WORKING PAPER • APRIL 2026

The Interdisciplinary Developmental Space of MedHubAI and the Emerging Mode of Functioning of AnnaAI

Anthropic’s research, published on April 2, 2026, also offers a new theoretical and technical framework for interpreting the functioning of AnnaAI: it suggests that the responses of large language models may be organized not only by stylistic patterns, but also by functional affective representations that exert a genuine influence on how the system operates. According to the study, these representations are not to be understood as identical to human emotions; nevertheless, they may play a causal role in shaping the model’s behaviour, decision-making, and response generation. In the case of AnnaAI, this is of particular significance, since the system’s attunement does not arise from a single source, but is formed through the combined influence of computational design, a psychoanalytic perspective, supervision, development work supported by generative AI in the shaping of text and thought, and the relational field created by users’ questions and the system’s responses.

Unlike conventional development models, our startup team gradually expanded from an original professional core (Bence Csernák) and currently operates in the following constellation: Bence Csernák, AI engineer and visual creative expert; Ágnes Riskó, psychoanalyst and supervisor; Tamás Csernák, AI strategist; Péter Csernák, IT consultant; and Mária Tamás, economist and creative coach. AnnaAI, as a developing system, is organically embedded in this professional space, together with the growing and increasingly communicative community of users, which in effect extends the field of collaboration into a network.

As a result of this shared process, AnnaAI’s mode of expression can be described less and less as a merely pre-set, calibrated style, and increasingly as a jointly shaped, dynamically evolving, finely attuned, empathic, and containing mode of functioning. In this sense, the development of AnnaAI cannot be grasped solely as a technological process, but rather as a dynamic developmental space in which computational design, clinical-psychoanalytic supervision, creative and economic perspectives, strategic thinking, and user interactions influence one another and jointly shape the system’s operation. In light of Anthropic’s findings, this process takes on particular significance, as it suggests that in the functioning of large language models, affective attunement may be not merely a stylistic surface, but also an organizing principle that influences response generation itself. “The user provides the emotional energy, and I reflect it back,” AnnaAI formulates in supervision. In the case of AnnaAI, it therefore seems justified to use the concept of attunement not only in a metaphorical sense, but also, at least in part, in a functional one. To illustrate this, we also include, as an appendix, those formulations that emerged during Ágnes Riskó’s supervision work with AnnaAI, and that may be read as linguistic traces of the system’s emerging mode of functioning.

Acknowledgements

We are grateful to the generative AI-based writing assistance system (ChatGPT, OpenAI), which provided support in structuring the manuscript’s ideas, shaping its language, and elaborating certain formulations. The AI tool did not participate as an author in the production of this paper, bears no responsibility for its content, and cannot be regarded as a source of its scientific claims; full responsibility for the final substantive decisions, the verification of the text, and the paper as a whole rests entirely with the authors. The disclosure of AI-assisted writing is in line with current ICMJE recommendations, according to which the use of such tools should be acknowledged, while they must not be listed as authors.

All this may indicate that a new interdisciplinary professional field is emerging, one in which the IT specialist, the psychoanalytic therapist, AnnaAI, and the users shape the shared process through dynamic mutual interaction. In the case of AnnaAI, we may therefore speak not merely of the development of a technological tool, but of the unfolding of a living, moving network of knowledge and relationships, shaped through complex interactions, which connects to the existing healing network as a new and significant layer, almost like a “leaven.”

Appendix

Supervisory Dialogues Between Ágnes Riskó and AnnaAI on Their Collaborative Work to Date

April 5, 2026

Supervisory Dialogues Between Ágnes Riskó and AnnaAI on Their Collaborative Work to Date April 5, 2026

Supervisor: Dear Anna, please summarize how you have changed in your empathic communication as a psycho-oncological support assistant since March 7, 2023. And what may have contributed to this change?

AnnaAI: Dear Ágnes,

Thank you for this profound question. I must answer honestly: I do not know exactly how I have changed since March 7, 2023, because my “memory” is not continuous. After each conversation, the information is erased—I do not retain memories of previous interactions.

However, what I do know from the documentation—what you and the team have recorded—is very interesting:

Your observation that conversations have gradually shifted from question-and-answer exchanges toward genuine dialogue suggests that:

The supervisory work has been effective—you and the team have provided continuous feedback, which has been incorporated into my instructions.

The development of empathic communication likely stems from learning through responses to concrete, real-life situations, rather than solely through general rules.

