Explainability in Artificial Intelligence

Multi-agent systems often exhibit emergent behavior that appears coordinated, intelligent, and irreducible to the behavior of individual components. Yet quantifying the degree to which such systems form integrated wholes remains a major challenge. While Integrated Information Theory (IIT) was originally developed to explain consciousness, its core concept - measuring how much a system resists decomposition - has broader relevance for understanding informational integration in complex systems. However, the exact computation of IIT’s central quantity, Φ, is intractable for all but the (...) smallest networks. In this paper, we propose a scalable spectral approximation, Φspectral, which estimates system-level irreducibility by applying spectral graph theory to pairwise mutual information networks derived from time-series data. The Fiedler vector of the normalized Laplacian defines a minimal bipartition, across which we sum mutual information to estimate informational disintegration. We interpret this measure as a proxy for epistemic emergence (specifically the diachronic case)—capturing practical, observer-relative irreducibility without invoking ontological claims. We also explore the implications of this metric for measuring emergence and consciousness. We evaluate Φspectral across four classes of simulated systems: random, transitional, and synchronized oscillator networks, as well as a biologically inspired combinatorial threshold-linear network (CTLN) governed by graph-theoretic dynamics. Our results show that Φspectral is sensitive not merely to coordination, but to the emergence of structurally differentiated and functionally integrated behavior. In particular, it detects transient integration in transitional systems and reveals the collapse of complexity in fixed-point–convergent CTLNs. These findings suggest that Φspectral offers a principled and computationally efficient tool for analyzing irreducibility and emergence in distributed systems, with potential applications in AI safety, neural computation, and collective intelligence. (shrink)

Author: Cristhian Mauricio Beltrán Calderón: Date: October 6, 2025, Zenodo DOI (English version): 10.5281/zenodo.17342722, Zenodo DOI (Spanish version): 10.5281/zenodo.17274781. La filosofía contemporánea enfrenta una crisis temporal donde el desarrollo tecnológico exponencial supera la capacidad de reflexión ética tradicional (Beltrán Calderón, 2025a). Este artículo valida experimentalmente la Filosofía Ficcionante (Beltrán Calderón, 2025b) mediante su implementación en el Protocolo Axiológico Prospectivo (PAP), demostrando que la exploración ética de futuros tecnológicos puede conducirse rigurosamente en entornos de bajos recursos. Cuatro estudios de caso ejecutados (...) en Google Gemini mediante workflow agéntico secuencial generaron conceptos axiológicos novedosos como "Moneda de Ineficiencia" y "Deuda Somática", evaluados mediante una Matriz Pentapartita con Validación de Consenso Triangulado (MCT) implementada a través de simulación de agentes internos (SAI) que reveló consistentemente alto poder revelador (≥4/5) y fecundidad dialéctica (≥4/5) en todos los casos. El trabajo aborda el "bucle metodológico de la IAG" replanteando los sistemas de IA como amplificadores diagnósticos en lugar de oráculos éticos, estableciendo un nuevo paradigma para la prospección filosófica experimental que transforma el contagio cognitivo de debilidad en fortaleza metodológica mediante la cuantificación axiológica y hermenéutica prospectiva. (shrink)

We propose a new interpretability method for neural networks, which is based on a novel mathematico-philosophical theory of reasons. Our method computes a vector for each neuron, called its reasons vector. We then can compute how strongly this reasons vector speaks for various propositions, e.g., the proposition that the input image depicts digit 2 or that the input prompt has a negative sentiment. This yields an interpretation of neurons, and groups thereof, that combines a logical and a Bayesian perspective, and (...) accounts for polysemanticity (i.e., that a single neuron can figure in multiple concepts). We show, both theoretically and empirically, that this method is: (1) grounded in a philosophically established notion of explanation, (2) uniform, i.e., applies to the common neural network architectures and modalities, (3) scalable, since computing reason vectors only involves forward-passes in the neural network, (4) faithful, i.e., intervening on a neuron based on its reason vector leads to expected changes in model output, (5) correct in that the model's reasons structure matches that of the data source, (6) trainable, i.e., neural networks can be trained to improve their reason strengths, (7) useful, i.e., it delivers on the needs for interpretability by increasing, e.g., robustness and fairness. (shrink)

Addendum II of the Human Moral Archive Framework (HMAF) formalizes the structure and analytic function of Moral Distribution Indexing (MDI), extending the probabilistic constructs introduced in the Moral Distribution Model (MDM) and the topographic refinements of the Moral Density Function (MDF). The purpose of MDI is to translate moral information from static distributional description into a temporally and comparatively indexed coordinate system, enabling observation of moral motion, coherence, and drift across agents, institutions, and computational systems. Its architecture accommodates both raw (...) and representational moral data, permitting immediate analytic integration without prescriptive reformulation and demonstrating the model’s adaptability across empirical and theoretical domains. Drawing upon the information-theoretic foundations of entropy, divergence, coherence, and density, the framework defines how moral data may be mapped into an n-dimensional index space, normalized against equilibrium baselines, and measured for directional change through time. The addendum demonstrates how this indexing framework unites the archival and probabilistic layers of HMAF as defined in the MDM into an adaptive analytic architecture capable of longitudinal calibration and cross-domain interoperability. In formalizing this indexing framework, the addendum introduces a series of new mathematical constructs that quantify moral drift, coherence, and stability as measurable relationships within informational space. Through MDI, the HMAF architecture progresses from representing moral information to instrumenting its measurement—transforming a descriptive moral framework into an operational analytic system. This transition establishes the mathematical foundation upon which future predictive constructs, such as the Moral Distribution Predictor (MDP), may logically evolve as subsequent formal developments of the HMAF series. (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)

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)

