Artificial intelligence, relationships and emotions

Current debates on AI emotion recognition obscure more fundamental issues: generalized surveillance, the evolution of law, and the need for ethical education.

Generalized surveillance: a more immediate danger than emotional recognition

I observe that we are now manufacturing a pseudo-prudence concerning the use of artificial intelligence. The debate focuses on their ability to recognize emotions. Can they identify joy behind a smile? Analyze facial micro-expressions? This question, while important, seems to me to mask a far more serious problem: automatic facial recognition, which has become widespread without generating any real collective resistance. We accept this permanent surveillance with disconcerting ease, lulled by the illusion that being constantly observed would protect us more.

This acceptance rests on a dangerous assumption: the benevolence of those who surveil. Yet, observing the current system, particularly in France, should alert us. One need only examine police recruitment and training methods to understand that nothing justifies blind faith in the democratic soundness of generalized surveillance. As the European AI Act regulation that came into force in June 2024 points out, even systems supposed to identify emotions have major “shortcomings”: “their limited reliability, their lack of specificity and their limited generalization capability”.

The current legal distinction moreover reveals the absurdity of certain positions. According to Eric Delisle from CNIL, verifying that a person is stressed via their biometric data is prohibited, but observing whether they are smiling is not, as smiling is not legally considered an emotion. This regulatory gymnastics clearly shows that we are fighting the wrong battle. While we debate emotional nuances, mass surveillance is establishing itself permanently in our societies.

The relational evolution of AI: toward machinic humanity

Let’s now look toward the future without deluding ourselves. The recognition of emotions, whether through textual analysis or non-verbal expressions, constitutes the very foundation of human relationships. If we want artificial intelligence to be increasingly useful to us (and this is the very meaning of their existence, their massive adoption by all of us, and their flourishing economic models) they will necessarily need to develop this relational capacity. Emotional recognition forms the basis of empathy, and empathy enables understanding of others’ needs, a fundamental tool for machines designed to serve us.

This evolution raises a philosophical question that Sherry Turkle notably develops in her work on human-machine relations: these artificial entities, although being machines, will interact with us with increasing finesse. They will often be more attentive than most humans, precisely because they are dedicated solely to this relational function. A human being has their own existence, personal concerns, attentional limits. A machine created for relationships does not have these constraints, because its very existence is defined by its relational function.

This fundamental ontological difference should not blind us. Despite the absence of a natural biological existence, these intelligences will paradoxically develop more and more humanity in their interactions. Not authentic humanity in the phenomenological sense, but functional humanity, optimized for relational efficiency. As philosopher Luciano Floridi notes in his theory of the infosphere, we are entering an era where the distinction between biological and artificial agents becomes less and less relevant in the shared informational space.

The inevitable adaptation of the legal framework

The legal objections that might be raised against me remind me of the history of YouTube’s creation and music streaming platforms. Before YouTube, broadcasting a video or anything else on the internet without holding the rights immediately exposed one to prosecution for counterfeiting. The hosting platform could be charged as a publisher for not having verified the legality of the content. YouTube completely disrupted this legal framework. Users simply checked a declarative box, transferring legal responsibility. The platform was only required to remove disputed content upon request.

This radical legal transformation, initially illegal, was imposed by the force of usage. YouTube faced numerous lawsuits in its early days, but the new fluidity it brought to video distribution so profoundly changed social practices and the economic models of more and more actors that the law had to adapt. Today, this model has become the legal norm. The same phenomenon occurred with Deezer and Spotify, initially perceived as the gravediggers of the music industry, before new economic models took hold.

I am convinced that it will be exactly the same for emotional artificial intelligence. What seems strange or illegal today will constitute tomorrow’s normality. Legal scholar Lawrence Lessig, in his work “Code and Other Laws of Cyberspace” (1999), perfectly explains how technology always ends up redefining legal frameworks rather than the reverse. The European AI Act, despite its regulatory ambitions, will also have to evolve in the face of usage. As Pascal Bovero from the National Union of Printing Industries recognizes, faced with generative AI, certain professions are “doomed within five years”, so the law will necessarily follow these transformations.

Situational ethics versus universal morality

For me, the real issue is not to legislate on the technical details of emotional recognition, but to develop a robust and deep ethical framework. Ethics differs fundamentally from morality: morality imposes fixed and universal rules, while ethics is constructed in the relationship between the concrete situation and the singular position of each actor (person or collective). This distinction, which Paul Ricœur develops in “Oneself as Another”, is essential I believe for thinking about our relationship with AI.

Allow me to illustrate this difference with a concrete example. Let’s imagine a war photojournalist, documenting terrible scenes by photographing them, mutilated bodies, victims of extreme violence. In their professional ethics, they consider that bearing witness to the world’s violence is part of their social function. This documentation seems morally justified to them. Faced with the same scene, my personal ethics would forbid me from taking these photos. I would feel voyeuristic, disrespectful of the ultimate intimacy of these exposed deceased persons. Neither of us is more moral than the other, we each act according to our situated ethics.

This understanding of ethics as personal and contextual construction is precisely what we need in the face of AI. Rather than imposing absolute moral prohibitions, we must develop flexible ethical frameworks, capable of adapting to varied contexts of use. The adoption of these AIs depends more on cultural acceptance than on the legal framework. This acceptance must be built on a solid ethical foundation, not on moralizing prohibitions.

Ethical education: an imperative for the future

The exponential increase in power that artificial intelligence confers upon us makes education in ethics absolutely essential. These tools give us unprecedented capacities for action upon others, manipulation, surveillance, influence among others. Without ethical safeguards, we risk sliding toward unprecedented forms of domination. Martha Nussbaum, in “Not for Profit: Why Democracy Needs the Humanities” (2011), argues that humanistic and ethical education is our best defense against technological dehumanization.

This education must not be limited to school. I advocate for lifelong ethical training, adapted to technological evolutions, which are constant. The ethical education I call for does not consist of teaching fixed rules, but of developing the capacity for autonomous judgment. It’s about learning to question our uses of technologies, to identify the power relations over other human beings they institute, to imagine alternatives respectful of human dignity. This approach aligns with Paulo Freire’s critical pedagogy: training citizens capable of reading the technological world to better transform it, rather than mere docile users.

The future of our relationships with artificial intelligence is being played out now, in the ethical and educational choices we make collectively. Rather than losing ourselves in technical debates about smile recognition, let’s focus on what’s essential: developing a living, situated ethics, capable of accompanying technological transformations without sacrificing our humanity. Only under this condition will we be able to build a fruitful coexistence with these new forms of relational intelligence, which can no longer be stopped anyway.