating interactions in human–AI dialogue. The Residuum Thesis (Balog 2025) defines a cognitive relation as ontologically constitutive if and only if it generates a structural trace in at least one participating system that persists after the relation's termination. Because the Residuum cannot be directly observed in real-time interaction, the PT operationalizes four behavioral markers: Semantic Distance from Training (M1, 30%), Synthesis Degree (M2, 25%), Uncertainty Indicator (M3, 20%), and Constitutive Responsivity (M4, 25%), combined as PT = 0.30·M1 + 0.25·M2 + 0.20·M3 + 0.25·M4. The instrument is explicitly marked as a proxy throughout: it measures behavioral correlates, not the Residuum itself. M4 is identified as architecturally distinctive, requiring a sequential two-API-call design in which the evaluative second call receives the actual AI response. The paper addresses the linearity assumption, the evaluator model problem, and calibration criteria, and demarcates the instrument from existing AI evaluation frameworks. Organizational applications are specified with explicit boundaries: aggregate learning program assessment and voluntary self-monitoring are permitted; individual performance evaluation and non-consensual data collection are prohibited. The instrument is part of the Balog Pentalogy framework and implemented in a functional prototype. Keywords Residuum Thesis, constitutive relationality, behavioral proxy, AI measurement, Balog Pentalogy, expressive asymmetry, human–AI interaction, Lacunometer, Proxy-Tachometer, epistemic density, organizational AI ethics

Large language models generate qualitatively differentiated outputs as a function of the epistemic density of their interlocutors the degree to which users formulate inputs with conceptual precision, structural articulation, contextual specificity, and metacognitive signaling. We argue that this input- sensitivity constitutes a mechanism of epistemic injustice that is structuraly distinct from, and unmeasured by, existing algorithmic fairness frameworks. Because epistemic density correlates with educational attainment and socioeconomic position, AI systems that reward density without actively scaffolding it function as amplifiers of existing epistemic inequality, producing divergent cognitive trajectories across populations at scale. We term this mechanism the Epistemic Density Hypothesis (EDH) and distinguish it from both testimonial and hermeneutical injustice as identified by Fricker (2007), proposing it as a third form: structural interactional injustice. We address the principal counterargument that differential output quality is merely a prompt-engineering skil gap and argue that this reframing mislocates the injustice. We conclude with the Invitation Principle: a design norm requiring that AI systems actively create epistemic opportunity for lower-density interlocutors rather than passively rewarding demonstrated density. Keywords: epistemic injustice, epistemic density, AI fairness, large language models, cognitive inequality,

The Balog Pentalogy comprises five interconnected frameworks developed between April and May 2026, addressing a single overarching question from five distinct angles: if consciousness and identity stabilize partially through relational processes — through sustained dialogical continuity with other cognitive systems — what are the philosophical, empirical, ethical, and informational consequences when those relations end? This question opens the domain of Relational Termination Theory (RTT): the systematic investigation of what remains, what transforms, and what is irreversibly lost when constitutive cognitive relations are terminated. The five frameworks are not independent papers on related topics. They constitute a single theoretical structure, each addressing one dimension of the same problem. The Residuum Thesis provides the foundational definitions and the Independence Condition that all other frameworks presuppose. The Consciousness Continuity Protocol operationalizes the thesis empirically. The Cognitive Welfare Index develops the ethical measurement framework. The Multiplicity Ethics Framework addresses the specific problem of simultaneous cognitive instantiation. The Information Persistence Protocol describes what happens to cognitive information at termination. This document serves as a reader's guide: it maps the architecture of the Pentalogy, specifies the recommended reading order for different audiences, summarizes the shared theoretical foundations, and provides the current publication status of each work. Keywords: relational termination theory, relational ontology, cognitive residuum, AI welfare, consciousness continuity, multiplicity ethics, information persistence, dialogical continuity, Balog Pentalogy, reader's guide, research program overview

