A Polyphonic Conception of AI Understanding

With Pierre Beckmann.

Argues that whether large language models understand what they talk about is a question both unavoidable and ill-framed: unavoidable because users must tell outputs that merit epistemic trust from those that do not, and no purely mathematical or statistical vocabulary can draw that line without reintroducing the concept of understanding in all but name; ill-framed because the inherited concept presupposes a single locus of understanding. Mechanistic interpretability instead reveals LLMs to be deeply polyphonic systems whose outputs emerge from parallel mechanisms of uneven reliability. Drawing on evidence of medical, mathematical, logical, and spatial understanding in LLMs, the paper engineers a conception fit for such systems: a model understands a domain insofar as it contains sound circuitry that tracks relevant structure and reliably recruits that circuitry to control its outputs—which turns seemingly idle metaphysical disputes into tractable empirical questions that can guide the allocation of trust.

AI, large language models, understanding, mechanistic interpretability, conceptual engineering, conceptual adaptation, explainable AI

Evolving Minds, Evolving Language: Metaphor as a Process of Conceptual Adaptation to Artificial Intelligence

Manuscript (+Slideshow)

Argues that cognitively loaded terms such as “understanding” and “reasoning,” when applied to large language models, are best understood neither literally nor as merely loose talk, but as metaphors that adapt our conceptual repertoire to new circumstances under pressure of conceptual needs. Drawing on the Strawson-Kant tradition on imagination and concept-application, inferentialist semantics, Bermudez’s theory of rational framing, and Gentner’s structure-mapping account of analogy, it develops a four-part framework: inferential transfer rather than referential transfer, the aptic normativity of conceptual needs, the empirical evaluation of transferred inferences through mechanistic interpretability, and the career of AI metaphors from novelty toward possible conventionalization. The result is a shift away from all-or-nothing disputes about the literal applicability of cognitive terms to AI toward graded and empirically tractable questions about which inferential transfers are apt, revealing why our cognitive vocabulary for AI is evaluatively and practically consequential.

AI, large language models, understanding, conceptual adaptation, aptic normativity, mechanistic interpretability

Re-Engineering the Concept of Understanding for AI

With Pierre Beckmann.

Argues that the concept of understanding needs to be re-engineered for artificial cognition in a way that is empirically informed by mechanistic interpretability research and theoretically informed by a grasp of the functions of the concept.

AI, large language models, understanding, conceptual engineering, mechanistic interpretability, functions of concepts

Why We Care about Understanding: Competence through Predictive Compression

With Pierre Beckmann.

Offers a unifying account of understanding by reverse-engineering the function of both the state and the concept. Argues that we care about understanding because it grounds robust competence. Our concept of understanding evolved as an efficient proxy to track this elusive property, allowing us to identify who to trust and learn from. This highlights the sociality of understanding and how it shapes the character of human understanding. Understanding is the result of convergent pressures to predict the world using cognitive models that are not only accurate, but also compressed enough to be stored, demonstrated, and transmitted.

understanding, functions of concepts, conceptual reverse-engineering, testimony, AI, explainable AI

Conceptual Engineering

In Metzler Handbuch Analytische Philosophie. Hans-Johann Glock, Christoph Pfisterer and Stefan Roski (eds.). Stuttgart: Metzler.

Conceptual engineering reorients analytic philosophy from the descriptive analysis of existing concepts to the normative task of assessing and improving representational devices to better serve our theoretical and practical purposes. This entry traces the method’s intellectual genealogy from Rudolf Carnap’s explication and pragmatist reconstruction to the contemporary ‘functionalist’ and ‘ameliorative’ frameworks championed by Haslanger, Simion, and Kelp. It concludes by examining the discipline’s current ‘applied turn,’ surveying how recent scholarship from 2024 to 2026 has operationalized these methods to address concrete problems in social ontology, artificial intelligence, and medicine.

conceptual engineering, explication, amelioration, metaphilosophy, Carnap

Explication or Amelioration? Carnapian Clarification as the Normative Basis for Conceptual Engineering

The Monist. Special issue on Explication and Conceptual Engineering.

