By Jana Baum (Center for Advanced Internet Studies (CAIS), Bochum and University of Leipzig)

The question of how given data can be processed autonomously and intelligently—or, in a certain sense, “meaningfully” or “correctly”—is not only central to research on artificial intelligence (see Russel & Norvig 2021), but also to the underlying field of cognitive science regarding human intelligence (Chalmers et al. 1992). The conception of AI-based systems as well as human cognition, in relation to concepts such as understanding, perception, and learning, thus share a common assumption: knowledge is (the result of) data processing. This basic assumption, however, remains philosophically underdetermined: What exactly does it mean to conceive of intelligence as data processing, and what implicit concepts of rationality, meaning, and world-reference are already presupposed therein?

This paper proposes to reframe this question by drawing on John McDowell (1994). Although McDowell does not belong to classical philosophy of technology, his development of Sellars (1956) critique of the “myth of the given” offers a starting point for the discussion. If, as McDowell argues, knowledge does not arise from raw, non-conceptual data (to think that it does is the “myth”) but is always already situated within the “space of reasons”, then it becomes questionable to what extent data-driven AI systems can be described as bearers of knowledge or intelligence at all. Rather, their mode of operation seems to rest on a tacit, philosophical assumption: that raw data can function as epistemically primary material without itself already being embedded in normative contexts of justification.

In this light, a twofold thesis will be developed. First, it can be shown that AI-based models are bound to the myth of the given, insofar as data is treated as a preceding, epistemically grounding instance. Second, McDowell’s philosophy allows us to critique this assumption without falling into a blanket rejection of technological developments. Instead, one could discuss to what extent AI should be understood as part of an expanded practice of justification in which human and technical operations are intertwined without technical systems themselves participating in the realm of reasons.

This paper thus positions itself at the intersection of epistemological and ontological questions in the philosophy of technology. It builds on existing debates regarding the status of AI—for instance, as a tool, infrastructure, or epistemic agent—while simultaneously demonstrating that even approaches in analytical philosophy seemingly distant from technology provide relevant resources for its reevaluation. In this way, AI is understood not as a rupture but as an occasionto renegotiate fundamental concepts such as data, experience, and rationality within various philosophical traditions.

By Claudia Montanaro (University of Amsterdam)

When engineers, philosophers, or cognitive scientists debate whether LLMs understand or produce ‘meaning’, they seldom share a common semantic ground, yet the same word circulates across disciplines as if they did. This study takes this polysemy as its starting point to trace the divergent theoretical commitments that a seemingly unified vocabulary of capability claims leaves hidden. Drawing on a systematic literature review and qualitative coding of 53 academic manuscripts about LLMs, we map how ‘meaning’ is conceptualized across disciplines and how these approaches inform claims about LLMs’ ability to understand, produce, and internally represent meaning. We identify five broad approaches: Meaning as Reference, Patterns of Relations, Pragmatics, Embodied Engagement, and Emergent Process. Cross-analyses show that no single framework dominates, as approaches circulate across fields while exhibiting characteristic concentrations, and opposing capability claims systematically track distinct theoretical commitments. We argue that capability debates are theory laden and that latent polysemy helps explain oscillations between deflationary “just next token prediction” framings and portrayals of LLMs’ meaning-making potential in interaction. Making background conceptions of ‘meaning’ explicit reveals how they condition, no less than benchmark results, interpretations of what LLMs are and do, and in doing so opens space for more reflexive interdisciplinary dialogue.

By Sofía Alcaine (University of Bonn)

This paper examines the conditions under which artificial intelligence – particularly, large language models (LLMs) – can be said to participate in the kind of interpretive dialogue that Gadamerian hermeneutics identifies as the incessant task constituting human existence. Drawing on Gadamer’s account of authentic experience and dialogue, his hermeneutic rehabilitation of Platonic dialectics and Aristotelian ethics, and his conception of Bildung as a lifelong process of self-formation, this paper argues that LLMs cannot be considered authentic participants in dialogue, and calls for the development of a hermeneutics of AI-mediated self-formation.

The paper begins by defining core elements of contemporary hermeneutics, such as experience, conversation, horizon, and prejudice, and providing a hermeneutic explanation of why knowledge-seeking is a constitutive necessity in human life. It then takes a closer look at the differences between understanding in the full sense of the term and the semblance of understanding enabled by interaction with LLM chatbots by questioning whether question-asking, orientation toward a Sache, and factical situatedness are even applicable to LLMs.

The second part of the paper analyzes the fundamental role of communication in human life as revealed by its etymology – composed of co- (shared) and munus (duty or obligation) –, framing communication as a shared obligation among human beings. This understanding of communication is then contrasted with the exchange that occurs between a human being and an LLM chatbot, focusing especially on the reception of probabilistically generated token sequences that are optimized for plausibility and satisfaction to the user. This analysis grounds the distinction between authentically communicative logos and tokenized logos.

Finally, drawing on a hermeneutic account of eudaimonia and on Gadamer’s revitalization of Bildung, the paper highlights the need for a hermeneutics of AI-mediated self-formation. Such a framework serves both as a defense mechanism against the anthropomorphizing illusions fostered by LLM design and as a means of encouraging responsible AI use by situating these technologies strictly within their ontological and epistemic limits. The goal of such a hermeneutics is to make us deeply aware of the existential necessity of inter-human dialogue, and so to encourage a consistent return to the finite, jagged, vulnerable, and laborious conversation and coexistence with others, through whose otherness we develop our humanity.

Ultimately, this paper contributes to a broader philosophical anthropology of our emerging AI-assisted form of existence by highlighting that understanding and communication are not mere skills to be delegated to machines, but rather the fundamental modes of being in which human life unfolds.