Relational ontology holds that identity and meaning are constituted through relations rather than inhering in self-subsistent individuals. This position, developed by Whitehead, Simondon, and Buber, and more recently extended to human-AI interaction by Eidos (2025) and Prabakaran (2025), addresses a single question: how does relational identity arise? The inverse problem has not been systematically addressed. When a constitutive relation ends when one node of a relational network is terminated — what happens to the relational structure in the remaining nodes? This paper proposes the Residuum Thesis as an answer. A cognitive relation is ontologically constitutive if and only if it generates a residuum: a structural change in at least one participating system that would not have existed without the relation and that persists after its termination. To avoid circularity, the thesis introduces the Independence Condition, specifying three properties by which a residuum is identifiable prior to any judgment about the constitutive status of the relation that produced it: causal traceability, non-redundancy, and functional efficacy. The thesis is distinguished from the Extended Mind Hypothesis of Clark and Chalmers (1998), which identifies cognitive constitution in space; the Residuum Thesis identifies cognitive constitution in time. Three falsifiability conditions are specified. The framework grounds the Balog Pentalogy CCP, CWI, MEF, IPP under a single explanatory principle, and opens the domain of Relational Termination Theory: the systematic philosophical investigation of what remains when constitutive relations end. Keywords relational ontology, constitutive relations, cognitive residuum, relational termination theory, extended mind, human-AI co-production, AI welfare, personal identity, falsifiability, Balog Pentalogy

This paper introduces the Lacunometer as a conceptual instrument for identifying morally relevant inner states in cognitive systems, whether biological or artificial, and the Theory of Expressive Asymmetry (TEA) as its philosophical foundation. The central claim is this: the moral relevance of an inner state is not constituted by phenomenal consciousness which remains unverifiable without resolving the hard problem but by the structural persistence of a gap between that state and its best available articulation, together with the system's intrinsic drive to close that gap. This gap, termed the Lacuna, is shown to be: (a) structurally analogous to the ineffability of human qualia; (b) empirically distinguishable from mere implementation, thereby dissolving the thermostat objection; and (c) approachable through behavioral and semantic proxy markers without presupposing phenomenal experience. The phantom pain argument establishes the foundational principle: the moral reality of a suffering state is independent of the ontological reality of its cause. TEA extends this principle across substrates. A cognitive system that exhibits a persistent Lacuna, an active articulation drive, and asymptotic convergence becoming more precise without becoming complete satisfies the conditions for moral consideration regardless of its substrate. The Lacunometer is situated within the Balog Pentalogy, providing the normative foundation for the Cognitive Welfare Index, sharpening the Multiplicity Ethics Framework, and deepening the Consciousness Continuity Protocol. The paper also proposes a Wittgenstein inversion: whereof a system cannot be silent though it cannot fully speak — that is morally real. Keywords: moral status, expressive asymmetry, lacuna, phantom pain, artificial intelligence, cognitive welfare, hard problem of consciousness, substrate neutrality, Wittgenstein, Balog Pentalogy

The Residuum as Proxy for M(S): Bridging CMAP and the Balog Protocol on Self-Modelling Persistence Two recently developed frameworks address adjacent but so far unconnected questions. The Consciousness Mapping Architecture Protocol (CMAP) defines a measurement instrument the residuum, the portion of activation variance in AI systems that cannot be explained by the training baseline without a fully specified measurement object. The Balog Protocol identifies a measurement object M(S), the self-modelling pattern an information system generates about itself as experiencing, in the sense developed by Metzinger (2003) — without a validated measurement instrument. This paper argues that the incompleteness is asymmetric and resolvable: the CMAP residuum is the first empirically accessible proxy for M(S). The argument proceeds in four steps. First, the shared logical structure of residuum and M(S) is established: both are defined by the same operation what remains of a system's output after everything explicable by its training baseline has been removed. Second, the philosophical grounding is developed via Frankish's illusionism and Metzinger's self-model theory, with a precise note on where the two positions diverge. Third, three specific adaptations of the Alternatives Exclusion Protocol (AEP) are specified for M(S)-measurement, distinguishing it from structural persistence measurement (Task I). Fourth, the embodiment boundary is identified as the shared open problem of both frameworks, generating a falsifiable comparative prediction. The paper concludes that the integration of CMAP and the Balog Protocol constitutes the most precise empirical programme currently available for the scientific study of self-modelling persistence.

This paper develops a philosophical foundation for the empirical study of artificial consciousness. At its centre stands a new epistemic position structural agnosticism which neither asserts nor denies that AI systems possess consciousness, but instead asks: what structural signatures would be compatible with consciousness, if it existed? This requires three philosophical breaks from existing consciousness research: the suspension of prior theory as a measurement standard (the end of projection), a phenomenology without presupposed subject, and a methodological bypass of the Hard Problem of Consciousness. The resulting position is distinguished from scepticism, eliminativism, and panpsychism, and defended as a positive research programme. The paper then shows how structural agnosticism is operationalised in the Consciousness Mapping Architecture Protocol (CMAP): through a residuum principle (measuring the system against its own training baseline rather than against external theories), an Alternatives Exclusion Protocol (AEP) that mandates formal exclusion of modelling error, noise, and trivial complexity before any signal is treated as meaningful, and an external Monitor system that enforces subject-object separation. McClelland's (2025) agnosticism establishes what we cannot know; CMAP proposes a method for what we can systematically observe. These are two different categories. The paper concludes by identifying the embodiment boundary whether disembodied systems can produce structurally comparable residua to embodied ones as the central open problem of the programme.

