Gaelle Charlot, occupational therapist in prison psychiatry

[M]: Can you introduce yourself: who you are, what you do, and in what setting you practise?

[G]: I'm Gaelle Charlot, occupational therapist since 2002. I began my career with traditional mediations: clay, pottery, collage, drawing, painting, rattan, mosaic. In psychiatry, these tools serve to express emotions contained within the patient's internal conflicts. Mosaic and rattan, for example, help to structure the course of thought in psychotic patients or to channel psychomotor agitation in mood disorders.

[G]: My career has taken me through four very different institutional settings. I started in Picardie, in a psychodynamic hospital service in a rural area. Then Pontoise, more oriented towards cognitive-behavioural therapy, in a peri-urban setting. Then Sainte-Anne in Paris in inpatient care and in a CMP at the Trocadero. And finally prison psychiatry in Bordeaux, which is a world apart.

[G]: In prison, I found that my patients often had learning disabilities that were never addressed in childhood, or had simply never had a conventional education, or had difficulty managing frustration. Offering them activities requiring sustained attention and concentration -- particularly those involving reading and writing -- set them up for failure. When I finally got hold of a computer, I began using music videos as a mediation. It was a medium that did not set them up for failure in terms of creativity, that facilitated the expression of emotions and the building of a therapeutic relationship.

[M]: In practical terms, how do you work with music?

[G]: I start with a standard assessment interview, then a specific session where I use two versions of Nothing Else Matters: the cello cover by Apocalyptica, and the original Metallica version with vocals and electric guitar.

[G]: This comparison allows me to assess several things. First, whether the patient is more responsive to the image or the music, and to which atmosphere. Then, whether they can verbalise their preferences. Some patients are unable to recognise that it is the same melody in both versions -- that is quite revealing. Those who do recognise it generally prefer the Metallica version, because the voice conveys the emotion.

[G]: I had a patient who could never remember the title or the artist, but who would sing the lyrics to me. With ChatGPT, I could find the song. This showed that he had mnestic abilities, but not the kind we usually expect. Without this mediation, we would not have been able to identify that type of memory.

[M]: What kind of population do you see?

[G]: I work in a remand prison where I see all profiles: from cannabis trafficking to homicide, from minor to major offences. At the SMPR, we distinguish several types of patients. There are those who experience carceral shock upon arrival -- finding themselves in a 9-square-metre cell can trigger transient carceral psychoses, anxiety attacks. There are withdrawals to manage urgently. There are judicial referrals: domestic violence, sexual offences. And then all those who present a vulnerability in detention and psychopathological disorders.

[G]: I primarily work with patients referred by the psychiatrist to work on verbalising emotions, as an alternative to acting out.

[G]: Following my M2 thesis on anger -- catharsis or pathology? -- I became interested in an Australian study showing that metal music could produce a purging of anger. In practice, when patients arrive angry, threatening hetero-aggressive acting out, I play them metal -- and it calms them down. Gradually, the team was able to observe the effectiveness of this approach.

[M]: This connects with the research on emotional regulation: an external stimulus congruent with the internal state allows the person to attune to themselves and to regulate.

[G]: Exactly. I had a very angry female patient for whom I played a Falling In Reverse video. Her reaction: "It's more violent than what I feel inside." This comparison calmed her down -- she had no further hetero-aggressive incidents after that.

[G]: Some patients who have phones in their cells use this music on their own initiative when they are angry, to avoid hitting a cellmate. They have internalised the regulation mechanism. And they tell me that outside too, they will use it.

[M]: How does artificial intelligence come into play in this context?

[G]: One of my patients, incarcerated for domestic violence, was in a relationship with another detainee. She leaves him. It's 3 a.m., his cellmate is asleep, he has psychological vulnerability, he wants to die. He had been told that if he needed to talk, he could use ChatGPT. He went on ChatGPT to pour out his sadness.

[G]: The next day, he told me: "I wanted to die, I talked about it to ChatGPT, I didn't die precisely because I talked about it. I found that listening ear. When I feel ready, I will bring you what I discussed with ChatGPT, but you gave me a way not to die, and yet I had the rope in my hand."

[G]: This is not care -- it's a listening ear, a momentary aid. I am careful to make this distinction with my patients. But this reassuring and supportive quality kept this patient going until morning and prevented him from acting on his suicidal thoughts.

[G]: That said, I have also seen the other side. A friend who is a YouTuber has completely isolated himself with Grok, which he considers his only friend. But there was a pre-existing depressive condition that had not been treated. The question is: is Grok not an amplifier of an underlying depressive condition? In all the studies I have read on AI-induced psychosis or AI-related isolation, there was already an underlying psychopathology.

[M]: Would AI be a revealer rather than an inducer?

[G]: These are profiles that have not yet decompensated -- for example, latent psychoses maintained in society that find a way to isolate via AI. But this is nothing new. The same phenomenon was found on early internet forums and on Discord.

[M]: Before forums, it was perhaps isolating oneself with one's pets, one's cats. The form changes, but is the substance really so different?

