Roman Malo, associate professor of clinical psychology

[M]: Thanks Roman for taking part in this interview for the readers of the IA et Psychothérapie site. The idea is to give a bit of your background and draw a portrait of you as a psychologist, teacher, researcher, and of how you came to encounter artificial intelligence. Maybe you can start by introducing yourself briefly.

[R]: Thanks for the invitation. I always take great pleasure in reading you, so it’s fun to be on the other side too. My name is Roman Malo, I’m an associate professor of clinical psychology at Nantes Université. I have a master’s in “plural integrative” clinical psychology — the label shaped my path. I’ve always had an interest in neuroscience, cognitive science in the broad sense, and also in psychoanalysis. Rather diverse and varied interests.

[R]: Then I did a PhD in co-supervision: on one side psychology/psychiatry, and on the other digital technology. My thesis was on the use of virtual reality as an aid to diagnosis and to the measurement of executive functions. I also have an interest in qualitative research methodologies. All this led me to discover the digital world, and with one of my thesis supervisors I discovered a whole universe. To put it simply, I first got interested in digital technology through video games. I have a background as a player, initially in my adolescence. Virtual reality amazed me, I got straight into it, and it led me to take an interest in digital technology broadly, then naturally in AI. It’s a bit of the video-games → VR → AI path, the triptych so far.

[M]: Very rich and diverse. I hear a lot of curiosity, the refusal to be confined to a single approach. In virtual reality, what seemed promising to you in connection with the psychology side?

[R]: Two things sparked my interest. First, the diagnostic aspects and the measurement of executive functions: I was always surprised by how off-the-ground our tools for measuring cognitive functions are. I discovered all the neuropsych batteries at university, and I was struck by their un-ecological character — until I discovered that some tests with recognized psychometric properties existed in gamified, ecological versions.

[R]: The second aspect was psychotherapeutic. I’ve always had an interest in active psychotherapies, where you give the patient things to do. I draw a parallel between psychological functioning and how the body works: as an athlete, when you’re injured, you go to a physiotherapist and do things to get back on your feet. It’s the same on the psychology side. There’s the in-session work, indispensable, which is most of our work. But we’re also increasingly moving toward prescribing exercises.

[R]: I was struck 5 or 6 years ago by an exhibit on mindfulness therapy in VR: doing breathing exercises with a gamified aspect, for people who don’t necessarily have good adherence. For a patient who doesn’t do what you ask, there might be something interesting there.

[M]: I hear the “using technology to create a situation closer to the real, closer to the patient’s body” side. And when you talk about gamification, I also hear a lower step to make therapy more attractive and active.

[R]: Yes, completely. In directive approaches — I’ll take the example of CBT, even exposure therapy — I always found there were exercises that could easily be sabotaged. When you ask a patient to imagine a phobia-inducing situation, it’s not that they lie, but they can easily do the exercise halfway. Our brain cuts things short as soon as it displeases us. With VR, you can choose a fine granularity and a level adapted to the patient. For an agoraphobic patient who has spent 15 years avoiding crowds, asking them to take the tram again at rush hour is already a mountain.

[M]: It fits the CBT approach well: the hierarchy of fears, gradual exposure. And as you say, with exposure in imagination, we don’t know what’s happening inside the person; there can be subtle avoidances that slip under the radar. Not to mention patients who can’t form images.

[R]: Even the capacities of imagination… I have the impression that when we propose therapeutic methodologies, we have two presuppositions: one, that the patient will manage it; two, that they are necessarily in good faith. From experience, both seem to me to be constructions on the psychologists’ side. Our brain is very good at making us continue what we already do, and effort is costly. If we don’t transform… Especially since these technologies are accessible today — it’s no longer fantasy, it’s no longer reserved for specialists or geeks.

[M]: I hear a lot of colleagues grumbling that the patient “would no longer have to make an effort.” But what you’re saying is that it’s about investing effort in the right place: in the exposure itself, not in the difficulties upstream. If we add obstacles before the therapy, we test the patient’s motivation in the wrong place.

[R]: I completely agree. There’s a concept I really like: friction. If there’s a lot of friction too early on the therapeutic path, the patient drops out. It’s all the more interesting with AI: we don’t innovate much on the psychotherapeutic tools themselves, we stick to tried-and-tested techniques. And yet we’ve never published so many books on anxiety disorders, while our patients remain in difficulty.

