AI

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

The Invented Inventor: Adapting Intellectual Property to Generative AI

Under review

As AI increasingly drives discovery, the concept of inventor faces severe strain. Recent judicial decisions, such as the Swiss Federal Administrative Court’s 2025 DABUS ruling, expose a deepening tension: courts demand intellectual creation by a natural person even as human contributions to AI-assisted discovery become increasingly nominal. This paper approaches the resulting tension from the standpoint of political philosophy rather than jurisprudence: the strain AI places on the concept of inventorship is too fundamental to be resolved by interpretative methods taking existing conceptual architectures for granted. Inspired by Hume’s genealogy of property, the paper reconstructs the historical “need matrices” that forged the concept of inventorship, tracing its evolution from Venetian guild economics through Romantic genius ideology to corporate R&D. This reveals the concept to be an overburdened bundle serving four social functions: incentivising innovation, disseminating knowledge, legitimating monopolies, and resolving priority disputes. It also clarifies the mismatch between the concept and the emerging realities of AI-driven discovery. To resolve this mismatch, we must disaggregate the concept of inventorship and develop specialised conceptual resources for each of these functions. If we invented the notion of inventor to perform certain functions, we can reinvent it to perform them better.

AI, intellectual property, property, conceptual adaptation, conceptual needs, genealogy, Hume

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

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

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

Dropping Anchor in Rough Seas: Co-Reasoning with Personalized AI Advisors and the Liberalism of Fear

Philosophy & Technology 38 (170): 1–7. 2025. Invited commentary. doi:10.1007/s13347-025-01006-z

A political critique of personalized AI advisors through the lens of the liberalism of fear. Highlights the asymmetries of power involved and argues that personalization risks stabilizing domination by translating structural injustices into individualized aspirational challenges. Three political constraints on personalized AI are then proposed: the priority of non-domination, the public contestability of operative norms, and the recognition of non-personalizable civic burdens.

AI, AI ethics, AI advisors, liberalism, non-domination, power, deliberation

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

Kein Sicherheitsnetz der Wahrheit: Warum Normativität für LLMs schwierig bleibt

meta(φ) 13 (1): 51—89. 2025.

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

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