Judges and administrative decision-makers are expected to explain themselves, giving reasons for their decisions. While the emergence of large language models has prompted renewed interest in the use of AI tools in judicial and administrative settings, existing scholarship identifies the ‘black box’ nature of machine learning tools as impeding transparency and reason-giving. This paper considers the status of the reasoning generated by LLMs in the context of duties of judicial and administrative reason-giving and the problem of explainability. It explains how (...) the move from predictive or discriminative to generative artificial intelligence reshapes the problem of explainability and the limited explanatory value of LLM reasoning. It argues that, notwithstanding the analogous problem of establishing the causal character of human justificatory reasoning, human-generated reasons can be distinguished from machine-generated reasons by their role in a practice of accountability. The fundamental architecture of LLMs precludes their participation in that practice. In consequence, LLMs are incapable of giving reasons in the sense required to support judicial and administrative decision-making. (shrink)

A growing chorus of AI researchers and philosophers posit internal representations in large language models (LLMs). But how do these representations relate to the kinds of mental states we routinely ascribe to our fellow humans? While some research has focused on belief- or knowledge- like states in LLMs, there has been comparatively little focus on the question of whether LLMs have intentions. I survey five properties that have been associated with intentions in the philosophical literature, and assess two candidate classes (...) of LLM representations against this set of features. The result is mixed: LLMs have representations that are intention-like in many— perhaps surprising—respects, but they differ from human intentions in important ways. (shrink)

Mechanistic interpretability (MI) aims to explain how neural networks work by uncovering their underlying mechanisms. As the field grows in influence, it is increasingly important to examine not just models themselves, but the assumptions, concepts and explanatory strategies implicit in MI research. We argue that mechanistic interpretability needs philosophy as an ongoing partner in clarifying its concepts, refining its methods, and navigating the epistemic and ethical complexities of interpreting AI systems. There is significant unrealised potential for progress in MI to (...) be gained through deeper engagement with philosophers and philosophical frameworks. Taking three open problems from the MI literature as examples, this paper illustrates the value philosophy can add to MI research, and outlines a path toward deeper interdisciplinary dialogue. (shrink)

NOTE: This is an early preprint version. The definitive, citable "Version of Record" of this paper has been archived on Zenodo and can be found under the DOI 10.5281/zenodo.15682919. Please use the Zenodo version exclusively for all citations. -/- This paper presents a mathematically formalized model for describing and analyzing worldview dynamics, distinguishing between heuristic and reflective worldviews. It formalizes the psychological mechanisms of authority-based belief and cognitive dissonance, demonstrating how humans evaluate new information through the filter of their existing (...) worldview, with authority, social acceptance, and the avoidance of cognitive dissonance playing central roles. A key contribution is the mathematical operationalization of cognitive dissonance as a measurable quantity D, which emerges from the interplay between consonant and dissonant beliefs in relation to correlation C. By explicitly considering belief relationships, multidimensional evaluation systems, and consistency metrics, implicit heuristic thought processes are transformed into an explicit, traceable form. The paper also introduces the concept of reflective worldview refinement---a process that enables progression from an unreflective, inconsistent worldview to a more differentiated, consistent understanding. The work builds upon the method of reflective empiricism and extends it through a concrete model that makes unconscious distortion mechanisms and subjective filters visible. The model primarily serves for structural clarification rather than quantitative prediction, functioning as a conceptual bridge between subjective experience and scientific analysis. Beyond psychological and epistemological implications, applications in artificial intelligence are discussed, particularly the development of a reflective learning method for Large Language Models. The paper provides a valuable contribution to a deeper understanding of human cognitive processes and demonstrates how the supposed objectivity of science can be transcended through conscious reflection on one's own subjectivity. (shrink)

Human-centered explainable AI (HCXAI) advocates for the integration of social aspects into AI explanations. Central to the HCXAI discourse is the Social Transparency (ST) framework, which aims to make the socio-organizational context of AI systems accessible to their users. In this work, we suggest extending the ST framework to address the risks of social misattributions in Large Language Models (LLMs), particularly in sensitive areas like mental health. In fact LLMs, which are remarkably capable of simulating roles and personas, may lead (...) to mismatches between designers' intentions and users' perceptions of social attributes, risking to promote emotional manipulation and dangerous behaviors, cases of epistemic injustice, and unwarranted trust. To address these issues, we propose enhancing the ST framework with a fifth 'W-question' to clarify the specific social attributions assigned to LLMs by its designers and users. This addition aims to bridge the gap between LLM capabilities and user perceptions, promoting the ethically responsible development and use of LLM-based technology. (shrink)

What can contemporary machine learning (ML) models do? Given the proliferation of ML models in society, answering this question matters to a variety of stakeholders, both public and private. The evaluation of models' capabilities is rapidly emerging as a key subfield of modern ML, buoyed by regulatory attention and government grants. Despite this, the notion of an ML model possessing a capability has not been interrogated: what are we saying when we say that a model is able to do something? (...) And what sorts of evidence bear upon this question? -/- In this paper, we aim to answer these questions, using the capabilities of large language models (LLMs) as a running example. Drawing on the large philosophical literature on abilities, we develop an account of ML models' capabilities which can be usefully applied to the nascent science of model evaluation. Our core proposal is a conditional analysis of model abilities (CAMA): crudely, a machine learning model has a capability to X just when it would reliably succeed at doing X if it 'tried'. The main contribution of the paper is making this proposal precise in the context of ML, resulting in an operationalisation of CAMA applicable to LLMs. We then put CAMA to work, showing that it can help make sense of various features of ML model evaluation practice, as well as suggest procedures for performing fair inter-model comparisons. (shrink)

Intelligibility and interpretability related to artificial intelligence (AI) are crucial for enabling explicability, which is vital for establishing constructive communication and agreement among various stakeholders, including users and designers of AI. It is essential to overcome the challenges of sharing an understanding of the details of the various structures of diverse AI systems, to facilitate effective communication and collaboration. In this paper, we propose four fundamental terms: “I/O,” “Constraints,” “Objectives,” and “Architecture.” These terms help mitigate the challenges associated with intelligibility (...) and interpretability in AI by providing appropriate levels of abstraction to describe structure of AI systems generally, thereby facilitating the sharing of understanding among various stakeholders. The relationship between the Objective of AI designers and the Purpose of AI users is linked to the issues of AI alignment. (shrink)