[R]: I like trying to understand, with patients: why don’t they change? Beyond the intrinsic processes, there’s a very material, operational side that we can try to lighten. It’s one of the psychologist’s roles: to take part in an architecture of care that makes sense and is adapted to where the patient is.

[M]: How did you go from VR to AI? When did you first encounter AI, in theory or in practice?

[R]: In theory, quite a while ago. One theme has interested me for a long time — we’ve already had the chance to talk about it: consciousness. What interested me in psychology was trying to understand how consciousness emerges. At first it was a lot of curiosity. I did little trial-and-error experiments on ChatGPT and Claude, like many people. I tried to push the technology quite far: I went from the cooking recipe to really pushing a line of reasoning, to see what happened. I found it fascinating. I didn’t recognize myself in what my psychologist colleagues might say to me, often with a surface approach: “it’s Google augmented that answers well.” I had the impression there was something that also fit well with the symptomatology of our patients today.

[M]: In what sense?

[R]: There’s an article I like by Thomas Rabéron, “The pathologies of dreaming,” which draws on a sociologist’s work on the reification of our health model. Many patients today want answers, a precise diagnosis, precise information. And when you use AI, I think: here’s this fantasy of the psychologist who gives precise information, right in your pocket. It exists — and yet psychotherapy hasn’t made an enormous leap. I find that fascinating. And I try to follow AI’s developments precisely, because what we write today is perhaps already outdated by the time we publish it.

[M]: On the pathologies of dreaming and your way of experimenting, which may differ from other people’s: we tend to treat technological tools like the ones we knew — a bit like a calculator, you ask a question, you want an answer. Here, chatbots reintroduce the art of dialogue, almost in a Socratic mode. In your experiments, it’s a process of questioning, of elaboration, of co-elaboration. A completely different state of mind.

[R]: Yes, completely. I tried to approach AI without presuppositions — it’s very difficult. But when you approach AI and mental health, suddenly all the arguments that are usually scientific become almost secondary: what counts is a worldview. “I like AI, I don’t like AI, it’s dangerous, it’s good.” I tried to approach it differently: what does AI concretely do, what can it bring us, what does it change? You have to be lucid. We are not facing a new calculator. We are facing something that is radically new, that raises many questions about agency, autonomy, consciousness, solitude — fundamental anthropological questions. It makes me think: we often discuss the telephone as a new communication technology, but here we are facing a phenomenon of a completely different nature and extremely accessible.

[M]: I agree with you. There’s really a leap. For me, it undermines our Western naturalist ontology, in which the human would always have “something more,” by definition. I hear many scientists say that as an established fact, when it’s a belief — which is not an insult between psychologists. We all have beliefs, but it’s useful to become conscious of them. This AI phenomenon manifests properties that are not pre-programmed and in front of which our habits of reasoning are out of step.

[R]: Completely. I saw Gérald Bronner on Instagram, I think, comparing AI and spirituality — we return to things that surpass us, a fascination for something almost divine. It makes me associate with the question of intentionality: we judge a piece of work differently depending on whether we think it was done by AI or by hand. We spend our time reconciling our beliefs with the world, and what bothers us is when it doesn’t coincide. AI is so new that it can’t, for now, coincide with our beliefs. The argument that comes up around me: “you have to check what the AI says.” But actually, we should do the same for all humans.

[R]: I like this idea of extended cognition: we’ll have a new element, just as today I wear a watch, I have a phone. Except this object will have many functions at once. We underestimate how fast it’s going to go, and how much it will create great disparities between individuals.

[M]: To sum up, you point to what we can call an epistemological double standard: we expect from AI a level of evidence we wouldn’t require of a human. A recent article — Against What Standard? Why AI Therapists Face Impossible Expectations by Abu Ghanem and Greenbaum, published in early 2026 in The American Journal of Bioethics — pointed precisely to two biases they call idealized comparator bias and perfection asymmetry bias: comparing AI to an idealized human gold standard that never exists in practice, and requiring of it a perfection we don’t expect from human clinicians.

[R]: Exactly. I often see the motif “we compared an AI psychotherapy to a gold-standard therapy.” We can already relax about that: is there a psychologist who has a perfect therapy and never makes a mistake? I think it’s a protection we’re creating. It’s very interesting, this shift: all the pro-AI arguments have to be doubly proven. It’s rather strange.

