We worry that AI will spread misinformation, but the real problem runs deeper: AI is incapable of telling the truth at all. Philosophers Bun-Sun Kim and Hongjoon Jo draw on Foucault and Heidegger to argue that humans speak truthfully because our finite, mortal existence is at stake in every word we say. AI, lacking a body, anxiety, or a conscience, risks nothing — it just recombines the internet's idle talk into statistically plausible text, with no self to reveal. Outsourcing our communication to AI doesn't just degrade information; it traps us in an endless loop of crowd-sourced mimicry, and threatens our capacity for genuine thought.

ChatGPT can answer complex questions and even seem to hold conversations. But can it tell the truth?

In an era where AI can answer virtually any human question, we must examine whether AI language can truly contain truth. Since the Dartmouth Conference of 1956, we’ve witnessed dramatic technological evolution—from the AI Winter of the 1970s and 80s to today’s sophisticated language models like ChatGPT that generate remarkably human-like text. As we increasingly delegate communication to artificial, rather than human, entities, a fundamental question emerges: Can AI’s artificial language capture the essence of truth conveyed by human discourse?

This article argues that, despite AI’s impressive capabilities, artificial language fundamentally cannot contain truth in the way that human language does. Drawing on the philosophical insights of Michel Foucault and Martin Heidegger, we examine the intrinsic limitations of AI discourse. Foucault reveals how language serves as a vessel for complex meanings, beyond mere information transfer, while Heidegger’s analysis of authentic discourse exposes AI’s inability to stake its existence on its words. In the post-truth era, recognizing these limitations becomes crucial as we navigate between human and artificial communication.

The folds of truth through Foucault’s lens

The identity of things, the fact that they can resemble others and be drawn to them, though without being swallowed up or losing their singularity—this is what is assured by the constant counterbalancing of sympathy and antipathy.

Michel Foucault, The Order of Things

Truth is formed through the process of the speaking subject caring for oneself. He maintains that truth does not originate from truth-value, but rather from the meaning one confers upon oneself in relation to oneself. For instance, consider the concept of pain. For example, a doctor might define pain through a standardized description of pain. But we don’t necessarily think that’s the truth about the pain that I experience. We define the pain we think of in our own language, but we leave room for the assumption that the pain I have experienced may or may not be the same as that of others.

So, when we ask AI the same question, can it also leave room in its language for sympathy, just as we have prepared? I believe it’s worth discussing this issue before relying on AI for the answer. To approach this question, let us turn to Foucault’s perspective.

Foucault analyzes the use of language in Western culture and traces how its modalities have transformed across historical epochs. The linguistic process that gives rise to empathy, according to this view, does not merely serve to designate a referent at a simple level; rather, it operates to convey meaning at a more intricate and relational level.