Psychotherapy Artificial Intelligence

Therapeutic Computational Creativity

At a glance: Framework proposed by Alison Pease and collaborators (2022) for thinking about the integration of creative AI — the kind that generates images, music, texts — into therapeutic settings. The central idea: AI can play the role of a digital “third hand,” facilitating the patient's creative process without replacing the therapist. This is not a therapeutic method, but a framework for deciding when, how, and under what conditions to use creative AI in sessions.

Relationship with computational creativity

Computational creativity (Boden, 1990) provides the general theoretical framework — what is computational creativity? This resource covers its clinical application — how to use it as a therapeutic tool? One illuminates the theory, the other guides practice.

Why This Concept Matters

Your patients are already using creative AI — Midjourney to draw, Suno to compose, ChatGPT to write. Some do it spontaneously, others could benefit from it in sessions. But how do you decide if it's relevant, and how do you frame its use?

The literature on AI in mental health focuses on chatbots (Woebot, Talkspace AI) — conversational AI. But creative AI is different: it produces artifacts (images, music, texts) that become shared therapeutic material, like clay or paint in art therapy. This framework fills that gap.

1. Distinguish creative AI from conversational AI: a therapeutic chatbot and a music generator don't raise the same clinical questions or ethical challenges
2. Position AI within the therapeutic relationship: neither a replacement for the therapist nor a simple gadget, but a creative third party serving the patient's process
3. Make informed decisions: have criteria for knowing when creative AI helps and when it hinders therapeutic work

The Digital “Third Hand”

In art therapy, Edith Kramer (1971) defined the concept of the third hand: the therapist intervenes discreetly in the patient's creative process — a brushstroke at the right moment, a color suggestion — without imposing their vision or diverting the meaning of the creation. It's facilitation, not direction.

Pease and collaborators propose that creative AI can play an analogous role:

Classic third hand

The therapist facilitates the patient's creation through discreet gestures, without imposing their own expression. The process remains the patient's.

Computational third hand

AI generates creative material from the patient's words or intentions. The human therapist remains the “container” of the relationship; AI is a tool in service of expression.

Why this matters for clinicians: this analogy anchors creative AI within a recognized art-therapeutic tradition. It is not a technological gimmick but a conceptual extension of an established practice. This makes the setup intelligible for practitioners trained in artistic mediations, even if they know nothing about AI.

Four Dimensions for Framing Use

Pease's framework proposes four axes for thinking about integrating creative AI in therapy. These axes are not steps but sliders that the clinician adjusts according to the patient and the therapeutic moment.

1. Co-creation or curation?

AI can be a co-creative partner (the patient provides words, the AI generates a song) or a curator (the AI selects and personalizes visual/auditory stimuli for therapeutic purposes). Between these two poles lies a spectrum of configurations.

In practice: Suno AI in co-creation mode (the patient writes, the AI composes) vs. an AI-generated playlist for relaxation (curation). The clinical implications differ: co-creation mobilizes patient agency, curation offers passive scaffolding.

2. Productive unpredictability

Creative AI never produces exactly what is expected. This unpredictability — often seen as a flaw — is here identified as therapeutically productive. It creates a space of surprise, of letting go, and works the central control/non-control tension in psychotherapy.

In practice: A patient types “my anger” into Suno and gets a gentle folk song. The surprise destabilizes them — then invites them to wonder why this interpretation touches them. Gaëlle Charlot calls this the “non-perceptible third”: an agent whose production you don't control, which forces you to take a stance.

3. Autonomy gradient

Where to place the slider between supportive tool (AI assists, therapist orchestrates) and autonomous agent (AI conducts elements of the session)? The framework recommends keeping the human therapist as pivot: AI remains in the role of tool or co-creator, not autonomous agent.

In practice: The therapist chooses when to introduce AI, which tool to use, and how to exploit the result. AI never decides the therapeutic direction.

4. Accessibility vs. creative struggle

Creative AI massively lowers the barrier to entry: a patient who “can't draw” can generate an image. But in art therapy, the difficulty of the creative process is sometimes the therapeutic vector. If AI does everything, the benefit of “creative struggle” disappears.