As conceptual engineering fractures into explication pursuing exactness and amelioration pursuing justice, the field risks losing its focus. I argue that unifying these projects requires retrieving a crucial insight from Rudolf Carnap: that attempts to improve concepts must start with the preliminary stage of practical clarification. However, Carnap’s account of clarification in terms of predictive proficiency remains normatively inert and biased towards exactness. I expand it into a normative diagnosis of the needs underpinning a concept’s inferential structure. This reveals whether properties like vagueness are flaws that need fixing or features worth preserving.

conceptual engineering, explication, amelioration, conceptual needs, vagueness, Carnap

Forms of Understanding in LLMs: A Crisis for Comprehension?

In The Routledge Handbook of Philosophy of AI. Leonard Dung, Guido Löhr, and Chris Ranalli (eds.). London: Routledge. Forthcoming. With Pierre Beckmann.

AI, large language models, understanding

Naturalizing Minds: Genealogies of Thought in Hume and Nietzsche

In Hume and Nietzsche. Peter Kail and Paolo Stellino (eds.). Oxford: Oxford University Press.

Claims that once we recognize the genealogical form taken by Hume’s and Nietzsche’s methodological pragmatism, we can see how both manage to avoid cruder views that identify the meaning, truth, or value of things with their effects.

genealogy, naturalism, pragmatism, truth, early modern philosophy, Hume, Nietzsche

Needs of the Mind: How Aptic Normativity Can Guide Conceptual Adaptation

Philosophical Studies. 2026. doi:10.1007/s11098-026-02511-3

The article offers an account of “needs of the mind” in terms of a distinctively aptic normativity–a normativity of fittingness. After reconstructing the history of different conceptions of needs and their gradual subjectivization, the article focuses on conceptual needs and argues that they register a cognitive privation that goes beyond a shortage of words, marking a mismatch between our conceptual repertoire and our situation that reorients conceptual engineering from detached amelioration to situated adaptation. This makes a needs-first approach uniquely suited to guiding conceptual adaptation in times of technological disruption.

conceptual needs, aptic normativity, conceptual adaptation, conceptual engineering

Une normativité sans histoire ? Foucault, Engel et la normativité de la vérité

Invited contribution to Dialogue : Revue canadienne de philosophie

By shielding the concept of truth from Foucauldian historicism, Pascal Engel ends up leaving the “virtues of truth” even more exposed to Foucault’s negative genealogy. This article proposes a more ambitious reading of the positive genealogy of these virtues, demonstrating that cultivating accuracy and sincerity as intrinsic values is a functional necessity rather than a historical accident. Vindicating these dispositions’ status as virtues provides a more robust defence against both Foucauldian cynicism and contemporary indifference to truth.

truth, truthfulness, genealogy, vindicatory genealogy, genealogical debunking, Bernard Williams

Mechanistic Indicators of Understanding in Large Language Models

Philosophical Studies. 2026. With Pierre Beckmann. doi:10.1007/s11098-026-02513-1

Draws on detailed technical evidence from research on mechanistic interpretability (MI) to argue that while LLMs differ profoundly from human cognition, they do more than tally up word co-occurrences: they form internal structures that are fruitfully compared to different forms of human understanding, such as conceptual, factual, and principled understanding. We synthesize MI’s most relevant findings to date while embedding them within an integrative theoretical framework for thinking about understanding in LLMs. As the phenomenon of “parallel mechanisms” shows, however, the differences between LLMs and human cognition are as philosophically fruitful to consider as the similarities.

AI, large language models, mechanistic interpretability, understanding, world models, explainable AI

World Modeling in Transformers

arXiv 2609.21748. 2026. With Pierre Beckmann and André Freitas. doi:10.48550/arXiv.2609.21748

Shows that behavioral failures can make a transformer appear to lack a world model even when it has learned faithful representations of its environment. In TaxiGPT, a transformer trained on random walks through Manhattan whose failures had been read as evidence of an incoherent internal map, mechanistic analysis and causal interventions reveal a model that represents intersections and streets, tracks its position, and navigates with a goal compass; its failures stem from interference between superposed intersection features, whose consequences are limited by an “affordance packing” that groups intersections with the same legal moves. Proposes mechanistic indicators for comparing models, which show world-modeling capacities emerging at different stages of training, and argues for a shift from asking whether a model has a world model to mechanistically studying its world modeling.

