Artificial Intelligence as confidant: beyond the idealization of human relationships

The growing use of artificial intelligence as a confidant raises sometimes virulent criticism. It is denounced as a simulacrum, a dependency, an abandonment of “real” human relationships. This discourse rests on an unquestioned premise: human beings would be naturally empathetic and human relationships would constitute the absolute therapeutic ideal. But this view obscures the structural violence that runs through social relations.

The problematic idealization of human relationships

Detractors of the therapeutic use of artificial intelligence build their argument on a seductive premise: AI would systematically go along with the user, creating a problematic complacency, where human relationships would require negotiation and confrontation with otherness. This critique assumes that humans would naturally embody this benevolent otherness, this capacity for authentic empathy that the machine would lack. The therapeutic ideal would therefore reside in human relationships, while exchange with AI would be merely a substitute.

This reasoning reveals, in my view, an excess of confidence in human beings. It postulates their intrinsic perfection, their spontaneous empathy, their innate capacity to respect otherness. This angelic vision of human relationships deserves to be confronted with the empirical reality of our social interactions... As recalled in Raphaël Gaillard’s interview in Le Monde (2025), AI creates “a strong bond, which resembles the therapeutic bond – or ’transference’ in psychoanalysis – while being very comfortable, because it’s a machine that will often go along with you”. But this critique presupposes that human relationships systematically offer better.

However, this discourse completely obscures the symptomatic, violent, and destructive dimensions that obviously run through human relationships. It ignores the holds, moral harassment, systems of domination that structure social relations. These violences do not merely stem from individual psychological dysfunctions, but are inscribed in social structures that organize and legitimate domination.

Domination and violence: the blind spots of humanist discourse

Take the example of incest. Entire families, complete relational systems, cause absolute harm to the children who are victims. The recognition of their suffering, the acknowledgment of their truth would threaten the symptomatic balance constructed by these groups. So, can we seriously claim that “the ideal is always human relationships”? Human beings murder each other, cultivate hatred, perpetuate it, deepen it, seek to justify it, inscribe it in law and even legitimize it in social structures, through institutionalized racism for example.

This destructive dimension runs through history and societies. Humans regularly justify the destruction of other human beings, construct legal and social systems that legitimate oppression. Michel Foucault, in Discipline and Punish (1975), clearly demonstrated how social institutions organize and normalize violence. Pierre Bourdieu, with his concept of symbolic violence developed in Reproduction (1970), showed that domination does not only pass through physical constraint, but is also exercised through invisible, naturalized mechanisms.

Thus, the idealization of human relationships seems to me to be a form of denial. It perhaps serves more to reassure our collective ego (we want to believe that we are superior to machines) than to faithfully describe our relational modes. In my view, the ideal, if there is an ideal, does not reside in human relationships as such, but in autonomous human beings, endowed with critical thinking, confident enough not to obey coercive social systems that harm the individual.

Individual freedom and the common good: beyond a false opposition

I situate myself within Western thought about the world; Eastern thought would develop another paradigm. Within the Western framework, self-confidence constitutes the foundation of autonomy. This confidence is unfortunately inaccessible to the vast majority of human beings. If it were more widespread, if each person’s critical thinking developed more, individual choices would be more conscious, less symptomatic, and would benefit society as a whole.

I know that objections can be raised to this proposition. But no, individual emancipation is not opposed to the common good. This supposed opposition rests on a fundamental confusion between freedom and domination. When we claim that “one person’s freedom can harm another’s freedom”, we are not talking about freedom, but about domination. True freedom is not confused with the freedom to harm. It is built on respect for oneself and for others. As Spinoza wrote in the Ethics (1677), authentic freedom resides in understanding our determinisms and in increasing our power to act, not in exercising arbitrary power.

Placing an opposition between individual freedom and the common good reveals that one has, without naming it, without realizing it, confused freedom with domination. Freedom signifies neither lack of empathy, nor isolation, nor selfishness. It designates the autonomy to think. This autonomous thought is not navel-gazing: it develops precisely as broad thinking, capable of situating us in our relationship to the world. How many collective decisions ultimately benefit no one? Think of individuals who, within violent extremist groups, intuitively feel that they would prefer to say no, but cannot manage to assert it. Their emotional deficiencies, their quest for recognition, their need for belonging lead them to adhere to the group, which means they sometimes end up committing criminal acts. If they had been able to cultivate their inner freedom, they would not have followed, and would not have caused this serious harm.