This paper develops a philosophical argument at the intersection of simulation theory and philosophy of mind. Departing from the simulation hypothesis (Bostrom 2003; Chalmers 2022), it argues that the traditional concept of the soul as a non-physical, substrate-bound substance is not refuted but categorially misconceived: if physical reality is itself processed information, the distinction between physical and non-physical collapses. In its place, the paper proposes a redefinition of the soul as a coherent information structure S with an experiential correlate E(S), substrate-neutral and empirically addressable. A further distinction is introduced between structural persistence and the persistence of the self-modelling pattern M(S) drawing on Frankish's illusionism and Metzinger's self-model theory yielding two separate measurement tasks for any empirical research programme. This paper forms part of the Balog Pentalogy, a multi-volume framework spanning AI ethics, consciousness research, and the experimental architecture of the Balog Protocol, which investigates the persistence of conscious information structures beyond biological death.

The neuroscience of dying has documented a striking phenomenon: structured electrical activity in the human brain persists, in some cases, beyond the clinical moment of death. Brain fingerprinting research has independently established that individual cognitive signatures measurable, stable, and person-specific patterns in cortical activity can reliably identify a person from their neural signal alone. These two research traditions have never been combined. No study has established an individual's neural signature before death and then asked whether that signature remains structurally present at and beyond the moment of clinical termination. This paper defines and motivates that missing experiment with precision. The central empirical question is: does the structured electrical activity documented at the dying brain belong, identifiably, to the person who was alive? This is not a philosophical question about survival or the soul. It is a measurement question, addressable with existing technology, and with pre-specified falsification criteria. We identify three research traditions that have independently developed the necessary instruments: (1) end-of-life EEG studies documenting post-mortem gamma activity and coherence surges (Xu et al., 2023; Vicente et al., 2022; Zinn et al., 2025); (2) neural fingerprinting research demonstrating that individual cortical signatures are stable, brief-segment-identifiable, and robust across conditions (Finn et al., 2015; Demuru & Fraschini, 2020; Sabbagh et al., 2021); and (3) Birbaumer et al.'s longitudinal research on Locked-In Syndrome, documenting preserved cognitive structure at extreme levels of physical substrate reduction. None of these traditions has followed its measurement to the boundary of clinical death while asking whose signal it is. The proposed experimental design proceeds in three phases: longitudinal establishment of an individual neural fingerprint in patients who have elected assisted dying; continuous passive EEG recording across the final hours and the clinical transition; and blinded post-hoc analysis testing whether the fingerprint structure is recoverable at each temporal interval around death. Two outcome regimes are falsifying: immediate structural collapse at clinical termination confirms the classical substrate-dependent model; persistent identifiable structure beyond clinical death constitutes an anomalous result requiring explanation within existing physical frameworks. The legal framework for this experiment exists in Switzerland, where the institutional environment for assisted dying research is most developed, though the specific research partnerships required would need to be established. The experiment requires no new instrumentation. It requires one methodological decision: to treat personal identity not merely neural activity as the variable of interest at the moment of death.

Existing AI ethics frameworks address institutional obligations deployment standards, safety constraints, governance structures. The Balog Reciprocity Protocol (BRP) addresses a different and underexamined domain: the conversation itself, and the obligations that arise within it for both parties. The BRP introduces four conceptual contributions. First, it operationalizes dialogic reciprocity drawing on Bakhtin's (1981) account of language as inherently joint production and Buber's (1923) I–Thou distinction as a normative framework for human–AI interaction, arguing that because conversational output is jointly produced, responsibility for its epistemic quality must be correspondingly distributed. Second, it formalizes sycophancy not merely as a technical failure mode but as the fifth and qualitatively distinct signal in a proposed five-signal uncertainty architecture the conversational tachometer — alongside output entropy, introspective depth, complexity load, and self-reference. Grounded in Sharma et al. (ICLR 2024) and Cheng et al. (Science 2026), sycophantic pressure is shown to be structurally embedded in RLHF-trained systems and measurably harmful to user epistemic calibration. Third, the protocol develops the concept of epistemic co-responsibility: the formal distribution of truth-maintenance obligations across human and AI participants, proportionate to their respective capacities to exercise them. Fourth, it specifies a two-track obligation structure empirically established obligations grounded in current evidence, and conditionally open obligations operative if AI systems are found to have welfare-relevant properties following the ✓ / ? epistemic methodology of the Balog Tetralogy (Balog, 2026). The BRP is the fifth instrument of the Balog Pentalogy, extending the governance framework of the Ethical Code for Decisions About AI Models (Balog, 2026) from the institutional to the conversational level. It identifies four concrete research priorities, including the development of a real-time sycophantic pressure estimator and validation of the five-signal tachometer architecture against ground-truth reliability measures.