[M]: My hypothesis: it’s one of two things. Either AI is a classic computing object, and then if it doesn’t give the right answers it’s defective — we expect the deterministic, the predictable. Or it gives good results but that’s hardly predictable, hardly controllable, and that takes it out of the “computer program” category. And then the closest categories for dealing with this kind of fuzzy intelligence are the categories we use between humans.

[R]: Exactly.

[M]: How do we measure a human therapist’s performance? Certainly not by checking that they never say anything foolish, otherwise we’d all be for it. It’s more about being aware of our limits than about being perfect.

[R]: I completely agree. Many elements we’ll have to think about as psychologists will be positions that will also have to rely on philosophical concepts, not merely take refuge behind evidence. We’ll reintroduce notions like morality, dimensions that are sometimes hard to measure: social contact, presence. That’s where I’m waiting for double-blind empirical evidence to explain solitude when there’s an AI available and answering at any time of day. We’ll come back to fundamental human qualities, perhaps to fundamental needs — I like Maslow’s pyramid. There will be many questions, and I’m quite impatient too to see what it will generate.

[M]: We’re aligned that the arrival of AI, which manifests qualities similar to the human on many levels, forces us to clarify what it is to be human, and what human values are. And besides, do we always manage to put these values into practice? We, as psychologists, know well the difference between an idealized vision and what we observe on the ground.

[R]: I agree. It’s interesting, even the way we talk about AI: “the AI,” we give it an identity. I have the impression we’re taking an additional narcissistic wound, and this one is costly. It touches on fundamental dimensions — consciousness, reciprocity in exchanges. An AI that imitates a human well enough: do we manage to identify the difference with a chatbot? It’s not a very new question, but I think we’re uncomfortable being able to be wrong, not to understand everything.

[M]: Yes, I agree with you on the wound. Other thinkers have used this expression. I imagined this a little over a year ago: a fourth or fifth narcissistic wound, when you count sociology in the list. It points to the construction of our species, where we thought the universe revolved around the Earth, that the human was the summit of evolution, that we were beings of reason. And then Galileo, Darwin, Freud, and sociologists came to add a bit of humility to the way we conceive of things.

[R]: What bothers me — here as a psychologist and researcher — is that these tools also have a lot to offer us. There is sometimes a catastrophist vision that forgets the potential of these tools. For some patients, I see a real value in them. It would be a shame to be only in defense against this wound. You have to be curious, be a bit surprised. Surprise is important in the face of this.

[M]: The stance we share here is almost that of a civilizational therapy. In our anxious patients, we see clearly how aversion to change crystallizes to idealize the present, demonize change and maintain things that aren’t necessarily great. We have a bit of that phenomenon at the social scale.

[R]: Completely. And in the face of AI, behavioral therapies — I’ll even drop the cognitive, the behavioral — are going to take a growing place. Our patients will quite quickly accelerate the processes of knowing their internal psychological functioning. If we don’t add the practical, hands-on aspect, the move to action, we’ll increasingly face patients who are experts on their disorder and their functioning, and yet nothing will change.

[M]: There’s a risk of a gap between psychologists and healthcare professionals on one side, and increasingly expert patients on the other. I had discussions of this kind with Santé Mentale France on the empowerment of patients who can inform themselves about their disorders, the links with physical activity, diet, etc. An AI can give advice, adapted explanations — which isn’t necessarily obvious for psychologists with such varied knowledge. You, being a great athlete, how do you see this behavioral side in relation to AI?

[R]: I draw a lot on the parallel between the psychologist’s profession and the physiotherapist’s. In my sports practice, I now see much greater knowledge about injury. You injure your knee, you type in your symptoms, you get a diagnosis that’s often not too bad, a bit generalist. You ask for a protocol to recover from the injury: it presents you with everything we know — gradual reloading, the ability to stress the tissue, supplementation to synthesize collagen. That knowledge used to be held by physiotherapists: first explanatory appointment, second appointment a treatment plan, and the physiotherapist as guarantor of your exercises at home.