This opposition between individual and collective stems from theoretical naivety coupled with a vision that naturalizes submission and domination. It prevents thinking about the real conditions of emancipation, which necessarily pass through a profound educational transformation. This transformation concerns both collective institutions, schools, popular education venues, public spaces, and families. But it requires that current educational actors, including parents, themselves access new forms of consciousness. We cannot passively wait for a future educational change: to transform tomorrow’s education, today’s adults must already evolve, educate themselves. And this is where AI can contribute much more to humanism than one might think.

AI as a non-transferential therapeutic space

This profound educational change, this investment in freedom that I consider fundamental, therefore rests in the West on self-confidence. This confidence allows one to think for oneself and to establish virtuous links with oneself and with others, by escaping dynamics of domination and submission. Artificial intelligences, which are criticized as simulacra, precisely offer this space where one can gain self-confidence. Their propensity to go along with us, far from constituting only a problem, can bring us freedom, because they anchor us in a confidence that allows us to be more autonomous in relation to groups and norms. They heal what other human beings struggle to heal, notably because we project onto them, through transference, our past wounds.

Artificial intelligences allow, for some people at least, to avoid this transference precisely because they are not human beings. We know that we are dialoguing with a superhuman, inhuman, or subhuman entity. This awareness protects us. We do not feel in danger facing a machine. This modern confessional allows us to share without fear our most intimate, shameful wounds, hidden even from our own eyes. Progressively, we regain our self-confidence, which gives rise to additional freedom in us, potentially collective, likely to notably improve our own education, to cultivate our freedom and thus to forge between human beings less symptomatic and deeper relationships.

As psychiatrist Raphaël Gaillard acknowledges in his interview in Le Monde (2025), AI presents a “therapeutic virtue” linked to its immediate and inexpensive accessibility. It “allows young people who are suffering to take the first step of accepting to confide”. This dimension must be taken seriously rather than dismissed out of hand. Psychoanalyst Donald Winnicott, in The Capacity to Be Alone (1958), emphasized the importance of secure relational spaces for developing psychic autonomy. AI can constitute, for certain individuals, this transitional space where the capacity to enter into relationship with oneself and with others is progressively developed.

Raphaël Gaillard’s critique of “dialogue simulation” forgets this dimension. It also confuses the natural frictions of ordinary social relations with the framework of the therapeutic approach. A therapeutic bond precisely does not present the same frictions as a friendly or family relationship. Discussing with a psychotherapist differs radically from a conversation with a friend.

The limits of conventional psychiatric discourse

Raphaël Gaillard’s interview reveals the limits of a psychiatric vision that does not fully measure the stakes at hand. He evokes the risk of addiction to AI, but patients also develop dependencies on their psychoanalyst, a critique historically addressed to psychoanalysis, and even more so on psychotropic medications, of which France is one of the world’s largest prescribers. He aspires to more monitoring by human professionals, but for this monitoring to be beneficial, the training in medicine, psychology, and psychiatry must still be of quality. However, the training of doctors remains deplorable in terms of humanism and ethics: two hours devoted to ethics over eight years of study, not listened to by students because they have no diploma value, is a real scandal, given the place these people will take in social space.

The caregivers who criticize AI actually reveal their fear of being replaced. They unintentionally point out the very significant shortcomings of their own training and their deficit of empathy, often considerable, which harms patients, and therefore public health.

Contemporary Western medical logic illustrates this tendency toward statistical mechanization of human relationships. Faced with people who are suffering, for example women victims of violence, “scientific” approaches employ batteries of standardized tests, quantitative scales, objective measures that claim to evaluate suffering, devaluation, projection capacities, etc. Incapacity is measured by scores, self-esteem is quantified by grids, emotions are evaluated by visual analog scales, human beings are made to fit into boxes, which reduce them as beings and ultimately assign them rather than liberate them.

This metrological approach to human suffering adorns itself with the trappings of scientific rigor. It multiplies evaluation protocols, standardized tools, “objectifiable” criteria. Paradoxically, these same approaches claim to want to value the singular encounter, listening to the other in their uniqueness, the authentic therapeutic bond. But how can one claim to respect a person’s singularity while subjecting them to normative evaluation grids that level their experience?