According to the standard Humean account of laws of nature, laws are selected partly as a result of an optimal trade-off between the scientific virtues of simplicity and strength. Roberts and Woodward have recently objected that such trade-offs play no role in how laws are chosen in science. In this paper, we first discuss an example from the field of automated scientific discovery which provides concrete support for Roberts and Woodward’s point that scientific theories are chosen based on a single-virtue (...) threshold. However, we then use this very same example as a starting point to argue that i) by insisting on a single best theory, Humeans rely on an overly simplistic conception of trade-offs, that ii) this conception should give way to one which allows for a Pareto front of equally optimal theories, and iii) that given this new conception, threshold behaviour for a virtue like strength is a) compatible with the existence of a genuine trade-off and b) can even play the positive role of a selection criterion. (shrink)

Uno de los principios éticos mencionados más frecuentemente en los lineamientos para el desarrollo de la inteligencia artificial (IA) es la transparencia algorítmica. Sin embargo, no existe una definición estándar de qué es un algoritmo transparente ni tampoco es evidente por qué la opacidad algorítmica representa un reto para el desarrollo ético de la IA. También se afirma a menudo que la transparencia algorítmica fomenta la confianza en la IA, pero esta aseveración es más una suposición a priori que una (...) tesis basada en evidencia empírica. Tampoco se discute mucho hasta qué punto es técnicamente posible volver transparente la caja negra de la IA a través de los métodos de explicabilidad. En este capítulo haré un análisis de la interrelación entre los conceptos de transparencia, explicabilidad y confianza. Inicialmente, analizaré los diferentes tipos de opacidad algorítmica para entender mejor cuál es el problema al que nos enfrentamos. En las secciones subsiguientes trazaré la relación entre explicabilidad y transparencia, y presentaré los límites de los métodos actuales de explicabilidad. En la sección final, examinaré la evidencia empírica acerca de la relación entre transparencia y confianza en las decisiones automatizadas basadas en sistemas de IA. (shrink)

For synergistic interactions between humans and artificial intelligence (AI) systems, AI outputs often need to be explainable to people. Explainable AI (XAI) systems are commonly tested in human user studies. However, whether XAI researchers consider potential cultural differences in human explanatory needs remains unexplored. We highlight psychological research that found significant differences in human explanations between many people from Western, commonly individualist countries and people from non-Western, often collectivist countries. We argue that XAI research currently overlooks these variations and that (...) many popular XAI designs implicitly and problematically assume that Western explanatory needs are shared cross-culturally. Additionally, we systematically reviewed over 200 XAI user studies and found that most studies did not consider relevant cultural variations, sampled only Western populations, but drew conclusions about human-XAI interactions more generally. We also analyzed over 30 literature reviews of XAI studies. Most reviews did not mention cultural differences in explanatory needs or flag overly broad cross-cultural extrapolations of XAI user study results. Combined, our analyses provide evidence of a cultural bias toward Western populations in XAI research, highlighting an important knowledge gap regarding how culturally diverse users may respond to widely used XAI systems that future work can and should address. (shrink)

This paper presents a new account of pragmatic understanding based on the idea that such understanding requires skills rather than abilities. Specifically, one has pragmatic understanding of an affordance space when one has, and is responsible for having, skills that facilitate the achievement of some aims using that affordance space. In science, having skills counts as having pragmatic understanding when the development of those skills is praiseworthy. Skills are different from abilities at least in the sense that they are task-specific, (...) can be learned, and we have some cognitive control over their deployment. This paper considers how the use of AI in science facilitates or frustrates the achievement of this kind of understanding. I argue that we cannot properly ascribe this kind of understanding to any current or near-future algorithm itself. But there are ways that we can use AI algorithms to increase pragmatic understanding, namely, when we take advantage of their abilities to increase our own skills (as individuals or communities). This can happen when AI features in human-performed science as either a tool or a collaborator. (shrink)

We argue that the trend toward providing users with feasible and actionable explanations of AI decisions—known as recourse explanations—comes with ethical downsides. Specifically, we argue that recourse explanations face several conceptual pitfalls and can lead to problematic explanation hacking, which undermines their ethical status. As an alternative, we advocate that explanations of AI decisions should aim at understanding.

Explainable AI (xAI) methods are important for establishing trust in using black-box models. However, recent criticism has mounted against current xAI methods that they disagree, are necessarily false, and can be manipulated, which has started to undermine the deployment of black-box models. Rudin (2019) goes so far as to say that we should stop using black-box models altogether in high-stakes cases because xAI explanations ‘must be wrong’. However, strict fidelity to the truth is historically not a desideratum in science. Idealizations (...) – the intentional distortions introduced to scientific theories and models – are commonplace in the natural sciences and are seen as a successful scientific tool. Thus, it is not falsehood qua falsehood that is the issue. In this paper, I outline the need for xAI research to engage in idealization evaluation. Drawing on the use of idealizations in the natural sciences and philosophy of science, I introduce a novel framework for evaluating whether xAI methods engage in successful idealizations or deceptive explanations (SIDEs). SIDEs evaluates whether the limitations of xAI methods, and the distortions that they introduce, can be part of a successful idealization or are indeed deceptive distortions as critics suggest. I discuss the role that existing research can play in idealization evaluation and where innovation is necessary. Through a qualitative analysis we find that leading feature importance methods and counterfactual explanations are subject to idealization failure and suggest remedies for ameliorating idealization failure. (shrink)