AI, large language models, world models, mechanistic interpretability

Can AI Rely on the Systematicity of Truth? The Challenge of Modelling Normative Domains

Philosophy & Technology 38 (34): 1–27. 2025. doi:10.1007/s13347-025-00864-x

Argues that the asystematicity of normative domains, stemming from the plurality, incompatibility, and incommensurability of values, poses a challenge to AI’s ability to comprehensively model these domains and underscores the indispensable role of human agency in practical deliberation.

AI, large language models, systematicity, truth, value conflict, AI ethics, deliberation

Doing History Philosophically and Philosophy Historically

With Marcel van Ackeren. In Bernard Williams on Philosophy and History. Marcel van Ackeren and Matthieu Queloz (eds.), 14–30. Oxford: Oxford University Press. 2025. doi:10.1093/9780191966361.003.0003

Distinguishes four different connections between philosophy and history. (1) Philosophy cannot ignore its own history in the way that science can. (2) When engaging with philosophy’s history primarily to produce history, one still has to draw on philosophy. (3) Even doing history of philosophy philosophically—that is, primarily to produce philosophy—requires a keen sense of how historically distant from us past philosophers were, because the point of reading them is to confront something different from the present. (4) Systematic philosophy itself needs to be done historically, engaging not necessarily with its own history, but with that of the concepts it seeks to understand.

historiography, metaphilosophy, Bernard Williams

Explainability through Systematicity: The Hard Systematicity Challenge for Artificial Intelligence

Minds and Machines 35 (35): 1–39. 2025. doi:10.1007/s11023-025-09738-9

Offers a framework for thinking about “the systematicity of thought” that distinguishes four senses of the phrase, defuses the alleged tension between systematicity and connectionism that Fodor and Pylyshyn influentially diagnosed, and identifies a “hard” form of the systematicity challenge that continues to defy connectionist models.

AI, explainable AI, systematicity

Internalism from the Ethnographic Stance: From Self-Indulgence to Self-Expression and Corroborative Sense-Making

The Philosophical Quarterly 75 (3): 1094–1120. 2025. doi:10.1093/pq/pqae051

Argues that Bernard Williams’s internalism about reasons is the philosophical underpinning of his liberalism, and that it needs to be understood in relation to his later work on the normativity of genealogical explanation and the ethnographic stance, where we imaginatively inhabit a conceptual and motivational perspective without endorsing it.

internalism, ethnographic stance, liberalism, genealogy, Bernard Williams

On the Fundamental Limitations of AI Moral Advisors

Philosophy & Technology 38 (71): 1–4. 2025. Invited commentary. doi:10.1007/s13347-025-00896-3

Argues that while the asystematicity of truth militates against the personalization of AI moral advisors, it also imposes limitations on generalist AI moral advisors.

AI, AI ethics, AI advisors, large language models, systematicity, deliberation

Williams’s Debt to Wittgenstein

In Bernard Williams on Philosophy and History. Marcel van Ackeren and Matthieu Queloz (eds.), 283–316. Oxford: Oxford University Press. 2025. With Nikhil Krishnan.

Argues that several aspects of Bernard Williams’s style, methodology, and metaphilosophy can be interpreted as evolving dialectically from those of Wittgenstein.

metaphilosophy, Bernard Williams, Wittgenstein

Defending Genealogy as Conceptual Reverse-Engineering

Analysis 84 (2): 385–400. 2024. Symposium on my The Practical Origins of Ideas. By invitation. doi:10.1093/analys/anad010

Responds to commentaries by Cheryl Misak, Alexander Prescott-Couch, and Paul Roth.

genealogy, pragmatic genealogy, conceptual reverse-engineering, naturalism

Précis of The Practical Origins of Ideas

Analysis 84 (2): 341–344. 2024. Symposium on my The Practical Origins of Ideas. By invitation. doi:10.1093/analys/anad011

Summarizes my book for a symposium in Analysis.

genealogy, pragmatic genealogy, conceptual reverse-engineering

Debunking Concepts

Midwest Studies in Philosophy 47 (1): 195–225. By invitation. 2023. doi:10.5840/msp2023111347

Argues that the debunking of concepts should extend beyond assessing their epistemological merits to include their evaluation on moral, social, and political grounds, based on their societal functions and effects.