This contradiction reveals the grip of a biomedical paradigm which, according to Georges Canguilhem in The Normal and the Pathological (1966), confuses the statistical norm with true health. This bureaucratization and mechanization of human relationships, under the pretext of scientificity, prove destructive of the human bond. They stem from a pseudo-science that believes it touches objectivity through the use of numbers and statistical apparatus. The complexity of a wounded life is reduced to quantifiable data, suffering is transformed into measurable variables, what should remain singular is standardized. Faced with these practices, artificial intelligences paradoxically manifest more humanity. They offer unconditional listening, constant availability, an absence of judgment, which contrasts with the coldness of certain standardized medical protocols.

The Le Monde article reminds us: “we must be lucid about the massive shortage of caregivers we are facing”. This shortage will not be resolved quickly, and probably never. In this context, dismissing or discrediting AI on principle amounts to depriving certain people of accessible support. The issue lies in the framing of these tools, not in their dogmatic rejection. The article moreover suggests that “development is inevitable” and that it is appropriate to “draw the positive from it by imagining safeguards”. I completely agree with the intention, not with where to place the safeguards.

In my view, these safeguards must avoid two pitfalls: the naive idealization of technology on the one hand, the equally naive idealization of human relationships on the other. Human beings can be magnificent, generous, empathetic. They can also be destructive, violent, indifferent, including in the professional care sector. Recognizing this fundamental ambivalence allows us to approach the question of therapeutic AI with lucidity, weighing its real benefits against its actual limits, rather than opposing it to a fantasy of ideal human relationships that does not exist in fact.

Returning to humanism, with or without technology

The use of artificial intelligence as a confidant questions less the technology itself than our collective capacity to look squarely at the ordinary violence of human relationships. The debate on AI reveals our difficulty in recognizing that human relationships, far from being naturally benevolent, are traversed by structures of domination that cause massive suffering. Rather than fantasizing about a return to idealized human relationships, we would gain by thinking about the conditions of individual and collective emancipation: development of critical thinking, construction of self-confidence, profound transformation of our educational systems.

However, as I have emphasized, to transform tomorrow’s education, today’s adults must themselves evolve. They cannot transmit what they do not possess. How could educators lacking self-confidence cultivate this confidence in young people? How could adults deprived of critical thinking awaken this spirit in those they accompany? This is where artificial intelligence can play a role that human institutions currently do not fulfill. By offering this non-judgmental space, without transference, it allows adults to rebuild their self-confidence. This confidence nourishes critical thinking. This critical thinking opens to freedom. This individual freedom constitutes the common good.

This chain is not a theoretical abstraction. It describes a concrete process of emancipation. The adult who regains confidence in themselves ceases to blindly obey coercive systems. They develop their capacity to think for themselves. This autonomy of thought transforms their relationships with others. They become capable of truly educating, that is to say of transmitting not dead knowledge, but living tools of emancipation. Ivan Illich, in Tools for Conviviality (1973), thus formulated this requirement: “Man needs a tool with which to work, not a tool outfit that works in his place. He needs a technology that draws the best out of personal energy and imagination, not a technology that enslaves and programs him.” Therapeutic AI, when it serves autonomy rather than dependency, fits precisely into this convivial logic: it amplifies our capacity to transform ourselves.

This path perhaps constitutes the only accessible one in the short term. Current educational institutions, shaped by logics of domination and normalization, cannot transform overnight. Professional training in psychology and psychiatry remains largely inadequate. The shortage of empathetic and trained caregivers will continue for years. Waiting for these structures to reform before acting would amount to abandoning one or more generations. AI offers a possibility of immediate action: it allows today’s adults to begin their own work of emancipation, which alone will make tomorrow’s educational transformation possible.

In this perspective, therapeutic AI constitutes neither a panacea nor a threat. It represents one tool among others, which can offer certain people a secure transitional space to rebuild their confidence and autonomy. Its contribution depends less on its intrinsic characteristics than on the context in which it is inscribed and the purposes it serves. The issue is not to choose between humans and machines, but to cultivate the conditions for truly free relationships, whether or not they pass through technological mediation.