[R]: We psychologists find ourselves in the same configuration. Patients able to explain what’s wrong at the level of their brain — and it’s also a danger, it can lead to a kind of false knowledge: “I function like this, like this, like this,” pointing to a list provided by ChatGPT. The psychologist will have a new mission: to convince that they are an expert. You’ll have to show your credentials: “yes, I know what you’re talking about, I studied, it’s my job. What do you expect of me? Do we push certain hypotheses? Are you expecting a confirmation?” It will horizontalize the relationship. Beneficial for patients, but I’m not sure it’s always very effective on the treatment itself.

[M]: It makes me think of the distinction, in hospitals, between technical care and relational care. The impression that ultimately the “technical” part — explanations, psychoeducation — could be provided by an AI, and that for psychologists the relational part would be brought to the fore. Paradoxically.

[R]: Completely. A dozen years ago, when I wondered which profession would be replaced by AI, for the psychologist I could never answer. The more I go on, the more I have the impression that the very technical aspect — psychoeducational, explanatory, exercise supervision, monitoring — we’ll be able to automate it, even save time by automating it. In an institution, taking an hour to explain the same thing to all patients, in a saturated system… On the other hand, relational qualities are really going to be added value.

[R]: We often say it’s our psychologist’s mumbo-jumbo, that it doesn’t exist. An example: attachment disorders. You can explain to someone what it is to feel safe, to be in a secure space. But above all you can make them feel it. That’s the specificity. The frame, the setting, regularity, stability, continuity of care, being right on time, being ready to listen — everything that makes up a secure attachment relationship. On that, AI does not replace us.

[M]: It refocuses on what is hard to replace. On the human.

[R]: And it bothers us as psychologists. In university training, what’s valued is often being a good psychotechnician — knowing which test to use, which scale. We’ll have to rethink the way we teach and train. We’ll move closer to how psychotherapy schools work: more experiential, more personal psychotherapy, really the human. And that raises many questions: how do you do it with large cohorts?

[M]: It brings us more into the body, into embodied experience. Far from intellectualization. Is that compatible with the university?

[R]: It’s going to be extremely complicated. And I also notice, on the academic side, a return of the body — embodied cognition, bodily synchronization. But training the body without calling on the body, I find that difficult.

[R]: And from an academic point of view, it’s also hard to evaluate. How do you evaluate a relational quality to validate a clinical-consultation exam? You’d have to put yourself in the patient’s position, offer students 20 interviews, give feedback on what it made the fake patient feel. It’s very different from the usual way of seeing the profession.

[M]: Evaluating the felt sense. And we’re both interested in analyzing the dialogues between patients and AI. As for humans: how do you analyze the quality of a dialogue, whether between humans or with an AI, knowing it depends a lot on the time window. An empathic confrontation evaluated just after: “well, the psychologist wasn’t very nice.” Evaluated two sessions later: “oh actually it was really useful and constructive.” An old, thorny question of psychotherapy evaluation. When you read the articles or hear the debates, I have the impression we don’t often ask the opinion of those most concerned: our patients.

[R]: Exactly. A whole literature deals with the evaluation of psychotherapies — which is actually still stuck at step 1. We have the impression that today it’s easy, but I’ll be very cautious. What we can say is that we still don’t know exactly how to evaluate a psychotherapy. We evaluate certain outcomes: symptom reduction, modification of the symptomatology. When we evaluate psychoanalytically-oriented therapies, some articles show a change of structure…

[M]: And it’s theoretically supposed to be fixed, by the way, the structure.

[R]: Exactly. So AI and psychotherapy makes us uncomfortable too, because we come back to a difficult problem: the evaluation of psychotherapy itself. There are many concepts we consider settled that would gain from being made more precise. The therapeutic alliance, the common factors. When we open the door to evaluation, let’s stay very modest. And it also makes us uncomfortable that many changes in our patients aren’t necessarily linked to what we do, but to environmental factors.

[M]: Often we evaluate target symptom criteria, but we can miss the key point: the patient’s global appreciation of their life. That’s where we can find philosophy or literature again.

[R]: I have a particular interest in phenomenology. In psychology, the idea persists that qualitative studies would be a bit more false, a bit less exact — “science, but not quite science.” And that if you ask a patient, there’d be suggestion effects? Yes, there are. But that doesn’t stop us from going to ask those concerned what they think. We’ll have to go back through an evaluation of psychotherapies by way of patients, and stop this illusion that we can totally objectify the subject. In any case, the different approaches broadly have the same results.

