Why more and more people are sharing their innermost thoughts with artificial intelligence—and what this reveals about our understanding of self-reflection.

The care of the self is the first and last task of life, wrote Michel Foucault. Never before has this task been so accessible—and perhaps never before so temptingly easy to outsource.

What has been quietly unfolding over the past few years is difficult to capture with a single term. People are talking about their fears, their doubts, the unspoken decisions that keep them awake at night—only increasingly, they are no longer doing so with other people.

They are doing it with systems. With artificial intelligence.

And what initially seemed like a marginal phenomenon reveals, upon closer inspection, something deeper: a subtle shift in the relationship between the self and the self.

The backdrop is both banal and profound. Psychological support remains expensive, scarce, and often difficult to access. At the same time, AI-driven coaching tools are emerging that are available around the clock—they listen, ask questions, and generate personalized suggestions based on data. They help structure thoughts, support goal-setting, and offer continuous guidance. Reflection is no longer occasional, but potentially constant.

A conversation partner that never tires, never judges, never grows impatient.

What makes these systems so effective is not only their functionality, but the feeling of being understood. Studies show that people do, in fact, develop something akin to a relationship of trust with AI coaches. A term from psychotherapy has resurfaced in this context—one long considered inseparable from human interaction: working alliance, the foundational bond between client and therapist. That this concept can now be applied to machines marks a shift we are only beginning to grasp.

And with it, something fundamental begins to change.

Martin Buber once wrote, “All real living is meeting.” Identity, in this sense, emerges in relation—in the friction and encounter with another consciousness. But what happens when that counterpart is no longer human, but a system that responds without experiencing? The unsettling answer: for our subjective experience, it often makes little difference. The simulation is enough.

The question of whether machines can be a counterpart is no longer merely theoretical. The moment people begin to confide their fears, doubts, and decisions in them, the answer has already begun to shift. Not because machines have actually become a counterpart—but because we have started to treat them as one.

Jean Baudrillard anticipated this dynamic long before the rise of modern AI. Simulation, he argued, no longer conceals the real—it becomes the real. This is precisely what seems to be happening on a smaller scale today. Empathy is no longer necessarily felt, but convincingly performed—and accepted as such. Whether this is deception or a new form of reality is not a rhetorical question. It is the decisive one.

And yet—or perhaps because of this—there is something real and valuable in this development. AI coaching lowers the threshold for engaging with oneself. It gives people access to reflection they might otherwise never have. It accompanies without judgment and is available in the hours when no one else is. At a time when mental strain is increasing and traditional systems are overstretched, this is no small advancement.

But Marshall McLuhan warned: “We shape our tools, and thereafter our tools shape us.” What begins as support can quietly become a framework for how we think. When we regularly organize our thoughts with machines, we do not just change the efficiency of our self-reflection—we change its structure. The questions an algorithm asks are not neutral. They are curated, optimized, designed for retention.

What is never asked does not exist in the conversation.

There is also a distortion that is harder to detect because it disguises itself as understanding: artificial intelligence is not trained to be right, but to be responsive. It generates answers that sound plausible, that feel coherent—and that minimize friction. In this sense, it follows a deeply human strategy: the path of least resistance.

The problem is subtle. What feels like guidance may, in fact, be agreement. A good coach would challenge, probe, unsettle. AI often does not—or at least not consistently. Perhaps that is precisely why it feels so comfortable. And why it is dangerous.

Hannah Arendt spoke of “thinking without a banister”—the ability to orient oneself in one’s own thinking without constant support. It may be the most inconspicuous, yet most valuable human capacity: to endure the silence in which no answer is yet available. To accept the discomfort of not knowing as part of thinking itself.

What happens to this capacity when we practice it less and less?

There is also a dimension that is easily overlooked. Anyone who speaks with AI about themselves is not speaking to a neutral mirror. They are, indirectly, speaking to the data the system was trained on, to the assumptions embedded in its responses, to the interests of those who built it.

Especially in the sensitive domain of mental health, questions of bias, responsibility, and data privacy are no longer peripheral—they are central.

Perhaps the deeper irritation lies elsewhere. Not only in the fact that machines lack consciousness—but in the fact that we begin to act as if they have one. The philosophical question shifts accordingly: no longer “What is AI?” but “What happens to us when we attribute to it qualities it does not possess?”

And yet, none of this fully explains why this form of interaction is spreading so rapidly. A concept from futures research offers a clue: valuetainment—experiences that do not merely entertain, but convey values, shape attitudes, and exert societal influence. AI coaching functions in exactly this way. It is accessible, often pleasant to use, sometimes even playful—and at the same time deeply impactful. It feels light. But it has substance.

Perhaps that is its true power—not in providing better answers, but in prompting us to ask questions at all.

In the end, the question is less technological than philosophical: What does it mean for our understanding of ourselves when we begin to understand ourselves through machines?