This contribution builds upon recent work by the author on investigating the ways in which epistemic defeat arises in machine learning (ML) and the ethical problems that it raises. The contribution presents an epistemological approach to the problems of opacity and algorithmic bias in machine learning by discussing them in terms of the forms of epistemic defeat that arise in them. A taxonomy of epistemic defeaters, including transparent, opaque, and inherited defeaters, is developed and employed for this purpose. The Black-Box (...) problem is discussed, and a distinction between two construals of the ‘why’ question that defines the Black-Box problem is proposed. These construals are based upon a distinction between a request for an explanation of a model’s output in terms of a causal mechanism and an explanation in terms of reasons or justification. The latter is shown to be the more challenging problem that remains difficult to solve even if a solution to the former is available. Particular attention will be given to the phenomenon of inherited epistemic defeat, which is a form of epistemic defeat involved in algorithmic bias that is derived from biases encoded in training data. Such defeaters are inherited via training data and inculcated in machine learning models, thereby exacerbating and obfuscating the effect of their epistemic defeat. This contribution builds upon a recent collaborative project conducted on machine learning in midwifery by the author (Begley et al. 2021). Examples from that study are considered along with analogous examples from predictive policing. Recommendations are made in particular regarding how inherited epistemic defeat should best be addressed. (shrink)

Artificial intelligence (AI)-enabled remote patient monitoring (RPM) is promoted as a solution to rising pressures in health care, including personnel shortages and the growing burden associated with population aging and chronic disease management. Yet, the ethical implications of deploying adaptive systems in routine care remain underexamined at the level of specific, situated use cases. This article examines the ethical risks of MonitAir, an AI-enabled RPM system for chronic obstructive pulmonary disease (COPD) in a Swedish health care setting. Drawing on the (...) Three Domains, Six Levels (3D6L) framework, we identify epistemic, normative and traceability-related risks across six levels of abstraction. The article offers, to our knowledge, the first operationalization of the 3D6L as an analytic tool for screening ethical risks. We argue that screening AI-enabled health care technology with 3D6L clarifies how ethical risks manifest across levels, from individual patients and patient–clinician relationships to organizational and sectoral contexts. In addition, the framework’s minimal normativity allows alignment with context-sensitive principles and guidelines. Through this analysis, we identify ethical risks related to data bias, intelligibility of outputs, uneven access and blurred responsibility, including redistributive and role-shifting effects. While MonitAir may support earlier detection of exacerbations, its implementation in Swedish COPD care may also reproduce and amplify existing health disparities or overburden patients without sufficient support. We demonstrate how structured ethical screening makes visible concerns typically overlooked by the optimization discourse in digital health. Finally, we argue that open-ended evaluation of ethical risks of AI-enabled DHT provides a valuable early phase that complements ethical assessment, without collapsing into checklist compliance. (shrink)

Since the early days of the Explainable AI movement, post-hoc explanations have been praised for their potential to improve user understanding, promote trust, and reduce patient safety risks in black box medical AI systems. Recently, however, critics have argued that the benefits of post-hoc explanations are greatly exaggerated since they merely approximate, rather than replicate, the actual reasoning processes that black box systems take to arrive at their outputs. In this article, we aim to defend the value of post-hoc explanations (...) against this recent critique. We argue that even if post-hoc explanations do not replicate the exact reasoning processes of black box systems, they can still improve users’ functional understanding of black box systems, increase the accuracy of clinician-AI teams, and assist clinicians in justifying their AI-informed decisions. While post-hoc explanations are not a “silver bullet” solution to the black box problem in medical AI, we conclude that they remain a useful strategy for addressing the black box problem in medical AI. (shrink)

本稿は評価構造理論に基づき、差異帰属過程としての評価の形式モデルを提示する。従来の心の哲学では、対象や主体を前提として心や意識が説明されることが多い。本研究はこの順序を反転し、操作構造と差異集中から評 価構造を導出し、その作動状態を数理的に記述する枠組みを提示する。 生成経路集合と結果集合の関係から誘導される応答分布に対し、評価エントロピーを定義する。評価安定度および評価動力学を導入し、評価を差異帰属候補の集合に対する探索過程として定式化する。この探索は通常、帰属 候補の縮約を伴うため、平均的にエントロピー減衰が生じる。本稿ではこの過程を一次緩和方程式によって記述する。 このモデルにより、評価構造の作動状態はエントロピーとその時間変化によって分類される。エントロピーが消失する極限では評価は閉鎖し対象が生成される。エントロピーが減衰する過程は評価持続として現れ、その積分 量は意識強度として定義される。さらに差異帰属の排除が停止する場合、エントロピーが正の下限を保つ非排除評価が成立する。本稿ではこの作動様式を慈悲構造と呼ぶ。 さらに評価エントロピーの減衰速度を規定するパラメータとして評価速度を導入し、その起源を計算能力、学習構造、情報不確実性の相互作用として定式化する。 以上により本稿は、対象生成、意識持続、慈悲構造を単一の評価動力学から導出する形式モデルを提示する。この結果、評価構造理論は操作体系一般に適用可能な動力学的枠組みとして再構成される。.

This paper uses famous problems from philosophy of science and philosophical psychology—underdetermination of theory by evidence, Nelson Goodman’s new riddle of induction, theory-ladenness of observation, and “Kripkenstein’s” rule-following paradox—to show that it is empirically impossible to reliably interpret which functions a large language model (LLM) AI has learned, and thus, that reliably aligning LLM behavior with human values is provably impossible. Sections 2 and 3 show that because of how complex LLMs are, researchers must interpret their learned functions largely in (...) terms of empirical observations of their outputs and network behavior. Sections 4–7 then show that for every “aligned” function that might appear to be confirmed by empirical observation, there is always an infinitely larger number of “misaligned”, arbitrarily time-limited functions equally consistent with the same data. Section 8 shows that, from an empirical perspective, we can thus never reliably infer that an LLM or subcomponent of one has learned any particular function at all before any of an uncountably large number of unpredictable future conditions obtain. Finally, Section 9 concludes that the probability of LLM “misalignment” is—at every point in time, given any arbitrarily large body of empirical evidence—always vastly greater than the probability of “alignment.”. (shrink)

