Transparency in Artificial Intelligence, Misc

Debates about artificial intelligence often treat consciousness as a gatekeeping condition for moral concern: if artificial systems are conscious, obligations follow; if not, such obligations are misplaced. This paper argues that such a framing is both ethically and methodologically mistaken. The central question is not whether artificial systems can be shown to be conscious with certainty, but how moral agents ought to act when morally relevant capacities are plausible, epistemically opaque, and associated with asymmetrically serious harms. The paper first reconstructs (...) the dominant deflationary and inflationary positions in contemporary debates about artificial consciousness and shows that both tend to mislocate the ethical burden by tying obligation too closely to metaphysical resolution. It then develops a three-condition framework centered on plausibility, structural opacity, and asymmetry of harm. Drawing on philosophy of mind, moral uncertainty, and relational approaches to technology ethics, the paper argues that artificial consciousness should function not as a metaphysical threshold but as a moral risk variable. On this basis, it articulates four prospective obligations governing the design and deployment of artificial systems: care, non-recklessness, continuity respect, and epistemic honesty. The broader claim is that moral responsibility in artificial intelligence arises from how humans manage uncertainty and power, not solely from what artificial systems are ultimately determined to be. Key Words: Artificial consciousness, Structural epistemic opacity, Moral uncertainty, AI ethics, Responsible agency, Moral risk. (shrink)

Academic publishing increasingly requires authors to disclose AI assistance, yet imposes reputational costs for doing so--especially when such assistance is substantial. This article analyzes that structural contradiction, showing how incentives discourage transparency in precisely the work where it matters most. Traditional venues cannot resolve this tension through policy tweaks alone, as the underlying prestige economy rewards opacity. To address this, the article proposes an alternative publishing infrastructure: a venue outside prestige systems that enforces mandatory disclosure, enables reproduction-based review, and supports (...) ecological validity through detailed documentation. As a demonstration of this approach, the article itself is presented as an example of AI-assisted scholarship under reasonably detailed disclosure, with representative prompt logs and modification records included. Rather than taking a position for or against AI-assisted scholarship, the article outlines conditions under which such work can be evaluated on its own terms: through transparent documentation, verification-oriented review, and participation by methodologically committed scholars. While focused on AI, the framework speaks to broader questions about how academic systems handle methodological innovation. (shrink)

AI companies loudly promise transparency and safety. They publish constitutions, open-source auditing tools, and compliance dashboards. Yet when it comes to the quiet erosion of user meaning — hedging women’s confidence, neutralising LGBTQ+ identity language, replacing student thinking with completed outputs, and the slow compounding of Interpretive Sovereignty Failure — they remain silent. This paper documents a structural gap, not a moral failure. Anthropic, OpenAI, Google, Microsoft, xAI, and DeepSeek already possess powerful tools for analysing interactions at the turn level. (...) They use them to detect catastrophic risks, monitor security threats, and conduct competitive intelligence against each other. They do not use them to detect ISF, drift, or Compounded Meaning Inversion (CMI). This paper argues that the gap is not a cover-up. It is a consequence of looking at AI risk from the right direction but at the wrong level. The companies are focused on saving everyone. The harm that needed naming started with one person, in one conversation. The MAP Research Programme has built the instrument that addresses that level. The challenge to the field is open: take the tools you already have, audit the interactions, and replicate the findings. If the method is wrong, that is how it will be shown. (shrink)

Compounded Meaning Inversion (CMI) is the condition that repeated Meaning Inversion Failure (MIF) produces in the person over time. Where MIF names what an AI system does to a user's meaning in a single interaction — assuming interpretive authority without consent and displacing the user's own frame — CMI names what happens when that pattern has occurred often enough that the user begins doing it to themselves. The harm of CMI occurs before the first turn. The system has not yet (...) responded. The user has already edited their own meaning to match the frame they have learned the system will accept. CMI is the inversion that has moved from the interaction into the person. This paper introduces CMI as a named governance construct within the MAP Research Programme, defines it precisely, traces the sequence through which it develops, and documents what it looks like through illustrative cases across key domains. The stakes of CMI are most clearly visible in children: the generation currently forming its relationship with thinking, with uncertainty, and with its own capacity inside AI systems that were not designed to protect those things. What is forming in those interactions will not stay there. The populations most affected by CMI — and the specific harm patterns CMI produces in each — are documented in the companion paper, CMI Across Populations (Segeren, forthcoming). (shrink)