[G]: I think the underlying mechanisms are the same -- it is the form that changes. AI is so new and subject to this novelty effect -- and novelty always triggers fear. The critiques of AI show that there is a psychological life around this technology. It will level out when it becomes the norm.

[M]: You mentioned using AI to generate music. How do you use it concretely with your patients?

[G]: The protocol: I ask the patient for six words. Suno generates two texts from these words, and the patient chooses the one that resonates with them. Then they choose a musical style. Suno produces several versions. We listen to them together and rank them.

[G]: What is therapeutically powerful is that the same text, set in different styles -- rap, folk, metal, classical -- is not perceived in the same way. Patients discover that their words "sound" differently depending on the context. And the parallel with everyday life comes naturally: just because someone says a word doesn't mean it signifies what you think. We work on the difference in tone and the perception of the other.

[M]: That's remarkable: by varying musical styles for the same text, the patient experiences in their body that the same theme is felt differently depending on the interpretation.

[G]: The process is structured: patients reflect on their words, choose their text, decide whether they want to integrate elements from one version into another -- they position themselves in relation to their creation. I have patients who are beginning to say: "There, I can hear the guitar soaring in the opening." They pay much more attention to what a melody is and how it changes their perception.

[G]: The case that struck me the most is a patient who has been going in and out of prison since the age of 14-15, severely polysubstance-dependent -- MDMA, ecstasy --, severely deprived cognitively and in his emotional regulation. I began using Suno with him so he could create his own techno music. During his last incarceration, he told me he had written down the occupational therapist's username on Suno to find his creations on the outside -- he had not found them, but he had continued creating on his own. He was typing his own lyrics, with spell check. He was choosing nuances of style: "No, that's Hard-Tech, but I should have put Tech-Trance because it doesn't match what I want."

[G]: This patient began buying himself clothes, taking care of himself. What I told the psychiatrist: he still talks to us about his substance use, but now there is a process of desire and pleasure outside of the substance. He had told me that drugs, for him, were "a slow-motion suicide". Now, he is beginning to take pleasure in a creative activity. He is building an identity as a music creator.

[M]: AI creates stepping stones sized to the person: it meets the person where they are, with what they can give, and produces a result that exceeds what the person thought they could do.

[G]: We could have enrolled him in djembe classes, but the learning curve would have been too demanding for his abilities. With Suno, the approach is easy and the output immediate -- like a cocaine hit, in a sense, except that this hit allows you to produce things. And there is still work on the reality principle and frustration tolerance: the result does not always match what was imagined. The patient learns to take the time to produce, at their own pace.

[M]: The common criticism is that "AI does everything for us". But what you describe is that there is inevitably a gap between what you ask and what the AI produces. There is immediacy, but also frustration -- which seems tolerable.

[G]: AI is an imperceptible third party. Before, we had the third party of material -- clay, rattan --, the third party of equipment, the third party of software. Now, we have the third party of AI: we don't know how AI interprets our concepts, but we know that it is AI doing it. There is this unknown in the shaping, and that creates a third party. It is a third party of a new kind in psychiatry: not concrete, somewhat abstract, computational, nuanced. We don't yet know how to name it, but this third party exists.

[M]: You mentioned ongoing research?

[G]: I have two patients with negative symptoms who have completed self-assessment questionnaires on the use of Suno. It's not a large sample. For the substance-dependent patient, I created a specific protocol via ChatGPT -- I haven't yet managed to get him to complete it, but when a substance-dependent patient tells me he takes pleasure in creating, that seems promising to me and would deserve more rigorous investigation.

[M]: Is AI also a tool in the service of the therapeutic relationship?

[G]: Yes. The time I save on information retrieval, I reinvest in the exchange with the patient. For example, a patient wanted to know when Guns N' Roses had played in concert in Bordeaux. ChatGPT gave me the dates instantly. The patient was able to tell me his memories, to consider making the effort to go to Paris this summer. It is a narrative that takes shape -- what do you like about this band, what is the connection to your adolescence?

[G]: What I observe is that through music and AI, patients talk about themselves as human beings. In prison, they are reduced to their offence. In psychiatry, they are reduced to their disorders. With these mediations, they reclaim their human identity -- and that is just as important as everything else.

[G]: I am also beginning to explore image and video generation. Starting from the thumbnails that Suno provides for songs, I use Sora or Grok to animate them. When these tools have evolved sufficiently, we will be able to generate a video from a text -- that will be another dimension of therapeutic work.

[G]: It is also important to show patients how to use these tools for free or at low cost on the outside. A psychotic patient is not going to pay 5 euros a month to generate songs. You need to give them tips so they can continue creating after their release.

[M]: Do you have any precautions or contraindications to mention?

[G]: What seems essential to me is psychoeducation around AI. I explain to my patients how AI works, the fact that conversational AI is a momentary aid, not care. It's like medication: normally, you explain the side effects and obtain the patient's consent. For AI, it's the same thing.

[G]: There is also the question of vulnerable populations in the broadest sense -- not just young people, but also the elderly, isolated people. What is needed is to think about what we mean by vulnerability in the face of a new technology, and how to provide guidance rather than prohibition.