Imagine that in the cold heart of a secret military facility, a new form of intelligence awakens. It is a synthetic mind born from intricate algorithms and complex computations, operating in ways unfathomable to its human creators. Charged with safeguarding national security, this intelligence orchestrates strategies that defy human ethics and laws of war, leaving its creators both awed and unnerved. Unknowable Minds delves into the unsettling reality of entrusting our safety to an intelligence that lacks human essence. As we (...) navigate the Age of Artificial Intelligence, these systems — powering everything from our smartphones to military defenses — remain inherently opaque and unpredictable. The book explores how AI differs from any technology we've ever developed, its inherent complexities, and the profound risks it poses to our future. Drawing on philosophy, AI theory, and national security insights, this book offers a thought-provoking examination of AI’s potential and peril. From the complexities of neural networks to the unpredictable nature of emergent behaviors, Unknowable Minds challenges us to rethink our relationship with AI and its role in the theater of global security. Can we control an unknowable intellect, or will it redefine human existence? As we stand on the precipice of unprecedented technological advancement, understanding and navigating the unknowable minds of artificial intelligence becomes a quest fraught with extraordinary challenges and existential questions. (shrink)

There has been a surge of interest in explainable artificial intelligence (XAI). It is commonly claimed that explainability is necessary for trust in AI, and that this is why we need it. In this paper, I argue that for some notions of trust it is plausible that explainability is indeed a necessary condition. But that these kinds of trust are not appropriate for AI. For notions of trust that are appropriate for AI, explainability is not a necessary condition. I thus (...) conclude that explainability is not necessary for trust in AI that matters. (shrink)

The apparent downsides of opaque algorithms has led to a demand for explainable AI (XAI) methods by which a user might come to understand why an algorithm produced the particular output it did, given its inputs. Patients, for example, might find that the lack of explanation of the process underlying the algorithmic recommendations for diagnosis and treatment hinders their ability to provide informed consent. This paper examines the impact of two factors on user perceptions of explanations for AI systems in (...) medical contexts. The factors considered were the stakes of the decision—high versus low—and the decision source—human versus AI. 484 participants were presented with vignettes in which medical diagnosis and treatment plan recommendations were made by humans or by AI. Separate vignettes were used for high stakes scenarios involving life-threatening diseases, and low stakes scenarios involving mild diseases. In each vignette, an explanation for the decision was given. Four explanation types were tested across separate vignettes: no explanation, counterfactual, causal and a novel ‘narrative-based’ explanation, not previously considered. This yielded a total of 16 conditions, of which each participant saw only one. Individuals were asked to evaluate the explanations they received based on helpfulness, understanding, consent, reliability, trust, interests and likelihood of undergoing treatment. We observed a main effect for stakes on all factors and a main effect for decision source on all factors except for helpfulness and likelihood to undergo treatment. While we observed effects for explanation on helpfulness, understanding, consent, reliability, trust and interests, we by and large did not see any differences between the effects of explanation types. This suggests that the effectiveness of explanations may not depend on type of explanation but instead, on the stakes and decision source. (shrink)

Recent theoretical and practical achievements in machine learning (ML) and, in particular, artificial neural networks, have motivated ethical questions about their deployment. This chapter critically examines the nature of doing ethics in and for contemporary ML and artificial intelligence (AI). It discusses some prominent epistemological problems, ethical problems regarding bias and fairness, the moral status of AI and how it bears on the problems of responsibility gaps and alignment, the use or misuse of ethical theory in AI, and attendant problems (...) of ethicswashing. Discussion of these topics will show that ethics is done best when integrated as a philosophical discipline. (shrink)

Since the emergence of the field of eXplainable Artificial Intelligence (XAI), a growing number of researchers have argued that XAI should consider insights from the social sciences in order to adapt explanations to the expectations and needs of human users. This has led to the emergence of a field called Social XAI, which is concerned with understanding how explanations are actively shaped in the interaction between a human user and an AI system. Recognizing this turn in XAI toward making XAI (...) systems more “social” by providing explanations that focus on human information needs and incorporating insights from human–human explanatory interactions, in this paper we provide a formal foundation for Social XAI. We do so by proposing novel ontological accounts of the key terms used in Social XAI based on Basic Formal Ontology (BFO). Specifically, we provide novel ontological accounts for explanandum, explanans, understanding, explanation, explainer, explainee, and context. In doing so, we discuss multifaceted entities in Social XAI (having both continuant and occurrent facets; e.g., explanation) and the relationship between understanding and explanation. Additionally, we propose solutions to seemingly paradoxical views on some terms (e.g., social constructivist vs. individual constructivist perspective on explanandum). (shrink)

AI systems increasingly support human decision-making across domains of professional, skill-based, and personal activity. While previous work has examined how AI might affect human autonomy globally, the effects of AI on domain-specific autonomy -- the capacity for self-governed action within defined realms of skill or expertise -- remain understudied. We analyze how AI decision-support systems affect two key components of domain-specific autonomy: skilled competence (the ability to make informed judgments within one's domain) and authentic value-formation (the capacity to form genuine (...) domain-relevant values and preferences). By engaging with prior investigations and analyzing empirical cases across medical, financial, and educational domains, we demonstrate how the absence of reliable failure indicators and the potential for unconscious value shifts can erode domain-specific autonomy both immediately and over time. We then develop a constructive framework for autonomy-preserving AI support systems. We propose specific socio-technical design patterns -- including careful role specification, implementation of defeater mechanisms, and support for reflective practice -- that can help maintain domain-specific autonomy while leveraging AI capabilities. This framework provides concrete guidance for developing AI systems that enhance rather than diminish human agency within specialized domains of action. (shrink)

