What does "authenticity" mean, in the context of the word being applied to an artificial agent?
We increasingly speak of artificial agents in human terms: they "understand," they "learn," they "decide." It is natural to ask whether such agents might also be "authentic." But this question is less straightforward than it appears. It looks, at first glance, like an empirical question: can we build systems sophisticated enough to qualify? This essay argues that it is fundamentally a conceptual one, about whether the conditions that make authenticity intelligible — finitude, embodiment, anxiety, situated freedom — can coherently be attributed to an artificial system at all. Drawing on both the existential-phenomenological tradition (Heidegger, Sartre, de Beauvoir) and the analytic philosophy of language (Wittgenstein, Ryle, Travis), I contend that applying "authenticity" to an AI is not false but risks being empty: a word detached from the form of life that gave it meaning.
To assess whether authenticity can be applied to an artificial agent, we must first establish what it means when applied to human beings. In Heidegger's existential analytic in Being and Time, authenticity (Eigentlichkeit) is contrasted with inauthenticity, the default mode of human existence. Most of the time, Heidegger argues, we exist in the manner of "the they" (das Man): we do what one does, think what one thinks, care about what one cares about (BT, §27). Our projects, values and self-understanding are inherited from the public world rather than owned as our own. This is not a moral failing. It is a structural feature of how human beings ordinarily exist.
What disrupts this absorption is anxiety (Angst), a mood fundamentally unlike fear. Fear has an object; anxiety does not. In anxiety, the familiar world loses its significance, and we are confronted with the groundlessness of our existence. Heidegger connects this to death — not as a biological event but as the ever-present possibility of no longer being (BT, §§46–53). Death is what individualises: no one can die my death for me. Authenticity, then, is the mode of existence in which I take over my own being in light of my finitude, rather than fleeing into the anonymity of "the they." It is not a permanent state but a repeated act of what Heidegger calls "resoluteness" (BT, §60): owning one's choices against the background of one's mortality.
Sartre radicalises this. If existence precedes essence, if there is no fixed human nature determining what we are, then every moment involves a choice for which we are wholly responsible. Authenticity means acknowledging this radical freedom rather than hiding in "bad faith" (mauvaise foi), the self-deceptive pretence that one's situation is given rather than chosen. The waiter who performs his role too precisely, as though he is a waiter rather than a free being who happens to wait tables, exemplifies bad faith (BN, pp. 59–60). He treats himself as a thing with a fixed nature when he is a being whose nature is to have none.
De Beauvoir adds something that Heidegger and Sartre miss. She retains Sartre's framework but insists that freedom is always situated: exercised in a body, a history, a web of social relations structured by power. One cannot authentically pursue one's own freedom while denying it to others, because freedom only acquires meaning in a world where it is mutually recognised. Her analysis of gender in The Second Sex (pp. 12–29) shows that imposed "essences" — claims about what women fundamentally are — function as instruments of oppression, denying an entire group the existential freedom that defines human existence. This makes authenticity irreducibly relational. It is not enough to confront one's own death or acknowledge one's own freedom. One must exist among others, in a body, within structures of power one did not choose, and act from that situation. This is a separate problem for AI authenticity, independent of finitude or anxiety: an artificial agent has no gender, no race, no embodied position in a social world. It does not encounter others as free beings whose freedom constrains and enables its own. Even a system that could simulate something like anxiety would still lack the situated, relational dimension that de Beauvoir shows authenticity requires.
What emerges from these accounts is not a checklist of properties but a set of structural conditions without which the concept of authenticity cannot get purchase. These include: finitude (existence bounded by death, which gives choices their weight); anxiety (a mood that reveals the groundlessness beneath everyday absorption); freedom (the capacity to negate what one is and choose otherwise); embodiment and situation (existence in a body, among others, within relations of power); and selfhood as an issue (one's own being is something one has to take up, not a given). These are not incidental features of authenticity. They are what make it the kind of phenomenon it is. Remove finitude and there is no urgency to own one's choices. Remove anxiety and there is no disruption of everyday absorption. Remove embodied situation and freedom becomes an abstraction with no content.
If this is what authenticity requires, then the question of whether it applies to artificial agents looks very different from how it looked at the start. It is no longer a question about capability. It is a question about the kind of being an artificial agent is. Current AI systems do not die in any Heideggerian sense. A system can be switched off, deleted or replaced, but none of these constitutes death as Heidegger understands it: the ownmost, non-relational, not-to-be-outstripped possibility that individualises a being. A system that can be copied, restored from backup or run across multiple servers lacks the irreplaceable singularity that death confers on human existence. The "possibility of impossibility," Heidegger's formulation for death (BT, §53), presupposes a being for whom its own being is at stake. It is far from clear that any current or foreseeable artificial system meets this condition.
Similarly, anxiety presupposes a being already absorbed in a world of significance that can then be withdrawn. Heidegger's analysis depends on the prior structure of "being-in-the-world": we encounter things as equipment within projects that matter to us, and anxiety reveals the contingency of this mattering. Dreyfus (1991) has argued, drawing on Heidegger, that disembodied AI systems cannot genuinely be in a world in this sense because they lack the embodied coping skills through which significance is constituted. Without a world to lose, there is nothing for anxiety to disclose.