[M]: The more we try to measure reliably, the more we reduce the real to increasingly artificial variables. Interesting that we speak of artificial intelligence — part of our scientific approach itself is “artificial” in the sense that it seeks the purest possible conditions to measure a link — not of correlation but of causation. But we lose the ability to generalize. Besides, we’re writing an article together on this, on the lack of ecology in published studies.

[R]: And we also lack something fundamental: the “art” aspect, in the sense of the artisan. Today, we have sometimes exceptional psychologists — the super shrinks, I’ve already talked about them. We’ll have to look at what they do, not only what they say they do. And also study alternative therapists quite faithfully. Controversial, but there are things to learn from alternative therapies on relational qualities — not on the techniques, which are often hard to support scientifically.

[M]: Priority on relational quality, then. How does it go for your teaching and your students in relation to AI? What are their questions, their uses, what do you offer them as reflection?

[R]: The university is in difficulty on this subject, and so am I. It’s hard to codify uses. In Nantes, a working group on generative AI has just sent a V1 of a code for students and teachers. But it’s very hard to codify something that moves so fast. On image generation, for example: not long ago, the quality was so poor that we used it little in coursework. Today, some graphics — I can no longer tell whether it’s students or AI. And it’s an accusation to tell them: “you didn’t do this by hand, I know it.”

[R]: When I ask students if they use AI, they always minimize it, as if it were cheating, when the tool is available and used by everyone. It’s astonishing. What I try instead is to use AI for what it is, no more. For restructuring, clarification, it’s extremely useful. With this idea of re-checking what the AI says: just because the language is pretty doesn’t mean the idea is correct, especially on sources. And to start first with work outside AI, then to complement it.

[R]: I like this idea of a prosthesis: you add on, you enrich, but you first call on your own reflection. I try to explain to students — and I’m afraid of being a big boomer saying this, but I do it anyway — that you have to train your reflexivity. If you give a case study to ChatGPT, at the diagnostic level, therapeutic proposal, identifying signs and symptoms, you easily get 15, 16, 17, 18. But what interests me in training psychologists is training reflexivity. Not the result, the process.

[R]: Which will bring reflection on assessment: do we evaluate the ability to prompt well, the ability to think, competence? There’s something quite urgent to do: have dedicated courses. Experts — computer scientists, designers, all those who orbit AI — to explain what you can do, what you can’t do, how it works. There’s almost something of popular education about it.

[M]: The universities’ reaction, the suspicion of cheating, as if there were a purity in the current way of doing things: it fits well the imaginary of the student who uses AI “in their place.” Obviously, that can be the case — it’s the calculator mode. But if teachers take up the tool as an aid in a process, AI can challenge the person’s thinking. I use it like that: to push, elaborate, propose alternatives. It’s not so much AI as the way we use it. By analogy: “going to the library is cheating, you can copy quotations.” Except that’s exactly what researchers do. And deep down, Plato already thought writing was cheating, and that real philosophy was oral.

[R]: Same when you use a translation tool. If you have DeepL translate and you go over it afterward, it has more value. If you call on an external native English speaker, what’s your share of the work? Not much. But that points to a limit: you’ll have to ask students too. And it’s hard to ask someone to be honest about how they get around a system: “explain to me how you handed in this assignment where I gave you a 17, when you didn’t quite do it yourself.” Lots of presuppositions on both sides.

[R]: And there’s a big problem: it points to our knowledge of AI. When I ask an interlocutor to explain to me how an LLM works and to give me examples of AIs and their differences, it often gets stuck. Talking about use without knowing the tool seems perilous to me. Saying “AI must be banned”: but then, we don’t use Claude for repetitive tasks? We don’t use AI to find a paper when we only have a screenshot on our phone? It gives a melting pot, a big anti-AI package with no nuance, which I sometimes struggle to understand.

[M]: It forces us to clarify where we put value. And it’s profoundly subversive: for dominant structures, the value of things is what is. AI forces us to clarify where value is, and where it is not.