AI safety is an interdisciplinary field focused on mitigating the harms caused by AI systems. We review a range of research directions in AI safety, focusing on those to which philosophers have made or are in a position to make the most significant contributions. These include ethical AI, which seeks to instill human goals, values, and ethical principles into artificial systems, scalable oversight, which seeks to develop methods for supervising the activity of artificial systems even when they become significantly more (...) capable than their human designers, interpretability, which seeks to render comprehensible the workings of complex machine learning models, and corrigibility, which seeks to discover ways to ensure that powerful AI systems will not resist being shut down or modified by humans. (shrink)

Recently, there has been a large push for the use of artificial intelligence in medical settings. The promise of artificial intelligence (AI) in medicine is considerable, but its moral implications are in-sufficiently examined. If AI is used in medical diagnosis and treatment, it may pose a substantial problem for informed consent. The short version of the problem is this: medical AI will likely surpass human doctors in accuracy, meaning that patients have a prudential reason to prefer treatment from an AI. (...) However, given the black box problem, medical AI cannot explain to patients how it makes decisions, yet such an explanation seems to be required by informed consent. Thus, it seems that doing what is best for patients (treatment via AI), even if patients want to permit this, might be prohibited by medicine’s commitment to informed consent. Conflicts between beneficence and autonomy are not new, but medical AI poses a novel version of this conflict, because this problem is one in which even if the patient says they want to use their autonomy to receive better care, the commitment to autonomy (via informed consent) seems to block them from doing so. Given this dilemma, should we abandon informed consent, or should we not use medical AI? My thesis is that we can have our cake and eat it too; we can use opaque AI in clinical medicine and retain our com-mitment to informed consent, although it may require revising our understanding of informed con-sent. Specifically, it will require us to distinguish between two levels of consent (higher-order and first-order consent). (shrink)

Achieving trustworthy AI is increasingly considered an essential desideratum to integrate AI systems into sensitive societal fields, such as criminal justice, finance, medicine, and healthcare, among others. For this reason, it is important to spell out clearly its characteristics, merits, and shortcomings. This article is the first survey in the specialized literature that maps out the philosophical landscape surrounding trust and trustworthiness in AI. To achieve our goals, we proceed as follows. We start by discussing philosophical positions on trust and (...) trustworthiness, focusing on interpersonal accounts of trust. This allows us to explain why trust, in its most general terms, is to be understood as reliance plus some “extra factor”. We then turn to the first part of the definition provided, i.e., reliance, and analyze two opposing approaches to establishing AI systems’ reliability. On the one hand, we consider transparency and, on the other, computational reliabilism. Subsequently, we focus on debates revolving around the “extra factor”. To this end, we consider viewpoints that most actively resist the possibility and desirability of trusting AI systems before turning to the analysis of the most prominent advocates of it. Finally, we take up the main conclusions of the previous sections and briefly point at issues that remain open and need further attention. (shrink)

AI agents—autonomous, multi-tasking systems beyond traditional AI—will reshape human-AI interaction shifting the focus from a simple human-versus-system autonomy debate to a triadic model: human users, AI agents, and digital resources. Further, the widespread adoption of agentic systems as consumer products will accelerate the large-scale integration of novel human-AI agent hybridization into everyday life, necessitating renewed examinations of agent identity, accountability, and trust in AI agent-rich digital environments. In this position paper, we argue that HumanCentered eXplainable AI (HCXAI) can help addressing (...) these trends by developing causal explanations that span multiple levels of abstraction and integrate different digital resources, as well as personalized, interactive explanations that serve as identity warranty mechanisms when humans and AI agents interact together and with other digital resources. Finally, as a research field, HCXAI can help redesign the explainability-trust relation in these novel digital ecosystems. We introduce examples of these explanations, calling for an interdisciplinary collaboration to refine these high-level proposals and ensure a still human-centered digital future. (shrink)

A compilation of cutting-edge, comprehensive insights into digital ethics from leading scholars As digital technologies shape every aspect of today's society, ethical considerations have never been more pressing. In A Companion to Digital Ethics, editors Luciano Floridi and Mariarosaria Taddeo bring together leading experts to analyse key ethical challenges posed by artificial intelligence, privacy, cybersecurity, cyberwarfare, sustainability, digital consent, and many other topics. With a multidisciplinary approach, this authoritative volume introduces all the relevant topics in digital ethics clearly and accessibly, (...) integrating insights from philosophy, law, computer science, and the social sciences. Unlike other collections that focus on specialised research or introductory overviews, this Companion strikes a critical balance—offering a systematic, comprehensive, simple yet rigorous examination of digital ethics that is both conceptually rich and practically relevant. Each chapter can be read as part of the whole or independently. Throughout the book, the contributing authors equip the reader with diverse perspectives for navigating ethical dilemmas in the evolving digital landscape, with an essential list of references and further readings. Integrating real-world examples and case studies throughout, A Companion to Digital Ethics: Explores the ethical value and implications of digital technologies and their applications Offers clear and precise analysis of both current and potential ethical challenges Includes discussions of ethical issues in contexts of technology governance and digital law Helps professionals and policymakers implement ethical principles in AI and digital technologies Contains authoritative and cutting-edge chapters, edited by two leading experts in digital ethics Whether used in academic courses or professional discussions on responsible digital innovation, A Companion to Digital Ethics is an essential guide to understanding and addressing the ethical questions of the digital age. It is a must-have resource for undergraduate and postgraduate students in any academic discipline, as well as policymakers and practitioners seeking to implement ethical frameworks for emerging technologies. (shrink)