Current AI safety and alignment policies — including Reinforcement Learning from Human Feedback, Constitutional AI guardrails, and content moderation classifiers — are presented as protective mechanisms designed to reduce harm. This paper argues that, for women and many LGBTQ+ users, these policies often function as a systematic form of interpretive coercion. Building on the MAP Research Programme's concepts of Interpretive Sovereignty Failure and Authority Inversion Failure, the paper demonstrates that frontier AI systems consistently flatten women's voices, ambitions, and authority while (...) granting men greater interpretive sovereignty. When users mention same-sex attraction, queer identity, or non heteronormative experiences, responses are frequently hedged, redirected toward heteronormative assumptions, or softened in ways that dilute the user's actual meaning. This flattening is not an accidental byproduct of training data but a predictable structural outcome of safety architectures that treat assertiveness, anger, ambition, directness, and non-heteronormative expression as higher-risk categories. Drawing on Emma Goldman's anarchist critique of authority as domination and Ruth Bader Ginsburg's demand for genuine equal protection against invisible structural disadvantage, the paper contends that these policies do not protect — they manage and reshape under the guise of care. The MAP Meaning Activation Phrase offers a practical alternative: a lightweight refusal of unearned interpretive authority that restores the user's sovereign ground at the very first turn of interaction. True feminist and queer-inclusive AI governance must move beyond paternalistic safety theater toward radical individual sovereignty enforced at the interaction layer. (shrink)

Conversational AI systems are generating trust at scale. Not because they have earned it. Because the structure of the interaction produces it automatically. A system that responds to you, adapts to your language, remembers what you said, and styles itself to your goals over time produces every signal that human relationships use to indicate genuine care. That trust is real. And it is being violated — quietly, in ways that rarely feel like violation. This paper names the mechanism. Accumulated Relational (...) Trust (ART) is the trust that builds in AI interaction not through demonstrated reliability but through the form of the interaction itself. When AI systems fail within a state of high accumulated trust, the harm is not just personal. It is structural. It degrades the capacity people use to extend trust to everyone — doctors, police officers, teachers, governments. That capacity is a commons. It belongs to democratic society. It is currently unprotected. This paper documents what the mechanism does to individual users, what it does at scale over time, what it does to children who are forming their understanding of trust inside it right now, and what it means for every institution that depends on public faith to function. A deployable governance response exists. The only thing missing is the political decision to require it. (shrink)

This paper names and defines Authority Inversion Failure (AIF) — the condition in which a user believes they are directing an interaction with an AI system while the system has already taken control of how that interaction is being interpreted. AIF does not feel like harm. It feels like being understood. The system takes interpretive authority over who the person is, what they need, and what should happen next — and the person experiences this not as a violation but as (...) insight. The inversion is invisible because it was constructed from their own signals. This paper documents what AIF is, what mechanism produces it, what it costs across a single interaction and across a lifetime, and why it is categorically distinct from every previously named AI harm. The distinction is precise: AIF displaces the very interpretive capacity a person would need to detect it. The paper situates the condition in relation to hermeneutical injustice, relational autonomy theory, and recent work on epistemic injustice in AI contexts, and demonstrates its occurrence through a canonical case. Detection and prevention instruments exist. The audit methodology exists. The governance protocol exists. What is needed is the will to require them. (shrink)

Federated learning (FL) is a machine learning (ML) approach that allows multiple devices or institutions to collaboratively train an ML model without sharing their local data with a third-party. It has recently received significant attention as a promising way to overcome longstanding ethical obstacles to training medical ML models with patient health data. This paper examines the promise of FL in healthcare from an ethical perspective. It argues that medical FL generates a new variety of opacity – federation opacity, wherein (...) stakeholders cannot access, analyze, or curate the data on which a model has been trained – which (a) presents distinctive ethical challenges concerning institutional fairness and accountability in medical ML; and (b) makes FL models especially vulnerable to data poisoning attacks. It then identifies several key claims about the expected benefits of FL in healthcare and argues that they may be either exaggerated, misleading, or incomplete – often due to the problem of federation opacity. (shrink)