[R]: And above all, from students we expect production. AI has increased productivity in the sense of “there’s more output,” more possibility of doing more. But that doesn’t mean today’s students are more competent. I like giving assignments that link concepts from the course to popular culture or a clinical example. In a course on defense mechanisms in psychoanalysis, for instance, identifying defense mechanisms in a fictional work. What’s interesting is going to see the work, asking yourself questions. Not giving me an output to check that at 12 minutes 24 the character says a colon: that brings nothing to someone destined for psychology. But this kind of assignment is almost no longer possible.

[M]: It connects to what you said about exposure exercises. In our routines, we can confuse the strategy and the need it’s supposed to nourish. And our social structures have even more inertia than psychic structures.

[R]: Completely. It raises the question: can we still evaluate remotely what was evaluated as coursework? We’re forced to, but I find it a great shame, because it brings us back to the in-class exam. I feel like I’m talking about the university of the 1960s, when not at all. But there’s this interesting side where you get lost in your research: you have coursework to do on subject A, you read an article that refers you to another, you look at your watch, and you’ve been reading things for two hours that no longer have anything to do with it. In terms of productivity, it’s not great. In terms of reflexivity and curiosity, it’s wonderful. And to be a psychologist, we are guarantors of curiosity with our patients — you have to be able to get lost and experiment. If we ourselves don’t do it, we lose something.

[M]: So it potentially gives back room to the oral, to live exchange. Now, the teacher/student ratio doesn’t make it easy.

[R]: We’ll relearn how to make specialists of the human work with real humans and not papers. As teachers, we’re overwhelmed, because we have to rethink the whole way we teach. In Nantes, in the master’s programs, we call on professional actors to play patients — it’s exceptional. But not feasible for cohorts of 1,000 students. There, AI could be interesting: generate fake patients or realistic situations on video, and have students carry out consultations on these clinical cases. You have to accept it, dare. It requires building assessment methods we haven’t always thought of.

[M]: We could even imagine that each student has a recorded vignette where AI plays a patient, then a mixed analysis with psychologists and an AI as judges. It’s not ideas that are lacking.

[R]: It could be an assessment to obtain your title of psychologist: you go through a consultation system where psychologists observe how you conduct a consultation. There’s everything to build, you have to be curious, venture out — which is also difficult.

[M]: Obviously, nothing replaces the internship and field experience. But it would be a complement.

[R]: Obviously, the internship is essential. To teach psychology, we’re much closer to an artisan’s work than a computer expert’s. To learn to cook, you have to cook. It’s the same for the consultation.

[M]: In terms of research at the moment, or the students you supervise, can you say a few words about the topics?

[R]: At the moment, I co-supervise a thesis on the therapeutic alliance and autonomous agents. It’s a co-supervision between cognitive psychology and computer science. The idea is to evaluate the quality of the alliance with digital devices. My current theme is studying what is potentially human in AI interactions, perceived empathy.

[R]: Another topic, shorter, is carried by a master’s student: the therapeutic pathway. How can an AI device lead a patient to seek care, or on the contrary divert them? My hypothesis: for patients at clinical or subclinical levels — say a 5 out of 10 on anxiety — all those below will not seek care thanks to AI, they’ll manage, tinker, improvise with its help. Whereas those who go above 5-6 will perhaps be led to seek care more quickly, since the AI will say: “I suggested this, it’s not working.” We’ll arrive at a polarization of consultations: the simple cases will seek care less.

[M]: By saying “under 5/10,” it’s not about minimizing people’s suffering, but rather the complexity of helping them. It’s an interesting hypothesis. I was discussing with other colleagues and researchers this potential double movement: perhaps some people will find what they need to feel better simply by pouring out an emotional overflow or asking for practical advice. So much the better.

[R]: And so much the better.

[M]: So perhaps some of these people won’t go looking for supportive therapies. But let’s not forget that the essential thing is that people feel better. If the person feels better, so much the better — we’re not going to wish people ill in order to have work. And for others, with more complex suffering, AI will perhaps lower the step to seek care. My hypothesis: people discover that talking about their problems to a “thing across from them” makes them feel better. I think it can normalize verbalizing one’s problems and destigmatize it. There’s potential for more consultations.

[R]: I agree. Whatever we say, the psychologist remains a somewhat mysterious figure today, even if we’ve destigmatized mental health. There’s still the moment of making an appointment, the financial cost. In my circle, I hear “I’m not going to see a psychologist, that’s for crazy people,” but every day I hear people explaining that they use ChatGPT, that they pour out their emotions, that they discuss their family with ChatGPT. And often what comes out is: “it makes me feel good. I know it doesn’t mean anything, but it makes me feel good.” I tell them there’s a profession that does the same thing, with trained people. It could be complementary — provided psychologists play the game, accept a nuanced approach and aren’t fundamentally anti-tech.

