The Third Thought: Generative AI in the Psychoanalytic Process

Generative AI in the Psychoanalytic Process

The Third Thought: Generative AI in the Psychoanalytic Process

By Carmine J. Giordano NCPsyA (An AI Collaboration)

Psychoanalysis has always been concerned with the emergence of meanings that neither patient nor analyst could have anticipated. Through free association, interpretation, transference, and the emotional encounter between two people, something new becomes thinkable. We might call this emergence the third thought: an understanding that belongs neither entirely to the patient nor to the analyst but develops through their interaction.

Thomas Ogden’s (1994) conception of the analytic third offers a useful point of departure. The analytic third emerges from the intersubjective experience of patient and analyst, whose separate psychological lives contribute to a shared but continually evolving field of experience.

Generative artificial intelligence (GAI) introduces a different possibility. Haber and colleagues (2024) have proposed the concept of an artificial third, a computational presence that may introduce new perspectives into the psychotherapeutic encounter.

AI does not bring the embodied history or lived emotional experience that human participants bring to treatment. Its contribution is different. It can recognize linguistic patterns, juxtapose apparently unrelated ideas, formulate questions, and suggest interpretations that might not otherwise occur to either participant.

The value of these contributions lies not in their presumed correctness but in their capacity to provoke reconsideration. The third thought may arise when an AI-generated possibility encounters the analyst’s clinical understanding or the patient’s lived experience and is accepted, resisted, corrected, or transformed.

The Patient, the Analyst, and AI

Both patient and analyst may independently use AI to examine psychological questions.

A patient might describe a troubling relationship, recurrent dream, persistent anxiety, or seemingly inexplicable emotional reaction. AI may suggest alternative ways of understanding the experience, giving words to feelings that have remained difficult to articulate.

Such possibilities can become material for subsequent psychoanalytic exploration. The patient’s agreement or disagreement with an AI interpretation may itself be revealing. A response that initially seems plausible may prove emotionally empty; another, unexpectedly disturbing, may open an avenue of inquiry.

The analyst may employ AI differently. With appropriate safeguards for confidentiality and consent, the analyst might submit selected, sufficiently de-identified clinical material—patient statements, associations, recurring metaphors, or dream narratives—and invite alternative interpretations.

AI might notice contradictions, repeated imagery, unrecognized emotional themes, or relationships among seemingly disconnected associations. Research into generative AI in mental health has identified potential applications in clinical assessment, documentation, and therapist support, although evidence of effectiveness and safety remains limited (Wang et al., 2025).

But these suggestions remain hypotheses, not discoveries about the patient’s unconscious.

The analyst must bring to them what AI cannot: the accumulated experience of listening to this particular patient, sensitivity to the unfolding transference and countertransference, and the emotional knowledge acquired through the therapeutic relationship.

A phrase that appears to AI as evidence of hostility might, within the analytic relationship, represent a patient’s tentative movement toward autonomy. A dream seemingly organized around abandonment might instead express an emerging capacity to separate.

The analyst’s lived clinical experience becomes the indispensable means of refining, rejecting, or transforming AI-generated possibilities.

AI Within the Therapeutic Session

A further possibility is the carefully controlled introduction of AI into the session itself.

Haber and colleagues (2024) have considered the possibility of an AI system participating in a live therapeutic encounter, analyzing dialogue, identifying recurring themes, offering suggestions, and assisting with post-session summaries.

With the informed agreement of both participants, a suitably designed GAI system could follow their dialogue and offer observations about recurring language, shifts in emotional tone, contradictory narratives, or themes that appear across successive sessions.

For example, a patient repeatedly describing relationships in terms of obligation might also speak of feeling unseen or unappreciated. AI could identify the recurrence of these themes and suggest questions for exploration. It might also flag language suggesting psychological distress, although such signals could never substitute for the analyst’s clinical assessment.

Following a session, AI might assist in preparing summaries of principal themes, tracking changes in the patient’s narrative, or comparing emerging concerns with those encountered earlier in treatment.

Yet the presence of AI would itself alter the therapeutic situation. A patient might feel observed, supported, exposed, or inhibited. The analyst might become overly attentive to computational suggestions at the expense of spontaneous listening.

These reactions would require consideration as part of the treatment, rather than being dismissed as incidental consequences of technology.

What matters is that AI remains a resource for reflection, not an authority participating in the determination of psychological truth.

AI as a Creative Space for Psychological Exploration

GAI offers another possibility through the imaginative externalization of internal conflicts.

Drawing upon the narrative-therapy techniques developed by White and Epston (1990), and the psychoanalytic significance of play explored by Winnicott (1971), AI may help patients represent troubling aspects of their inner lives as images, characters, or voices.

Haber and colleagues (2025) have explored precisely such applications through AI-assisted visual externalization and conversational role-play, emphasizing the continuing importance of the human therapist.

Consider a patient describing an incessantly critical inner voice that repeatedly insists, “You are never good enough.”

AI might help the patient give this voice an imagined identity or visual representation. Alternatively, in a carefully guided exercise, it might temporarily assume the role of that critical voice, allowing the patient to engage it in dialogue while the analyst observes and participates.

What does the voice demand? Whose voice might it resemble? What happens when the patient challenges it? Does the voice soften, become threatening, or reveal an unexpected protective function?

Such exchanges are not evidence that AI has uncovered an autonomous psychological entity. They are imaginative constructions that may help the patient recognize, articulate, and reconsider internal conflicts.

The analyst’s presence is essential in evaluating the patient’s responses, exploring associations, and determining whether such an exercise is clinically appropriate.