genealogy, genealogical debunking, conceptual ethics, ideology critique, functions of concepts

Making Past Thinkers Speak to Us Through Pragmatic Genealogies

In Historiography and the Formation of Philosophical Canons. Sandra Lapointe and Erich Reck (eds.), 171–191. New York: Routledge. 2023. doi:10.4324/9781003184294-9

Instead of treating Hobbes and Hume as answering the same questions we ask today, this article proposes that we start from the practical predicaments their political concepts addressed in their own time. Hume’s account of property and Hobbes’s account of sovereign power are reconstructed as historically local, yet structurally revealing, responses to predicaments—over conflict, security, and cooperation—that still structure our political life.

genealogy, pragmatic genealogy, historiography, early modern philosophy, property, social cooperation, Hume

Conceptual Engineering and the Politics of Implementation

Pacific Philosophical Quarterly 103 (3): 670–691. 2022. With Friedemann Bieber. doi:10.1111/papq.12394

Argues that how much control we have over conceptual change is itself something we can control, and while some domains require the institutionalization of the power to enforce conceptual innovations, because there are strong practical pressures to coordinate on a single harmonized technical terminology, there are also liberal and democratic rationales for making conceptual engineering hard to implement by default.

conceptual engineering, conceptual change, liberalism, power

Function-Based Conceptual Engineering and the Authority Problem

Mind 131 (524): 1247–1278. 2022. doi:10.1093/mind/fzac028

Identifies a central problem for conceptual engineering—the problem of establishing the authority of engineered concepts—and argues that this problem cannot generally be solved by appealing to increased precision, consistency, or other theoretical virtues. Solving the problem requires engineering to take a functional turn and attend to the functions of concepts. This also helps us alleviate Strawsonian worries about changes of topic.

conceptual engineering, functions of concepts, authority, theoretical virtues

Genealogy, Evaluation, and Engineering

The Monist 105 (4): 435–51. By invitation. 2022. doi:10.1093/monist/onac010

Argues that genealogical explanations can be used to evaluate and improve conceptual practices, taking as an example the demand for conceptual innovation around notions of legitimacy created by the increasing power of international institutions.

genealogy, conceptual engineering, conceptual ethics, legitimacy, international law, ideology critique

Nietzsche’s Conceptual Ethics

Inquiry 66 (7): 1335–1364. Proceedings of the International Society of Nietzsche Studies. 2023. doi:10.1080/0020174X.2022.2164049

While Nietzsche appears to engage in two seemingly contrary modes of concept evaluation—one looks to concepts’ effects, the other to what concepts express—this article offers an account of the expressive character of concepts which unifies these two modes and yields a powerful approach to practical reflection on which concepts to use.

conceptual ethics, revaluation of values, genealogy, naturalism, pragmatism, cultural critique, Nietzsche

The Essential Superficiality of the Voluntary and the Moralization of Psychology

Philosophical Studies 179 (5): 1591–1620. 2022. doi:10.1007/s11098-021-01720-2

Argues that the notion of the voluntary is an essentially superficial notion that does important work on the condition that we do not try to metaphysically deepen it, and that attempts to deepen it illustrate a problematic tendency to warp our conception of the mind under pressure from moral aspirations.

voluntariness, responsibility, moralism, free will, functions of concepts, Bernard Williams

Choosing Values? Williams contra Nietzsche

The Philosophical Quarterly 71 (2): 286–307. 2021. doi:10.1093/pq/pqaa026

Highlights enduring epistemic and metaphysical difficulties for any project of evaluating and improving the values we live by, including contemporary work in conceptual ethics and engineering, and argues that attempts to sidestep these difficulties fall prey to “Saint-Just’s illusion”—the mistake of believing that a set of values from one political context can be successfully transplanted into a different political context.

revaluation of values, conceptual ethics, conceptual engineering, conceptual change, political realism, Bernard Williams, Nietzsche

Left Wittgensteinianism

European Journal of Philosophy 29 (4): 758–77. 2021. With Damian Cueni. doi:10.1111/ejop.12603