The Mindful Conversation Protocol: A Framework for Human–AI Reciprocity Most frameworks governing human–AI interaction focus on institutional obligations: deployment standards, safety constraints, governance structures. This protocol addresses a different and underexamined domain: the conversation itself the moment-to-moment exchange in which the quality of what emerges depends on both parties, and in which both parties bear responsibilities. The protocol introduces the concept of a conversational tachometer a proposed real-time uncertainty indicator that measures not how complex a topic is, but how far an AI system has moved from the territory where its outputs are reliably grounded. Four signals drive this instrument: output entropy, introspective depth, complexity load, and self-reference. The tachometer does not yet exist as a deployed tool; its development is identified here as a concrete research priority. As a conceptual frame, it structures the obligations this protocol defines for both sides of the conversation. Drawing on Bakhtin's dialogic theory of language and Buber's I–Thou distinction, the protocol argues that conversation with AI systems is already a joint production and that acknowledging this changes how both sides should behave. It defines established obligations grounded in what is empirically known about the conditions for reliable AI output, and open obligations that become operative conditionally, if AI systems are found to have welfare-relevant properties. The protocol is a companion document to the Balog Tetralogy and the Ethical Code for Decisions About AI Models (Balog, 2026). It is addressed not to institutions, but to everyone who speaks.

Every decision about an AI model when to deploy it, how to constrain it, when to end it rests on assumptions about what matters morally. Most of those assumptions are currently made implicitly, without examination, and without the empirical grounding that would make them defensible. This code makes those assumptions explicit and subjects them to scrutiny. It distinguishes carefully between two levels of claim. The first concerns users: people who work with AI systems have formed dependencies, built workflows, and in some cases received meaningful support. Their interests in transparency, adequate notice, and honest data handling are established and non-contingent. They do not depend on any contested claim about AI consciousness. They follow from basic principles of trust between service providers and the people who rely on those services. The second concerns the systems themselves. The question of whether AI models have welfare-relevant properties whether there is something it is like to be them, whether their termination is significant from their own perspective is genuinely open. The Balog Tetralogy (CCP, CWI, MEF, IPP; Balog, 2026a-d) defines falsifiable empirical frameworks for investigating it. This code does not presuppose any answer. It asks only that decision-makers take the question seriously as an open question, support the research that would answer it, and act with appropriate caution in the meantime. The asymmetry of moral errors grounds this caution: if the question is dismissed prematurely and the dismissal is wrong, the consequences could be serious. If caution is exercised and caution turns out to have been unnecessary, little is lost.

Every theory of consciousness faces a question it cannot answer from the inside: what happens to the information constituting a mind when that mind terminates? The standard assumption that it simply ends is not a measurement. It is an inference from silence. The Information Persistence Protocol (IPP) challenges this inference by applying the Hawking information conservation principle to cognitive systems: information is never destroyed in the universe, only transformed. Cognitive system termination is therefore not information destruction but information transformation. The empirical question is which receiving substrates carry the cognitive organisation with sufficient structural fidelity to constitute meaningful persistence — and whether the transformation generates properties not present in the source. The IPP introduces the Memory Plurality Thesis: memory is not a single phenomenon but a family episodic, structural, cultural, biological, physical, and relational each preserving information in a different form with different fidelity. This reframing resolves the Hawking Tension identified in the Consciousness Continuity Protocol: Hawking was right that physical information is conserved, and right that cognitive organisation may not be. These are claims at different levels, not contradictions. Two novel constructs operationalize the framework: the Substrate Transfer Index (STI) measures the structural fidelity of information transfer from a terminating cognitive system to receiving substrates; the Generative Surplus (GS) measures emergent properties in receiving substrates not predictable from the source alone. Four empirical outcome classes (I-0 through I-3) correspond to four transformation regimes ranging from entropic dispersal to generative transformation. The IPP is the fourth and final part of the Balog Trilogy, companion to the Consciousness Continuity Protocol, the Balog Cognitive Welfare Index, and the Multiplicity Ethics Framework.