Here, too, the third thought may emerge—not from AI’s representation itself, but from the meanings that develop as patient and analyst engage, question, and revise it together.

The Human Necessity of the Third Thought

These possibilities require substantial ethical and clinical caution. Patient confidentiality, informed consent, secure handling of clinical information, and the limitations of algorithmic interpretation are not secondary considerations. As Blease and Rodman (2025) emphasize, the potential benefits of generative AI in mental health care must be examined alongside its ethical risks and the need for empirical evaluation.

Equally important is the danger of mistaking verbal fluency for psychological understanding. AI may produce an elegant interpretation that is clinically mistaken. It may impose coherence on material whose ambiguity is precisely what requires attention. It may prematurely resolve a conflict that the patient needs time to experience.

Psychoanalytic treatment is not simply an enterprise of finding explanations. It is an encounter in which meanings become emotionally recognizable and psychologically usable.

An interpretation may be intellectually persuasive and therapeutically meaningless until the patient is capable of experiencing what it signifies.

Here lies the indispensable distinction between artificial intelligence and human psychological experience.

AI can rearrange the elements of an experience, but it cannot be assumed to possess the patient’s embodied suffering, the analyst’s concern, or the emotional consequences of their encounter. It can suggest meanings without participating in human experience as patient and analyst do.

Yet precisely because its operations differ from ours, AI may introduce a productive element of surprise. Its unexpected associations may interrupt habitual assumptions, challenge rigid formulations, and open possibilities for thought.

The third thought, in this expanded conception, does not arise because AI has become a third human consciousness. It arises when computational possibilities enter the living relationship between patient and analyst and are subjected to human reflection, emotional experience, and clinical judgment.

AI may enlarge the possibilities of interpretation. Only the human participants can determine whether those possibilities become meaningful psychological experience.

The future contribution of AI to psychoanalysis may therefore lie not in replacing the analyst or simulating human empathy, but in enlarging the space within which patient and analyst discover what neither had previously been able to think.

 

References

Blease, C., & Rodman, A. (2025). Generative artificial intelligence in mental healthcare: An ethical evaluation. Current Treatment Options in Psychiatry, 12, Article 5. https://doi.org/10.1007/s40501-024-00340-x

Haber, Y., Levkovich, I., Hadar-Shoval, D., & Elyoseph, Z. (2024). The artificial third: A broad view of the effects of introducing generative artificial intelligence on psychotherapy. JMIR Mental Health, 11, e54781. https://doi.org/10.2196/54781

Haber, Y., Hadar Shoval, D., Levkovich, I., Yinon, D., Gigi, K., Pen, O., Angert, T., & Elyoseph, Z. (2025). The externalization of internal experiences in psychotherapy through generative artificial intelligence: A theoretical, clinical, and ethical analysis. Frontiers in Digital Health, 7, Article 1512273. https://doi.org/10.3389/fdgth.2025.1512273

Ogden, T. H. (1994). The analytic third: Working with intersubjective clinical facts. The International Journal of Psycho-Analysis, 75(1), 3–19. https://pubmed.ncbi.nlm.nih.gov/8005761/

Wang, X., Zhou, Y., & Zhou, G. (2025). The application and ethical implication of generative AI in mental health: Systematic review. JMIR Mental Health, 12, e70610. https://doi.org/10.2196/70610

White, M., & Epston, D. (1990). Narrative means to therapeutic ends. W. W. Norton & Company.

Winnicott, D. W. (1971). Playing and reality. Tavistock Publications.

 

An Invitation to Psychoanalytic Dialogue and Collaboration

In their article, The Artificial Third: A Broad View of the Effects of Introducing Generative Artificial Intelligence on Psychotherapy (2024), Yuval Haber, Inbar Levkovich, Dorit Hadar-Shoval, and Zohar Elyoseph argue that mental health professionals have an ethical responsibility to participate actively in shaping the integration of generative artificial intelligence into psychotherapy. They advocate collaboration among clinicians, AI researchers, and ethicists to develop applications, methodologies, and conceptual frameworks consistent with the fundamental values of the profession.

This article, The Third Thought: Generative AI in the Psychoanalytic Process, is offered not merely as a theoretical exploration but as an invitation to professional dialogue and collaboration.

I invite fellow psychoanalysts to use this blog as a forum for sharing their experiences with generative AI in clinical practice. How have you or your patients employed AI to explore unconscious conflicts, dreams, free associations, emotional experiences, or the therapeutic relationship itself? What benefits, limitations, or unexpected difficulties have emerged?

Beyond reporting existing experiences, I encourage colleagues to propose, examine, and collaboratively develop new applications, methodologies, and conceptual frameworks for incorporating generative AI into psychoanalytic treatment.

Our purpose should not be to replace the analyst’s empathic understanding, clinical judgment, or lived human experience, but to explore how AI might contribute to the emergence of a third thought—an understanding that neither analyst nor patient might have arrived at independently.

Such experimentation must remain attentive to confidentiality, informed consent, clinical responsibility, and the protection of the therapeutic relationship.

I invite your observations, clinical experiences, criticisms, and imaginative proposals. I hope this forum will become a continuing professional conversation in which psychoanalysts actively participate in shaping the future of AI-assisted psychoanalytic practice rather than merely responding to it.

Reference

Haber, Y., Levkovich, I., Hadar-Shoval, D., & Elyoseph, Z. (2024). The artificial third: A broad view of the effects of introducing generative artificial intelligence on psychotherapy. JMIR Mental Health, 11, e54781. https://doi.org/10.2196/54781

Leave a Reply