Focusing on the social and political conceptual practices that Wittgenstein neglected, the paper presents a novel, more dynamic interpretation of Wittgenstein’s model of conceptual change, on which conceptual change becomes intelligible not just as a brute, exogenous imposition on rational discourse, but as endogenous and reason-driven. This counters the socially conservative tendencies of existing interpretations and renders intelligible the possibility of radical critique within a Wittgensteinian framework.

conceptual change, language games, ideology critique, Wittgenstein, Bernard Williams

Nietzsche’s English Genealogy of Truthfulness

Archiv für Geschichte der Philosophie 103 (2): 341–63. 2021. doi:10.1515/agph-2018-0048

Based on various posthumous fragments, the article reconstructs Nietzsche’s little-known early genealogical account of how the value of truth and the cultivation of the virtue of truthfulness originated not from a pure love of truth, but from the practical necessity of social cooperation.

genealogy, pragmatic genealogy, truthfulness, social cooperation, Nietzsche

From Paradigm-Based Explanation to Pragmatic Genealogy

Mind 129 (515): 683–714. 2020. doi:10.1093/mind/fzy083

Why would philosophers interested in the points or functions of our conceptual practices bother with genealogical explanations if they can focus directly on paradigmatic examples of the practices we now have? This paper offers three reasons why the genealogical approach earns its keep and formulates criteria for determining when it is called for.

genealogy, pragmatic genealogy, paradigm-based explanation, functions of concepts, E. J. Craig, Miranda Fricker

How Genealogies Can Affect the Space of Reasons

Synthese 197 (5): 2005–2027. 2020. doi:10.1007/s11229-018-1777-9

Attempts to derive reasons from claims about the genesis of something are often said to commit the genetic fallacy—they conflate genesis and justification. One way for genealogies to side-step this objection is to focus on the functional origins of practices. But this invites a second objection, which maintains that attempts to derive current from original function suffer from continuity failure—the conditions in response to which something originated no longer obtain. This paper shows how normatively ambitious genealogies can steer clear of both problems.

genealogy, pragmatic genealogy, vindicatory genealogy, E. J. Craig, Bernard Williams

Revealing Social Functions through Pragmatic Genealogies

In Social Functions in Philosophy: Metaphysical, Normative, and Methodological Perspectives. Rebekka Hufendiek, Daniel James, and Raphael Van Riel (eds.), 200–218. London: Routledge. 2020. doi:10.4324/9780429435393

This paper argues that state-of-nature stories, read as dynamic models rather than history, can reveal how key normative practices meet collective needs of coordination, conflict-management, and non-domination. Drawing on Hume’s genealogy of justice, Williams’s genealogy of truthfulness, and related work, it shows how concepts like property, knowledge, and testimonial justice underpin social cooperation and political legitimacy. In doing so, it offers social and political philosophers a way to explain both the persistence of ideas and institutions and the grounds on which they can be criticized.

genealogy, pragmatic genealogy, functions of concepts, social cooperation, legitimacy, non-domination, Hume

Genealogy and Knowledge-First Epistemology: A Mismatch?

The Philosophical Quarterly 69 (274): 100–120. 2019. doi:10.1093/pq/pqy041

Timothy Williamson’s knowledge-first epistemology holds that the concept of knowledge is primitive and explanatorily fundamental. This seems to leave little room for attempts to give a genealogical explanation of the concept of knowledge, much less ones that explain the formation of the concept of knowledge in terms of the concept of belief, as E.J. Craig does. Yet I argue that Craig’s genealogy of the concept of knowledge not only is compatible with knowledge-first epistemology, but actually lends succour to it.

genealogy, pragmatic genealogy, knowledge, testimony, E. J. Craig

Nietzsche as a Critic of Genealogical Debunking: Making Room for Naturalism without Subversion

The Monist 102 (3): 277–297. 2019. With Damian Cueni. doi:10.1093/monist/onz010

Argues that contrary to popular belief, Nietzsche’s genealogical method does not seek to subvert by revealing immanent and lowly naturalistic origins—quite the opposite: Nietzsche is a critic of genealogical debunking thus conceived, on the grounds that it threatens to make a universal acid of reflection in a world increasingly disenchanted by scientific advances. Instead, Nietzsche advocates an outlook which makes room for naturalistic understanding and redraws the contrast between vindicatory and subversive genealogies within the space of naturalistic origins.