In practice: For cognitively impaired or severely struggling patients, AI accessibility is an asset (no failure experience). For others, maintaining a degree of creative frustration is therapeutically necessary. The clinician assesses case by case.

Two Clinical Illustrations

Suno AI in forensic psychiatry (Charlot, 2025)

At the SMPR in Bordeaux-Gradignan, Gaëlle Charlot uses Suno AI in occupational therapy with incarcerated patients. A substance-dependent patient, failing with traditional mediations, creates songs from a few keywords. He progressively develops an identity as a “music creator.” Charlot varies musical styles for the same text, working on perception and emotional differentiation.

Framework reading: co-creation (patient writes, AI composes) + productive unpredictability (the result always surprises) + accessibility (no musical skills required) + minimal AI autonomy (therapist orchestrates each session).

DeepThInk — AI-assisted digital art (Du et al., 2024)

An artistic creation system developed with art therapists over 10 months of co-design. Uses the ETC model (Expressive Therapies Continuum) to structure AI/patient interactions at different expression levels: kinesthetic, perceptual, cognitive, symbolic. The most rigorous empirical study applying these principles to date.

Framework reading: curation (AI adapts stimuli to the patient's expression level) + calibrated autonomy gradient (AI guides, therapist supervises) + creative struggle preserved (system doesn't generate in place of the patient).

Therapeutic Chatbot ≠ Creative AI

An AI chatbot providing therapeutic text responses (Woebot, Talkspace AI) does not fall within this framework. The difference is fundamental:

Conversational AI

• Verbal responses
• No artistic co-production
• No productive unpredictability
• Simulates human interlocution

Creative AI (this framework)

• Generates artifacts (images, music, text)
• Shared patient-AI co-production
• Unpredictability structurally integrated
• Extends artistic mediations

Points of Caution

This framework does NOT claim that:

All use of creative AI in sessions is therapeutic — the clinical setting, therapeutic relationship, and practitioner judgment remain decisive
Creative AI can replace the art therapist — it is explicitly positioned as a tool under human supervision
Co-creation with AI is equivalent to manual creation — the embodied and sensory dimensions of creation (clay modeling, painting) are not digitally reproducible

Limitations to keep in mind:

Tool dependency: the goal is for the patient to access their own capacities for expression, not to become an expert user of Suno AI
Agency: if AI “does everything,” the creativity belongs to the AI, not the patient. Maintaining active patient involvement is essential
Confidentiality: texts and images generated by the patient pass through commercial services (OpenAI, Suno). Sensitive clinical data must not be entered into these tools
Cost and access: most creative AI tools are paid services. Who funds them? The patient, the institution, the therapist?

Other Perspectives

Pease's framework belongs to an Anglo-Saxon art-therapeutic tradition. Other perspectives complement the reflection.

Winnicott: Digital transitional space

The transitional object (neither me nor not-me) illuminates what Charlot calls the “non-perceptible third”: the AI-generated result is neither entirely the patient's nor entirely the machine's, creating a potential space in the Winnicottian sense.

For clinicians: If your patient says “that's my song” about a Suno production, they are investing a transitional space. This is clinically significant.

Phenomenological critique: The absent body

Digital creation does not engage the body like clay modeling or painting. For Merleau-Ponty, creativity is embodied. What AI gains in accessibility, it loses in sensory engagement.

For clinicians: For patients where the relationship to the body is central (eating disorders, dissociation), prioritize embodied mediations. Creative AI is better suited for patients where the body is an obstacle to expression (motor disability, inhibition, cognitive deficits).

Therapeutic computational creativity is an orientation framework, not a protocol. Its value lies in providing guideposts for pioneering clinicians experimenting with creative AI — not in replacing clinical judgment.

This Concept in Our Tool Pages

Therapeutic computational creativity is concretely embodied in generative AI tools used in sessions.

Suno

Direct application case — music creation as creative mediation in forensic psychiatry