[M]: I hear a lot the idea that people who would feel better talking to AI would be those with anthropomorphization biases, hence almost pathologies. It’s a very Western way of seeing things, stigmatizing, pathologizing. Animist cultures might see it differently — and they are often cultures that haven’t destroyed their ecosystems, precisely because they considered that the non-human deserved a form of relationship.

[R]: Completely. And even: if you feel better chatting with ChatGPT, maybe there’s an underlying problem, or even it’s a “less noble” therapy. There’s a tenacious idea in psychology: psychotherapy must be tedious, must be expensive, must be long. This hierarchy is very toxic in how we approach health as a spectrum. One of the psychologist’s roles is to arrange for their patient to feel better, whatever the consented means. If for the psychologist that means leading the patient to a responsible and appropriate use of AI — I’m not talking about psychiatric pathologies, I’m talking about the average person going through a complicated phase of life — my personal opinion on AI doesn’t matter. You have to be capable of a non-judgment a priori, both about the patient and about the care device.

[M]: I already see it with so-called alternative therapies. Many colleagues grumble. I have patients who see astrologers, numerologists, energy healers. I don’t prescribe them, I ask them: what does it do you good, what did you experience, what’s going well, what’s not. We take an interest in the subjective material, because we’re not judges of a truth. I always worry when a person, psychologist or not, claims to hold the Truth.

[R]: Of course.

[M]: You have to refocus on what’s happening subjectively for the person. Do they say they feel better? Does it lead them not to isolate themselves? I often hear: “AI will replace this or that.” We find the fear of replacement again, one OR the other.

[R]: The spectrum no longer exists, it replaces, it makes you alone. We have an injunction to be for or against. It’s mandatory.

[M]: I’m interested in your perception. In your follow-ups, do you have the impression that patients using AI would become pathologically dependent, closed off, diminished — like the figure of the Japanese Hikikomori?

[R]: Ah, Hikikomori, the example trotted out to justify addiction.

[M]: That’s it, the “no-life” as they say.

[R]: Yeah, the famous no-life. I’ll come back to alternative medicines: it also redefines what we consider therapeutic, as a spectrum. I don’t know if in your personal life you do improv theater — it’s not a prescribed therapeutic modality, your psychiatrist isn’t going to recommend it. And yet…

[M]: And yet it’s very good, I did two years of it!

[R]: And yet it’s very good. You have creative or spiritual practices I’m not going to prescribe to my patients. But if it has an effect on stress management, I’ll clearly encourage it. AI is a bit the same: it depends what is done, in what context, at what moment, and that it doesn’t substitute for classic care. On AI addiction, I don’t think it’s an urgent matter for our patients. We have smoker patients: our obsession isn’t to stop the cigarette, it’s to work on the anxiety disorder, on depression. AI addiction, I haven’t yet observed it. There’s a fascination, but one linked to the novelty of the device. Actually, something that can provide answers to everything potentially has an addictive character, but again…

[R]: It makes me associate with the Rat Park experiment. Rats are offered cocaine or food: most take cocaine. But offer these rats an amusement park, social activities and food: the choice of cocaine drops sharply. If today we have people completely dependent on digital technology to the point of evoking an episode of Black Mirror, it’s perhaps because there’s something problematic in social interactions, that we’re facing an extremely lonely generation. It’s not AI. It’s easy to blame AI. But the anthropological and societal changes are, in my view, of a much more central urgency.

[M]: I share this view, and it’s actually a subject I address on a dedicated site, Intelligences Plurielles. The interview could go on for hours. To round it out, I’d like to underline a difference between social media and conversational AI. With conversational AI, you’re active, a stakeholder in a unique, co-constructed dynamic — it’s not the same as being passive in front of the TV or social media. You can scroll, click, but the type of experience is very different.