Bad faith, too, requires a particular structure of selfhood. Sartre's analysis depends on a being that is simultaneously facticity (what it is) and transcendence (the capacity to negate and surpass what it is). An artificial agent, as currently conceived, does not stand in this kind of relation to itself. It has no facticity to flee from and no transcendence to deny. It processes inputs and generates outputs, but the reflexive structure of bad faith — lying to oneself about one's own freedom — requires a form of self-relation that does not obviously obtain.
The phenomenological tradition shows that the structural conditions for authenticity do not obtain in artificial agents. But there is an even deeper problem, and it comes from a different direction entirely. The analytic philosophy of language suggests that the difficulty is not just that AI lacks what authenticity requires, but that the word "authenticity" itself may lose its grip when applied outside the human context that gives it meaning.
The later Wittgenstein argues that words derive their meaning from their use within "language-games" embedded in shared "forms of life" (PI, §§23–25). If we ask what "authenticity" means, we must look at the practices in which the word has its home: confronting mortality, resisting social conformity, taking responsibility for one's situated freedom. Travis's occasion-sensitivity thesis deepens this point. The same sentence can express different thoughts on different occasions of use; what a word contributes to a thought depends on how it is being applied. "Authentic," applied to a human being who faces death and lives within structures of power, may not express the same thought when applied to a system that does neither. It is not that the application is false. It is that the word, in this new context, may fail to determine any thought at all.
Ryle's concept of a category mistake is also pertinent. To attribute authenticity to an AI may be like asking whether the number seven is heavier than the number three: not wrong, but confused — applying a concept to the wrong logical category. Authenticity belongs to the category of beings whose existence is an issue for them, who are "always already" in a world, who face death and freedom. Applying it to a system that belongs to a different category does not produce a false claim. It produces something closer to nonsense.
Brandom's inferentialism provides a further angle. If meaning is constituted by inferential roles within practices, then "authenticity" carries its meaning through connections to concepts like mortality, anxiety, responsibility and situated freedom. Strip away those connections, as one must when applying the term to a system that lacks these features, and the word retains its surface form but loses its inferential substance. It becomes, in Brandom's framework, a noise without a normative role.
There is a real question here about whether a system with a non-restorable architecture, genuinely embedded in a social world, with something like preferences it could reflect upon, might develop functional analogues of finitude, anxiety and self-relation close enough to count. Could that be enough? It deserves serious consideration. But notice what the objection asks us to do: reverse-engineer the conditions of authenticity as a checklist of features and then build a system that ticks the boxes. That move itself reveals the problem. Authenticity, as the existentialists describe it, is not a set of properties. It is a way of existing, structurally tied to the kind of being we are. This is not a claim about current technological limitations that future engineering might overcome. It is a claim about what authenticity is: a phenomenon that belongs to beings who are born, who die, who exist among others in bodies they did not choose. A system designed to replicate its features would be doing something closer to performing authenticity than living it. And the question would remain: are we preserving the philosophical content of the word, or borrowing its prestige while emptying it of meaning?
Two traditions — Continental existential phenomenology and analytic philosophy of language — converge on the same answer from different directions. From the phenomenological side, authenticity requires structural conditions (finitude, anxiety, situated freedom, mutual recognition) that do not obtain for artificial agents as currently conceived. From the philosophy of language, applying the word "authenticity" outside the form of life in which it has its home risks producing not a false proposition but an empty one. The question is not whether AI can be authentic. It is whether the word "authentic," applied to an AI, still means anything at all. The answer, I think, is that it does not. And that matters — because the temptation to apply it anyway tells us something important: not about the capacities of artificial agents, but about how readily we extend our deepest concepts beyond the existential conditions that gave them life, and how much we stand to lose when the words we live by are hollowed out.
Bibliography
Brandom, R. (1994) Making It Explicit: Reasoning, Representing, and Discursive Commitment. Cambridge, MA, Harvard University Press.
De Beauvoir, S. (1997) The Second Sex. Trans. Parshley, H. M. London, Vintage.
Dreyfus, H. L. (1991) Being-in-the-World: A Commentary on Heidegger's "Being and Time", Division I. Cambridge, MA, The MIT Press.
Heidegger, M. (1962) Being and Time. Trans. Macquarrie, J. & Robinson, E. Oxford, Blackwell.
Ryle, G. (1949) The Concept of Mind. London, Hutchinson.
Sartre, J.-P. (1958) Being and Nothingness. Trans. Barnes, H. E. London, Methuen.
Travis, C. (2008) Occasion-Sensitivity: Selected Essays. Oxford, Oxford University Press.
Wittgenstein, L. (1953) Philosophical Investigations. Trans. Anscombe, G. E. M. Oxford, Blackwell.
This essay was written for my Contemporary Philosophy and AI course at Imperial College London. Thank you Prof. Cosmin Badea for the feedback and for inspiring in me a love for wisdom.