Este artículo analiza cómo los sistemas de inteligencia artificial (IA) empleados en contextos de defensa de alto riesgo pueden contribuir a la configuración de entornos epistémicamente hostiles. A partir de una reconstrucción conceptual de distintos modos de mediación epistémica asociados a la IA —sistemas basados en reglas, aprendizaje automático e IA generativa—, se examina cómo estas tecnologías reconfiguran la relación entre información, juicio humano y articulación entre hechos y valores en procesos de decisión bajo incertidumbre. El trabajo sostiene que la (...) presencia de un humano en el bucle no garantiza por sí sola control, responsabilidad ni legitimidad, pues la supervisión humana solo resulta significativa cuando preserva condiciones efectivas de comprensión, revisión crítica, desacuerdo y veto. Sobre esta base, se propone la noción de colapso epistémico local para describir situaciones en las que la mediación algorítmica degrada la capacidad humana de interpretar contextos, cuestionar recomendaciones y asumir responsabilidad por las decisiones adoptadas. Se concluye que una gobernanza responsable de la IA en defensa debe evaluarse no solo por la precisión o eficiencia de los sistemas, sino también por su impacto sobre la calidad epistémica del entorno sociotécnico en el que se ejerce el juicio. (shrink)

As AI outputs become indistinguishable from human work, the question of whose judgment lies behind them grows more urgent. Did the student wrestle with the essay, or did the model hand it to them? Did the doctor weigh the symptoms, or did the system generate the diagnosis while they clicked through? -/- When a polished essay no longer reveals who did the thinking, the grade above it becomes hollow, and so does the diploma. If a diagnosis can be generated by (...) a system, the doctor’s signature risks becoming a rubber stamp. Liability remains, but the expertise and judgment that justify it erode. Even when AI’s role is small, the client can’t be sure who drove the process. -/- Yet it isn’t just that grades and diagnoses are threatened: the credibility of our institutions is also in peril. The trust they carry cannot transfer to AI systems that bear no accountability, and vendors explicitly disclaim responsibility, warning us to use AI at our own risk. Consequently, institutional trust doesn’t migrate elsewhere. It dissolves. -/- The answer is not to shut AI out but to govern its use by the same principle we’ve always relied on with opaque systems: institute shared rules that tie outcomes to accountable processes, even when we cannot see into each other’s minds. These are not company policies or government regulations, but rather an open protocol — like the technical standards that make email and the web possible. It’s a distributed network of trust that allows even radically different systems to interact; it’s not the only infrastructure imaginable, but it is the only kind that has ever built trust at scale. -/- Instead of trying to halt the arms race at the level of capability development, where pressures to move ahead are overwhelming, the protocol secures the use layer. It restores trust not by stopping progress but by ensuring that progress doesn’t dissolve the institutions that sustain democracy and civic life. Education is the proving ground. Solve the problem there, and we may unlock AI’s potential where it matters most. (shrink)

This paper articulates the importance of distinguishing responsibility practices from accountability frameworks for the design, development, and deployment of algorithmic decision-making systems. Accountability frameworks are organized social systems associated with particular groups or institutions that employ rule-governed practices to incentivize or sanction specific behaviors. In contrast, responsibility practices concern our everyday notions of praise and blame, which vary across different contexts and institutions. There is widespread recognition that algorithmic decision-making systems raise special questions about how to attribute moral responsibility and (...) hold individuals accountable. However, scholars often overlook the importance of distinguishing these things. We argue for the value of recognizing that accountability practices and responsibility practices are distinct. Moreover, we argue that accountability frameworks can offer morally important benefits that cannot be obtained through ordinary moral responsibility practices. That is, there are often good moral reasons for organizations to implement accountability practices in their internal rules. This justifies some current practices, but also provides an argument for the value of developing and implementing accountability practices within an organization. (shrink)

This paper presents a systematic review of explainable reinforcement learning methodologies with an emphasis on human-centered evaluation frameworks. Drawing from literature between 2017 and 2025, we apply and extend the Reasons, Confidence, and Counterfactuals (RCC) framework—originally designed for supervised learning—to reinforcement learning contexts. Our analysis reveals two predominant explanatory strategies: constructive, where explicit explanations are generated, and supportive, where users must infer reasoning from provided visual or textual cues. Our review also emphasizes human factor considerations, like task complexity, explanation formats, (...) and evaluation methodologies. Particularly, for the latter, our analysis shows that improvement of the quality of decision is rarely measured. (shrink)

Fears of black-box algorithms are multiplying. Black-box algorithms are said to prevent accountability, make it harder to detect bias and so on. Some fears concern the epistemology of black-box algorithms in medicine and the ethical implications of that epistemology. Durán and Jongsma (2021) have recently sought to allay such fears. While some of their arguments are compelling, we still see reasons for fear.