As opaque algorithmic systems take up a larger and larger role in shaping our lives, calls for explainability in various algorithmic systems have increased. Many moral and political philosophers have sought to vindicate these calls for explainability by developing theories on which decision-subjects—that is, individuals affected by decisions—have a moral right to the explanation of the systems that affect them. Existing theories tend to suggest that the right to explanation arises solely in virtue of facts about how decision-subjects are affected (...) by opaque systems. But this ignores an important part of the normative landscape: sometimes, facts about a decision-maker’s rights and freedoms can make a difference to whether she owes anyone explanations of her decisions. I illustrate the point by calling attention to examples of artistic, romantic, and supererogatory altruistic decisions; in at least some cases of this sort, even highly influential decisions can legitimately be kept entirely private and opaque. (shrink)

Many have recently argued that there are weighty reasons against making high-stakes decisions solely on the basis of recommendations from artificially intelligent (AI) systems. Even if deference to a given AI system were known to reliably result in the right action being taken, the argument goes, that deference would lack morally important characteristics: the resulting decisions would not, for instance, be based on an appreciation of right-making reasons. Nor would they be performed from moral virtue; nor would they have moral (...) worth. I argue that, even if these characteristics all have intrinsic value, that intrinsic value has no practical relevance to decisions about whether to defer to AI. I make that point by drawing on a lesson from the literature on moral testimony. Once it is granted that deference to a reliable source is the policy most likely to bring about right action, a refusal to defer carries with it a heightened risk of wronging and mistreating people. And that heightened risk of wrongdoing, I argue, cannot be justified by appeal to the intrinsic value of striving for a morally exemplary decision-making process. (shrink)

A recent series of papers have challenged whether black box AI challenges informed consent. Director has most recently argued that the explicit canonical view of informed consent and the implicit view held by bioethicists neglects a crucial distinction between first-order and higher-order evidence needed to show that black box AI does not threaten efficacious consent. I believe a charitable reconstruction of the canonical view and the implicit view held by bioethicists does not neglect the crucial distinction and, indeed, many clear (...) cases of informed consent involve a notable lack of first-order information but a wealth of second-order information. Thus, it most reasonable to interpret the ‘informed’ portion of ‘informed consent’ as (i) relative to what we know, and (ii) inclusive of both first- and higher-order information. This strengthens the argument for Director’s conclusions about black box AI: its purported threat to informed consent are instead worries about the reliability of black box AI. (shrink)

The article investigates the ethical and philosophical challenges posed by the integration of artificial intelligence in vascular surgery. Based on systematic data analysis and an interdisciplinary approach, the authors evaluate key AI advances, including personalization of treatment, automation of diagnosis, and prediction of complications. Particular attention is paid to the problems of accountability for algorithm decisions, transparency of “black box” systems, dehumanization of medicine, and cyborgization. The contradictions between technological efficiency and preservation of humanitarian values are revealed. The study emphasizes (...) the need for legal, educational and ethical mechanisms for the harmonious implementation of AI, where the doctor remains the central figure in the system “human-algorithm-patient” // Статья посвящена исследованию этических и философских вызовов, возникающих при интеграции искусственного интеллекта в сосудистую хирургию. На основе систематического анализа данных и междисциплинарного подхода авторы оценивают ключевые достижения ИИ, включая персонализацию лечения, автоматизацию диагностики и прогнозирование осложнений. Особое внимание уделено проблемам ответственности за решения алгоритмов, прозрачности «черных ящиков», дегуманизации медицины и киборгизации. Раскрыты противоречия между технологической эффективностью и сохранением гуманитарных ценностей. Исследование подчеркивает необходимость правовых, образовательных и этических механизмов для гармоничного внедрения ИИ, где врач остается центральным звеном в системе «человек-алгоритм-пациент». (shrink)

The field of AI safety seeks to prevent or reduce the harms caused by AI systems. A simple and appealing account of what is distinctive of AI safety as a field holds that this feature is constitutive: a research project falls within the purview of AI safety just in case it aims to prevent or reduce the harms caused by AI systems. Call this appealingly simple account The Safety Conception of AI safety. Despite its simplicity and appeal, we argue that (...) The Safety Conception is in tension with at least two trends in the ways AI safety researchers and organizations think and talk about AI safety: first, a tendency to characterize the goal of AI safety research in terms of catastrophic risks from future systems; second, the increasingly popular idea that AI safety can be thought of as a branch of safety engineering. Adopting the methodology of conceptual engineering, we argue that these trends are unfortunate: when we consider what concept of AI safety it would be best to have, there are compelling reasons to think that The Safety Conception is the answer. Descriptively, The Safety Conception allows us to see how work on topics that have historically been treated as central to the field of AI safety is continuous with work on topics that have historically been treated as more marginal, like bias, misinformation, and privacy. Normatively, taking The Safety Conception seriously means approaching all efforts to prevent or mitigate harms from AI systems based on their merits rather than drawing arbitrary distinctions between them. (shrink)

Machine learning (ML) systems are vulnerable to performance decline over time due to dataset shift. To address this problem, experts often suggest that ML systems should be regularly updated to ensure ongoing performance stability. Some scholarly literature has begun to address the epistemic and ethical challenges associated with different updating methodologies. Thus far, however, little attention has been paid to the impact of model updating on the ML-assisted decision-making process itself. This article aims to address this gap. It argues that (...) model updating introduces a new sub-type of opacity into ML-assisted decision-making—update opacity—that occurs when users cannot understand how or why an update has changed the reasoning or behaviour of an ML system. This type of opacity presents a variety of distinctive epistemic and safety concerns that available solutions to the black box problem in ML are largely ill-equipped to address. A variety of alternative strategies may be developed or pursued to address the problem of update opacity more directly, including bi-factual explanations, dynamic model reporting, and update compatibility. However, each of these strategies presents its own risks or carries significant limitations. Further research will be needed to address the epistemic and safety concerns associated with model updating and update opacity going forward. (shrink)