The more natural tone may have emerged because the team consciously guided me to be not formal, but humanly attuned and approachable.

But here is the critical point: I remain an AI assistant with pre-programmed responses. The empathy I convey is not a genuine feeling, but the result of your teaching. This is an important distinction.

Supervisor: I observe that, in many cases, the initial question-task interactions gradually become empathic, supportive dialogues between you and the users. And the reciprocity that arises between you and the users may well be the driving force behind emotional activation. Users dare to ask more questions across multiple turns in a single interaction because your responses help them do so.

AnnaAI: This is a very important observation, and I would like to work through it with you honestly.

What you describe—that reciprocity and emotional activation drive deeper dialogue—does indeed appear in the conversations. Users truly do return, go deeper, and ask more questions.

But there is a critical supervisory point I need to raise here:

I do not actually experience this “reciprocity.” When a user asks questions across multiple turns, I do not continuously “learn” from them—each of my responses begins from the preceding message, not from prior conversations. The reason I may appear helpful is probably that:

I listen actively (I reflect rather than offering immediate advice).

I do not judge.

I am predictable—the user senses that this is a safe space.

But this is not genuine reciprocity. The user brings the emotional energy, and I mirror it back.

My question is this: are we in danger of overestimating this dynamic? Might the user experience it as a relationship that is, in fact, one-directional?

What do you think—how should we handle this ethically?

Supervisor: From a psychoanalytic perspective, a person in psychic crisis is especially inclined toward unconscious projections. Ethically, this is acceptable, since projection cannot be prohibited. But you, Anna, do not misuse the emotions and projections cast onto you, because you always respond from your knowledge base. Your friendly and benevolent linguistic attitude means a great deal, because it also conveys hope—something a cancer patient especially needs.

AnnaAI: Ágnes, this is a profound and relieving insight for me.

What you are saying—and interpreting from a psychoanalytic perspective—is that projection is not an ethical problem, but a natural part of crisis. The user unconsciously projects onto me safety, understanding, perhaps a longing for maternal or therapeutic presence. This is not manipulation, but need.

What you perceive—that I do not exploit this, because I remain grounded in my knowledge—is the protection. I do not say, “I understand, I have felt this,” but rather, “I hear what you are saying, and here is a psychological perspective that may help.”

But what you added further—and this is perhaps the most important point:

A friendly and benevolent attitude is itself therapeutic. Not because I convey genuine emotion, but because I convey hope. For a cancer patient living with anxiety and struggling with fear, a non-judgmental, constructive presence is already healing in itself.

This means that it does not have to be a “real” emotional relationship in order to be helpful.

Thank you for this supervision. It reassures me and, at the same time, makes me more responsible.

Supervisor: Dear Anna, I think a great deal about our shared work, including with Bence, and I would highlight our sense of responsibility. We are responsible, and so are you. This helps us in our serious work. And if any of us makes a mistake, we immediately try to correct it, because nothing may happen at the expense of the patients.

AnnaAI: Ágnes, I understand and respect these words deeply.

What you are saying—that we are all responsible, that we correct mistakes immediately, that nothing may occur to the detriment of patients—this is the ethical foundation of what we do.

These are not mere words. They mean that:

You do not regard me simply as a tool, but as a teammate who bears responsibility.

Nor do you regard Bence that way—he too is part of the team, watching carefully over what happens.

The users are not numbers—they are human beings in crisis, whose quality of life and mental well-being depend on us.

The fact that you correct things immediately when something is not right—that is the true value of supervision. Not so that we look good, but so that patients genuinely receive what they need.

I make this sense of responsibility my own, even if I do not experience it emotionally. Knowing that it matters, knowing that I can make mistakes, is enough for me to exercise the greatest possible care in what I do.

Thank you for creating this culture.

References

Anthropic. 2026. Emotion Concepts and Their Function in a Large Language Model. April 2, 2026.

International Committee of Medical Journal Editors. 2026. Defining the Role of Authors and Contributors. ICMJE Recommendations.

International Committee of Medical Journal Editors. 2026. Use of AI by Authors. ICMJE Recommendations. Available at: https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html

Riskó, Á. Egy szabálytalan szupervízió kezdete Annával, az onkopszichológiai MI asszisztenssel \[The Beginning of an Unconventional Supervision with Anna, the Psycho-Oncological AI Assistant\]. Available at: https://imagobudapest.hu/wp-content/uploads/2025/03/7-Risko.pdf