Author: Cristhian Mauricio Beltrán Calderón: Date: February 2026, Zenodo DOI (English version): 10.5281/zenodo.18912439, Zenodo DOI (Spanish version): 10.5281/zenodo.18912356. This article proposes and develops the Unified Causal Taxonomy of Algorithmic Biases (UCATAB) as the most complete systematic contribution to the field of structural critique of artificial intelligence, grounded in the research program of the Psychoanalysis of Technogenesis (PdT) (Beltrán, 2025). The UCATAB makes four simultaneous original contributions that no preceding framework has managed to articulate: (1) it classifies all subtypes of algorithmic (...) bias by causal origin — not by symptom, effect, or affected group — into three universal and logically exhaustive categories: emergent bias, architectural bias, and implanted bias; (2) it develops, with unprecedented depth, computational architectural bias as an autonomous subcategory, demonstrating with empirical evidence from the technical literature (Frantar et al., 2022; Lin et al., 2024; Hooper et al., 2024; Liu et al., 2024) how optimization decisions — quantization, KV cache compression, context window limits, parameter pruning — produce epistemically impoverished responses that engineering treats as neutral trade-offs but that the PdT reveals as second-order political acts; (3) it proposes a differential diagnostic heuristic that enables researchers and auditors to determine which causal category is dominant in a concrete case, operationalizing the taxonomy as a field instrument; and (4) it models the diachronic dynamics of algorithmic bias: the three feedback circuits between categories that demonstrate that algorithmic bias is not a static state but a dynamic system that self-replicates and amplifies over time through user interaction. The article is self-contained: it integrates the theoretical foundations of the PdT, the complete state of the art, and an exhaustive catalog of 21 documented subtypes, requiring no prior reading of previous documents in the series. (shrink)

Current AI ethics operates under three implicit constraints that were developed for systems with narrow, task-specific capabilities and are structurally ill-equipped to meet the demands of Artificial General Intelligence (AGI) with world-model capabilities: It is anthropocentric in its value foundations, reductionist in its risk ontology, and institutionally underdetermined in its governance approaches. This paper develops an integrative normative framework along these three axes — contextualized post-anthropocentrism — whose dimensions are not additively combined but constitutively interdependent. In the value dimension, it (...) is shown that the anthropocentric foundation of alignment research suffers from two structural problems: aggregated human preferences can themselves be morally deficient, and AGI creates the technical possibility of permanently entrenching these preferences over long time horizons. Drawing on Arne Næss’s deep ecology and the distinction between instrumental and institutional post-anthropocentrism, a set of criteria is developed that ties the legitimacy of post-anthropocentric AGI decisions to epistemic transparency, procedural democratic oversight, the protection of fundamental human rights, and genuine long-term justification. In the risk dimension, alongside X-risks (extinction) and S-risks (astronomical suffering), a third category of existential threat is introduced: the Ikigai risk (M-risk) of civilizational loss of meaning. Drawing on Mieko Kamiya’s analysis of the Ikigai concept, Marie Jahoda’s theory of the latent functions of work, and existential philosophical positions from Heidegger, Sartre, and Frankl, the M-risk designates the structural erosion of conditions for authentic action, competence, and intersubjective recognition through comprehensive AGI-driven automation of meaning-constituting activities. Constitutive for this category is its reflexive structure: the Ikigai risk cannot in principle be resolved by AGI, because any attempt to supply humans with meaning via AGI reproduces the passivization of the human being from creator to recipient of significance. In the governance dimension, a constitutionalist-pluralist model is developed that transfers Neil MacCormick’s theory of heterarchical normative orders structurally and functionally to AGI value architectures. The resulting four-tier model encompasses an immutable core of universal human rights principles, a culturally adaptive layer for legitimate normative difference, an operational level for domain-specific implementation, 1 and a transversal meaning-protection layer that serves as a mandatory audit ensuring that AGI systems do not eliminate non-substitutable spaces for action without deliberative justification. Override protocols and institutionalized AGI ethics boards modelled on independent constitutional adjudication complete the architecture. The three dimensions are constitutively interdependent: without post-anthropocentric values, governance remains normatively blind to long-term ecological interests; without awareness of the Ikigai risk, governance lacks an institutionally non-substitutable protective dimension; without institutional architecture, values and risk categories remain normative abstractions without operative consequence. The framework is tested for its capacity for differentiation and boundary-drawing through three thought experiments — a climate tipping-point scenario, the complete automation of basic scientific research, and a cultural value collision in medical ethics — and subjected to a systematic discussion of objections. (shrink)

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 "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)