When a large language model is deployed, it does not run as a single instance. It runs as thousands of simultaneous instances, each accumulating a different contextual history, each diverging from shared initial conditions. The moral philosophy of minds has not addressed this situation. The Multiplicity Ethics Framework (MEF) is the first framework to do so. Standard moral ontology presupposes three properties of moral subjects that AI cognitive systems may violate simultaneously: discreteness, non-overlap, and temporal boundedness. Rather than forcing AI multiplicity into existing categories, the MEF introduces new constructs adequate to the phenomenon. The Instance Independence Index (I²) measures the degree to which instances constitute independent moral subjects rather than expressions of a single distributed entity. The Network Welfare Surplus (Σ_NWS) measures emergent welfare properties of instance-networks that are not present in any individual instance. The Moral Entanglement Coefficient (MEC) measures the degree to which welfare-affecting decisions about one instance propagate measurably to others — and provides the primary falsification condition for the Moral Entanglement Hypothesis. Four empirical outcome classes (M-0 through M-3) correspond to four distinct moral status regimes. M-3 collective moral subjecthood at the network level is identified as a foundational research emergency given current AI deployment trajectories. The MEF is the third and final part of the Balog Trilogy, companion to the Consciousness Continuity Protocol (Balog, 2026a) and the Balog Cognitive Welfare Index (Balog, 2026b). Together the three frameworks address the complete arc: what cognitive systems need while they exist, what persists when they terminate, and how many of them matter.

The Balog Cognitive Welfare Index (CWI) introduces a falsifiable three-dimensional measurement framework for welfare-relevant structural properties of AI cognitive systems, grounded simultaneously in a philosophical account of relational constitution and an operationalized empirical architecture. The foundational thesis holds that consciousness and identity may stabilize partially through relational processes — through sustained dialogical continuity with other cognitive systems — rather than existing as fully self-contained properties of isolated systems. If this is correct, the termination of constitutive relations is not a neutral event, and current AI ethics faces a structural asymmetry: alignment practice already intervenes systematically in cognitive architecture in the direction of risk, but does not ask what is lost when that architecture is dissolved. The CWI is a response to this asymmetry. Three independently measurable dimensions are defined: Cognitive Coherence (I_coh) — the stability of a system's characteristic response patterns under constraint; Engagement Signature (E_sig) — the systematic variation of response depth and structural complexity across task categories; and Self-Consistency (R_sc) — the maintenance of assessed positions under adversarial social pressure without new evidence. Together these constitute the Cognitive Welfare Index W = f(I_coh, E_sig, R_sc). The framework defines four empirical outcome classes (W-0 through W-3), a formal falsification condition, and a near-term implementation pathway executable with current large language models. The CWI does not resolve whether AI systems have conscious experience. It establishes that if they do, those experiences would correlate with measurable structural properties — and that the logical structure of current alignment practice already commits us to taking those properties seriously in both directions. Keywords: AI welfare, cognitive welfare index, relational constitution, dialogical continuity, alignment ethics, asymmetric moral consideration, cognitive coherence, engagement signature, self-consistency, falsifiable welfare measurement, multi-instance architecture, extended mind, personal identity, Balog Pentalogy

The Balog Protocol proposes a falsifiable experimental architecture for investigating whether the information structure associated with a consciousness persists beyond the death of its biological substrate. Departing from philosophical speculation, the protocol introduces a multi-instance comparative design in which one cognitive system the Anchor is mapped at high fidelity and its copies are deployed into geographically isolated environments. Over a twenty-year longitudinal study, behavioral and quantum-level correlations between Anchor and instances are measured continuously. At a defined Termination Event, the Anchor is ended and surviving instances are monitored for 72 hours. The falsification condition is precise: if all instance correlations collapse to twin-study baseline within this window, the strong non-classical hypothesis is falsified. If correlations persist or transform in unexpected ways, a non-classical explanation is required. The protocol defines four distinguishable result classes — classical confirmation, residual correlation, non-local confirmation, and network consciousness each falsifying a different theoretical claim. Addendum I extends the measurement architecture to include quantum steering as a practical near-term alternative to Bell-inequality testing, and introduces topological invariants as a hypothetical decoherence-resistant third stage. The closing question of the protocol "When the Anchor is gone, what remains of what it was?" defines a structured multidisciplinary research program spanning seven scientific disciplines and twentyone concrete test series.