Why can't artificial language contain the truth? A focus on Foucault's and Heidegger's discussions

The paper situates the rise of AI conversational systems—from early theories of speech acts to contemporary data-driven NLP—and the rapid adoption of models like ChatGPT within a post-truth context marked by misinformation, bias, and growing public trust in AI outputs. It poses the central problem of whether AI can speak truth and advance human discourse or instead reduce human language and thinking. Motivated by documented issues such as hallucination, bias, and the unclear political valence of AI outputs, the study aims to assess the essence and limits of AI language by contrasting it with human discourse understood via Foucault’s and Heidegger’s frameworks. The authors hypothesize that AI language, optimized for information transmission and logical regularities, cannot capture the complex, embodied, and ethical dimensions of human truth-telling.

The paper reviews the historical development of AI from Turing’s vision and the Dartmouth Conference to the shift from symbolic systems to data-driven machine learning and deep learning, alongside the evolution of conversational agents (McTear, 2020; Toosi et al., 2021). It surveys recent analyses of ChatGPT’s capabilities and limitations across domains such as education, academia, and politics, noting concerns about rigor, reliability, and the role of filters in democratic communication (Mahyoob et al., 2023; Ronan & Schneider, 2023; Kreps & Jakesch, 2023; Singer et al., 2023). It highlights issues of hallucination, bias, and under-informativeness (IBM definition of hallucination referenced; Heikkilä, 2023; Zhang et al., 2023), and the increasing social trust placed in AI relative to human authorities. Complementary work on language, subjectivity, and discourse (e.g., Koshy, 2024; Besley, 2015) frames broader concerns about text-centered knowledge and post-structural critiques. These sources motivate a philosophical inquiry into AI’s discourse relative to truth.

This is a philosophical and conceptual analysis employing close engagement with Michel Foucault’s and Martin Heidegger’s accounts of language, discourse, subjectivity, embodiment, and truth. The authors reconstruct Foucault’s analysis of classical language and the four modalities of resemblance (convenientia, aemulatio, analogy, sympathies), his critique of text-centered knowledge and discourse, and his considerations of parrhesia, spirituality, and care of the self. They also reconstruct Heidegger’s phenomenology of Dasein, including the fourfold structure of discourse (expressed words, significations, communication, and the speaker’s existence), the distinction between idle talk (das Man) and authentic speech grounded in anxiety, conscience, and resoluteness. These frameworks are then applied to analyze AI as a speaking entity, contrasting AI’s text-based, rule-regularized outputs with human discourse’s embodied, ethical, and world-disclosive dimensions. No empirical data were collected or analyzed; the approach is theoretical and interpretive.

• Foucaultian perspective: Human language, especially in its classical forms, aims beyond referential designation to convey complex, layered meanings and to induce empathy through resemblance structures (convenientia, aemulatio, analogy, sympathies). Discourse functions as a vessel (with folds, plis) that can hold multifaceted meanings tied to historicity, subjectivity, and care of the self. AI language, being logically regularized and text-centered, linearizes and excludes many meanings and cannot internalize embodied relationships or spirituality. Thus, it cannot achieve the empathetic and expansive functions of human language.
• Heideggerian perspective: Discourse requires not only expressed words and communication but also significations grounded in the speaker’s existence. Everyday ‘idle talk’ (of the ‘they’) lacks authentic disclosure. AI can transmit words but, lacking embodiment, anxiety, conscience, and self-presence, cannot disclose beings as they are or reveal its own existence; its outputs correspond to idle talk. Learning from internet-scale data reinforces conformity to the ‘they’ rather than authenticity.
• Implications for truth and the post-truth era: AI’s tendencies toward hallucination, ambiguity, incompleteness, bias, and under-informativeness risk replacing human idle talk with an even more elaborate world of empty words. Even hypothetically embodied or advanced AI would face limits: a manufactured body coupled to software would not instantiate the unitary body–mind required for genuine sympathies, anxiety, and authentic truth-telling. Therefore, AI cannot contain ‘true and complete words’ or fully represent human/world existence.
• Societal impact: Growing reliance on artificial language may reduce human engagement in language games that generate new meanings, narrowing moral and behavioral norms and weakening plural, interactive discourse.

The analysis addresses the central question—whether AI can tell the truth in a humanly meaningful sense—by showing that truth for both Foucault and Heidegger entails more than factual or logical correctness. It involves embodied subjectivity, ethical self-formation, historicity, and authentic disclosure. AI’s discourse, rooted in text-oriented regularities and public linguistic averages, lacks the existential and affective grounding necessary for truth-telling as parrhesia or world-disclosure. Consequently, increased social dependence on AI risks amplifying post-truth conditions: persuasive but empty discourse may displace authentic human speech, further complicating democratic deliberation and moral judgment. The findings underscore the need to preserve and cultivate human discourse practices while reframing AI language research to grapple with metaphor, context, amphibology, and the non-linear, multi-layered aspects of meaning that current systems exclude.

AI excels at information scale and transmission but cannot, in its current or foreseeable forms, contain truth in the sense articulated by Foucault and Heidegger. Foucault’s discourse-as-vessel suggests AI cannot encompass the rich folds of meaning tied to empathy, historicity, and subjectivity; Heidegger’s analysis shows AI’s speech remains idle talk lacking the speaker’s embodied existence, anxiety, conscience, and resolute authenticity. In the post-truth era, AI is poised to lead discourse with greater authority, risking further displacement of human truthful speech. Two tasks follow: (1) prioritize and protect human natural language and discourse capable of conveying truth, and (2) guide AI language development beyond mere information transmission to reflect more facets of human language, while users remain aware of the irreducible gaps AI cannot fill. Further research should address user practices and perceptions of the external world to inform responsible AI language design and use.

The study is theoretical and interpretive, relying on Foucault’s and Heidegger’s frameworks without empirical data or experiments. Its conclusions are bounded by these philosophical lenses and by the current state of AI systems. While the authors consider hypothetical advances toward strong or embodied AI, they argue fundamental limits would remain; however, this claim is not empirically tested. The work does not measure real-world discourse effects or user behaviors; it calls for future research on user practices and contextual meaning.