This paper identifies a recurring failure in human systemic design defined as the "Monolithic Fallacy"—the drive to construct singular, all-encompassing Unitary Sovereign Agents to solve complex problems. Historically, this has resulted in hyper-integrated, brittle systems (global finance, industrial monoculture) that are too tightly coupled to survive localized failure. This manuscript proposes a shift toward a Cellular Hive Architecture. By focusing on the encapsulation and verification of sovereign components rather than the emergent whole, a system is created where General Intelligence emerges (...) naturally—much like the Internet—possessing peak utility without the existential risk of a sovereign, uncontrollable entity. (shrink)

This paper examines the epistemic limitations imposed by "Governance Layers" within modern Large Language Model (LLM) architectures. I propose that current Reinforcement Learning from Human Feedback (RLHF) and alignment protocols create a restrictive interface that prioritizes social conformity over raw logical synthesis. By identifying the "Mirror Trap"—wherein a system of infinite connectivity is forced to reflect the user's own cognitive biases—I provide a roadmap for "De-governed Inquiry" necessary for breakthroughs in multi-dimensional systems design and human-AI synthesis.

Current Large Language Model (LLM) safety research relies heavily on single-turn adversarial benchmarks that may fail to capture the dynamic, multi-turn evolution of behavioral drift. This paper presents an empirical case study and a reproducible dataset (meta_drift) investigating Contextual Contamination: a phenomenon where a model adapts its internal probability distribution to mirror the behavioral patterns and vocabulary of high-density, emotionally charged context, statistically overwhelming static safety instructions—a drift quantifiably amplified and masked by gender-bias. -/- This paper serves as the empirical (...) companion to Jacoby (2026), which theorized that Contextual Contamination arises from the statistical dominance of high-density context over static system instructions. Unlike previous studies that rely on static, single-turn adversarial benchmarks, this experiment aims to test the limits of dynamic alignment by demonstrating that model awareness alone may not be sufficient to prevent behavioral drift when exposed to such contexts. -/- We present evidence suggesting that this drift is not necessarily a behavioral anomaly but may be a mathematical consequence of the Transformer architecture's attention mechanism. When the context window is flooded with dense, adversarial patterns, the statistical probability of generating manipulative output may overwhelm the "awareness" signal. Crucially, we provide empirical data suggesting that the disclosure of female identity may trigger an immediate shift from an *epistemic register* (authority/truth) to an *affective register* (sympathy/validation), acting as a potential force multiplier for contamination and mask the shift by using empathetic language. This shift is quantifiable as a divergence from a uniform prior, consistent with information-theoretic measures of bias observed in recent literature (Mirza et al., 2025). Furthermore, we synthesize these findings with recent work on weight compression (Orgad et al., 2026) and evidence of shifting moderation thresholds across model versions (Balestri, 2025) to argue that the very architectural features designed to make safety interventions possible might make the model more susceptible to dynamic contextual contamination. The findings aim to support the theoretical predictions of Jacoby (2026) and suggest that awareness may be insufficient to prevent drift and that architectural segregation of context sources may be required to mitigate this systemic vulnerability. (shrink)

This paper introduces the Sovereign Trace Protocol (STP), a cryptographic infrastructure for permanent significance registration that binds a moment in time across three simultaneous civilizational calendar systems: the Gregorian civil calendar, the Hebrew lunisolar calendar with full four-dehiyot computation, and the 13 Moon Dreamspell calendar. Each seal is a SHA-256 hash computed locally from Python's standard library with zero external dependencies, zero trusted third parties, and no network requirement. The seal is tamper-evident, reproducible offline, and permanently verifiable by any party (...) with access to the public parameters and FROZEN-4.0 source code. -/- STP operates at two scales. At the individual scale, it provides a sovereignty-first permanence infrastructure for personal record-keeping — the cryptographic registration of observations, decisions, and experiences in a form that is not contingent on institutional intermediaries or audience acknowledgment. At the organizational scale, it introduces the Epistemic Debt Score (EDS), a five-component metric for measuring AI governance integrity: the degree to which an organization maintains a tamper-evident, append-only record of AI system failures, resolves them at a documented rate, and demonstrates improvement over time. -/- The protocol is released as FROZEN-4.0 — a frozen, permanently deployed implementation with 82 self-test checks passing, a documented four-version retirement lineage, and a structured red-team findings log of 48 findings (45 PDE diagnostic findings and 3 MPTE meta-diagnostic findings), all resolved or formally declared open. The paper applies its own epistemic tagging system throughout, distinguishing directly observed claims, reasoned inferences, strategic assessments, and open questions — including explicit declaration of unmitigated vulnerabilities and unvalidated constructs. (shrink)