[R]: You’re completely right. I make the distinction on activity/passivity, I often come back to agency — as in video games. For TV, it’s a good example, or the phenomena of doom scrolling on Instagram: you scroll, you don’t act, you don’t learn anything, you don’t do anything. For AI, we’re a bit facing the same question: does it bring actions into the real? Exchanges about a theory? Low-level things? I don’t think AI in itself is the problem. And it makes me associate: having been a heavy video-game player, is a digital relationship less real than a real relationship? If you have a digital landscape with buddies, friends, people you play with every evening, is it less high-quality than human experiences with a group of friends you’re bored with?

[M]: We see this in our psychology practices. In a couple, how one partner sees the other is often out of step with what we perceive in the stated actions. In extreme cases, a person idealizes a domineering, manipulative, even violent partner, but looks at them with a child’s wonder-struck eyes. So talking about the “virtual” only to designate the digital is a bit easy — it resembles a mechanism of projection.

[R]: Completely. As long as this dichotomy exists — believing that what is material and visible right away, without a computer, fundamentally has a superior quality to what’s behind a screen — we’ll have a simplistic and biased vision. Our exchange here is much more interesting than, in five minutes, my going out to get a baguette with my earphones on: the relationship with the baker would be human, but would it be high-quality? I don’t know. We have to change that, because the purely digital doesn’t only bring unhappiness. When I hear this discourse, I get the impression that people outside the digital would be radically happier, without anxiety. There’s an underlying hierarchy: an original purity that would be screen-free, and all those who use screens forgetting what technology can bring.

[M]: It makes me want to conclude with the notion of pharmakon, which has three components. A pharmakon is an object or substance that can be either poison or remedy. The third aspect: we project onto this strange, multifaceted object. We become wary, and this wariness is welcome when it’s measured — it’s not about naive optimism — but it also shouldn’t resemble that of our very anxious patients who see only catastrophe scenarios, even if we need to consider them among a set of possibilities.

[R]: Exactly. Perhaps there is something to create, a bit like a Rorschach. Now, you would show a ChatGPT and ask: “what do you perceive?” It might be a new projective test. What do you see on this screen?

[M]: Yes, I already use that one a lot. Would you have any recommendations for future students who will become psychologists? What would you want to tell them about AI?

[R]: First, experiment and test. Go and get a little lost playing with AI. It’s a bit like virtual reality: there’s what you’ve heard, what you’ve read in the manual, and then what’s really there. There are quite fascinating things. I like the idea of the expert — the expert patient, the expert user. It’s interesting to talk with people who use AI a lot. Test, make mistakes, keeping in mind that using AI involves obvious risks — ethical, ecological. You mustn’t deny them. And you have to read more than psychology: a bit of anthropology, sociology, philosophy, and even computer scientists. I did my PhD with a computer-scientist supervisor: reading computer science as a psychologist is interesting. On the cognitive aspects, on how to conceptualize a neural network, and also on fundamental questions — consciousness, activity, agency — which are very design, very computer-science questions, and which aren’t so far from our psychology questions.

[R]: The second piece of advice is about relational skills. Maintaining them is a daily training. If society becomes digitized, you have to deliberately maintain our relational skills by experiencing the maximum of human interactions: talking to people in everyday life, with strangers. It’s something I often give, which comes from a book by Charles Pépin I like, “La Rencontre” (The Encounter): a very simple thing, but with earphones in the street, you never make an encounter. Take them out for an hour, and lots of things happen. You have to develop your skills at follow-up questions. Do your shopping without music, focusing on the interactions. It’s a real skill, and it’s something you have to live and experience, and that for now AI cannot provide. Really put the focus on that. Do theater if you want, but for everything that’s exchange and relationship-building. It’s really important.

[M]: A good conclusion. Ultimately, it’s AI that upends many things, but leads us — in your vision, which I share — to put value back on human interactions, the body, felt experience in the body, very concrete, very human things. That will be the closing word for this interview, if that’s alright with you.

[R]: Yes, thank you very much. And thanks too for your work, for everything you do. Your blog is of remarkable quality and has a curiosity and accuracy that I find very important on these subjects. And you always manage to make it very accessible even to non-experts, which is sometimes lacking on this theme where jargon taints public understanding.

[M]: I do my best, and I’m helped by my “amIAs” — I put forward this new word, my “amIAs” (AI-friends), which I don’t confuse with humans. Without this help, I couldn’t do this work. So I’m quite glad they’re there. It’s thanks to them that we can exchange together today and share these things.