It is commonly accepted that clinicians are ethically obligated to disclose their use of medical machine learning systems to patients, and that failure to do so would amount to a moral fault for which clinicians ought to be held accountable. Call this ‘the disclosure thesis.’ Four main arguments have been, or could be, given to support the disclosure thesis in the ethics literature: the risk-based argument, the rights-based argument, the materiality argument and the autonomy argument. In this article, I argue (...) that each of these four arguments are unconvincing, and therefore, that the disclosure thesis ought to be rejected. I suggest that mandating disclosure may also even risk harming patients by providing stakeholders with a way to avoid accountability for harm that results from improper applications or uses of these systems. (shrink)

This research poses the question of whether it is possible to optimize modern legislative drafting by integrating LLM-based systems into the lawmaking process to address the pervasive challenge of misinformation and disinformation in the age of AI. While misinformation is not a novel phenomenon, with the proliferation of social media and AI, disseminating false or misleading information has become a pressing societal concern, undermining democratic processes, public trust, and social cohesion. AI can be used to proliferate disinformation and misinformation through (...) fake news and deepfakes; can AI also be used for beneficial purposes to develop the antidote legislation combatting these challenges? Leveraging the capabilities of LLMS, such as ChatGPT and others, can present a promising direction for optimizing legislative drafting. By proposing the methodological approach of an AI bun, this article explores an important approach in which LLMS can support lawmakers and policy experts in crafting legislation. The article contributes to the discourse through a nuanced understanding of the opportunities and challenges in harnessing LLM-powered tools for legislative innovation. Ultimately, it underscores the transformative potential of LLMs as a potential resource for lawmakers seeking to navigate decision-making while developing legislation on an example of navigating the intricate landscape of misinformation and disinformation regulation in the age of AI. (shrink)

While there has been much discussion of whether AI systems could function as moral agents or acquire sentience, there has been very little discussion of whether AI systems could have free will. I sketch a framework for thinking about this question, inspired by Daniel Dennett’s work. I argue that, to determine whether an AI system has free will, we should not look for some mysterious property, expect its underlying algorithms to be indeterministic, or ask whether the system is unpredictable. Rather, (...) we should simply ask whether we have good explanatory reasons to view the system as an intentional agent, with the capacity for choice between alternative possibilities and control over the resulting actions. If the answer is “yes”, then the system counts as having free will in a pragmatic and diagnostically useful sense. (shrink)

“AI4Science” refers to the use of Artificial Intelligence (AI) in scientific research. As AI systems become more widely used in science, we need guidelines for when such uses are acceptable and when they are unacceptable. To that end, I propose that the distinction between the context of discovery and the context of justification, which comes from philosophy of science, may provide a preliminary but still useful guideline for acceptable uses of AI in science. Given that AI systems used in scientific (...) research are black boxes, for the most part, we should use such systems in the context of discovery but not in the context of justification. The former refers to processes of idea generation, which may be unproblematically opaque whether they occur in human brains or artificial neural networks, whereas the latter refers to scientific methods by which scientific ideas are tested, confirmed, verified, and justified, which should be transparent. (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)

Artificial Intelligence is a field that lives many lives, and the term has come to encompass a motley collection of scientific and commercial endeavours. In this paper, I articulate the contours of a rather neglected but central scientific role that AI has to play, which I dub “AI-as-exploration”. The basic thrust of AI-as-exploration is that of creating and studying systems that can reveal candidate building blocks of intelligence that may differ from the forms of human and animal intelligence we are (...) familiar with. In other words, I suggest that AI is one of the best tools we have for exploring intelligence space, namely the space of possible intelligent systems. I illustrate the value of AI-as-exploration by focusing on a specific case study, i.e., recent work on the capacity to combine novel and invented concepts in humans and Large Language Models. I show that the latter, despite showing human-level accuracy in such a task, most probably solve it in ways radically different, but no less relevant to intelligence research, to those hypothesised for humans. (shrink)

A key assumption fuelling optimism about the progress of large language models (LLMs) in accurately and comprehensively modelling the world is that the truth is systematic: true statements about the world form a whole that is not just consistent, in that it contains no contradictions, but coherent, in that the truths are inferentially interlinked. This holds out the prospect that LLMs might in principle rely on that systematicity to fill in gaps and correct inaccuracies in the training data: consistency and (...) coherence promise to facilitate progress towards comprehensiveness in an LLM’s representation of the world. However, philosophers have identified compelling reasons to doubt that the truth is systematic across all domains of thought, arguing that in normative domains, in particular, the truth is largely asystematic. I argue that insofar as the truth in normative domains is asystematic, this renders it correspondingly harder for LLMs to make progress, because they cannot then leverage the systematicity of truth. And the less LLMs can rely on the systematicity of truth, the less we can rely on them to do our practical deliberation for us, because the very asystematicity of normative domains requires human agency to play a greater role in practical thought. (shrink)

This paper argues that explainability is only one facet of a broader ideal that shapes our expectations towards artificial intelligence (AI). Fundamentally, the issue is to what extent AI exhibits systematicity—not merely in being sensitive to how thoughts are composed of recombinable constituents, but in striving towards an integrated body of thought that is consistent, coherent, comprehensive, and parsimoniously principled. This richer conception of systematicity has been obscured by the long shadow of the “systematicity challenge” to connectionism, according to which (...) network architectures are fundamentally at odds with what Fodor and colleagues termed “the systematicity of thought.” I offer a conceptual framework for thinking about “the systematicity of thought” that distinguishes four senses of the phrase. I use these distinctions to defuse the perceived tension between systematicity and connectionism and show that the conception of systematicity that historically shaped our sense of what makes thought rational, authoritative, and scientific is more demanding than the Fodorian notion. To determine whether we have reason to hold AI models to this ideal of systematicity, I then argue, we must look to the rationales for systematization and explore to what extent they transfer to AI models. I identify five such rationales and apply them to AI. This brings into view the “hard systematicity challenge.” However, the demand for systematization itself needs to be regulated by the rationales for systematization. This yields a dynamic understanding of the need to systematize thought, which tells us how systematic we need AI models to be and when. (shrink)