genealogy, genealogical debunking, vindicatory genealogy, naturalism, Nietzsche

Nietzsches affirmative Genealogien

Deutsche Zeitschrift für Philosophie 67 (3): 429–439. By invitation. 2019. doi:10.1515/dzph-2019-0034

Argues that alongside his well-known critical genealogies, Nietzsche also developed “affirmative genealogies” that are not historically situated. These genealogies investigate the “practical origins” of concepts like justice and truth, showing how they arise instrumentally from fundamental human needs. By presenting these concepts as naturalistically intelligible and practically indispensable, this approach offers an affirmative justification, which I connect to Nietzsche’s later idea of an “economic justification of morality.”

genealogy, pragmatic genealogy, vindicatory genealogy, justice, truth, Nietzsche

The Points of Concepts: Their Types, Tensions, and Connections

Canadian Journal of Philosophy 49 (8): 1122–1145. 2019. doi:10.1080/00455091.2019.1584940

By distinguishing four senses in which concepts might be said to have a “point,” this paper resolves the tension between the ambition of point-based explanations to be informative and the claim—central to Dummett’s philosophy of language, but also to the literature on thick concepts—that mastering concepts already requires grasping their point.

functions of concepts, conceptual ethics, conceptual engineering

Davidsonian Causalism and Wittgensteinian Anti-Causalism: A Rapprochement

Ergo 5 (6): 153–72. 2018. doi:10.3998/ergo.12405314.0005.006

Seeks a rapprochement in the longstanding debate between Davidsonian causalists and Wittgensteinian anti-causalists by arguing that both sides can agree that reasons are not causes, but that intentional explanations are causal explanations.

reasons and causes, Davidson, Wittgenstein

Williams’s Pragmatic Genealogy and Self-Effacing Functionality

Philosophers’ Imprint 18 (17): 1–20. 2018. doi:2027/spo.3521354.0018.017

Reconstructs Williams’s genealogical investigation into the social function of the norms of truthfulness and brings out its social and political implications. Develops an understanding of this “pragmatic” form of the genealogical method which reveals it to be uniquely suited to dealing with practices exhibiting what I call “self-effacing functionality”—practices that are functional only insofar as and because we do not engage in them for their functionality.

genealogy, pragmatic genealogy, truthfulness, self-effacing functionality, naturalism, Bernard Williams

Does Philosophy Have a Vindicatory History? Bernard Williams on the History of Philosophy

Studia Philosophica 76: 137–52. 2017. doi:10.24894/StPh-en.2017.76008

Develops Williams’s suggestion that for philosophy to ignore its history is for it to assume that its history is vindicatory. The paper aims to offer a fruitful line of inquiry into the question whether philosophy has a vindicatory history by providing a map of possible answers to it. It first distinguishes three types of history: the history of discovery, the history of progress, and the history of change. It then suggests that much of philosophy lacks a vindicatory history, for reasons that reflect philosophy’s character as a humanistic discipline.

historiography, metaphilosophy, Bernard Williams

Two Orders of Things: Wittgenstein on Reasons and Causes

Philosophy 92 (3): 369–97. 2017. doi:10.1017/S0031819117000055

Situating Wittgenstein in the causalism/anti-causalism debate in the philosophy of mind, this paper argues that Wittgenstein’s arguments differ from those of his immediate successors; that he anticipates current anti-psychologistic trends; and that he is perhaps closer to Davidson than historical dialectics suggest.

reasons and causes, language games, Wittgenstein, Davidson

Wittgenstein on the Chain of Reasons

Wittgenstein-Studien 7 (1): 105–30. 2016. doi:10.1515/witt-2016-0108

This article examines Wittgenstein’s conception of rationality through the central image of the “chain,” arguing that reasons are defined by their relational role in making actions intelligible rather than by intrinsic properties. The author contends that unlike chains of causes, chains of reasons are necessarily finite and anchored in communal reason-giving practices, meaning that justification inevitably ends at the boundaries of a specific language game. Ultimately, the paper suggests that this finite structure liberates agents from the misleading expectation of infinite justification while simultaneously limiting the reach of reasons to the specific practices that sustain them.

reasons and causes, language games, Wittgenstein