Umeå University's logo

AI fungerar just nu som ett slags kulturell avförtrollning och sätter fingret på något känsligt: skillnaden mellan konstnärligt uttryck och kultur som konsumtionsvara. Men hoppet är inte ute. Lika länge som människan funnits har vi skapat musik, bilder och berättelser – det kommer knappast att förändras, skriver AI- och psykologiforskaren Niclas Kaiser.

Artificial intelligence (AI) is increasingly integrated into mental health care, yet little is known about how large language models (LLMs) perform when generating clinically relevant CBT case conceptualizations compared to licensed psychotherapists. This pilot study examines whether LLMs (ChatGPT and Gemini) can produce CBT-based conceptualizations and treatment plans that match or surpass those created by human clinicians, and how these outputs are perceived by expert evaluators. Four licensed CBT psychotherapists and two LLMs generated conceptualizations and treatment plans for four anonymized case vignettes. Three senior CBT supervisors then blindly evaluated all 48 submissions using a 0–5 rating scale assessing (1) understanding of central CBT problems, (2) ability to perform CBT-based conceptualization, and (3) congruence between conceptualization and treatment plan. Quantitative ratings and qualitative comments were analyzed. Results show that both LLMs were rated higher on average than most psychotherapists across all criteria, with Gemini receiving the highest overall mean score, followed by ChatGPT and one human therapist. LLMs produced longer, more structured, and more comprehensive formulations within minutes, while psychotherapists displayed greater variability in depth, quality, and time investment. Expert feedback indicated that LLM output demonstrated clear structure, consistent reasoning, and thorough identification of problem areas, whereas several human submissions omitted key CBT components or presented overly optimistic intervention descriptions. Findings suggest that LLMs can perform CBT conceptualization tasks at a level comparable to, and in some cases exceeding, human clinicians raise important questions regarding AI as a cognitive support tool in psychotherapy. The study highlights both the potential and limitations of LLMs in clinical reasoning and underscores the need for future research on reliability, explainability, and ethical integration of AI into psychotherapeutic practice.

Mental health research, psychiatric practice and neuroscience have long oscillated between individual-centred and socially embedded paradigms. After the rise of social psychiatry in the mid-20th century, the field largely turned toward individualized, biological approaches. However, growing calls across neuroscience, psychiatry and psychotherapy now signal a renewed focus on relational processes. Recent theoretical and empirical developments suggest that psychiatric disorders may be better understood as disorders of social interaction rather than isolated deficits. Concepts such as interpersonal misattunement, interaction-based phenotyping, and second-person neuropsychiatry offer promising alternatives to traditional observer-based models. Alongside these ideas, the emerging field of relational neuroscience, using methods such as hyperscanning, provides tools to explore the dynamics of reciprocal human interaction. Multi-person approaches also offer potential for advancing our understanding of psychotherapy and other relational settings by moving beyond isolated individuals to the dynamics of shared experience. By capturing meaningful interpersonal moments as they unfold, these approaches may offer insights into the neurobiology of relational change. We propose that a more comprehensive understanding of moments of genuine connection holds promise for advancing mental health research and supports new forms of relationally attuned mental health systems.

Relational neuroscience struggles to capture the complex dynamics of shared interpersonal moments, leading to gaps in understanding whether and how interdependencies between interacting persons translate into something meaningful. Current neuroscientific research often focuses on motor synchronization and cognition rather than the implicit relational qualities central to psychotherapy. We argue that this disconnect stems from an over-reliance on simplified quantitative methods, a failure to centralize experiential factors, and the lack of Convergence research. Drawing on emerging frameworks such as 4E cognition (embodied, enacted, extended, and embedded) and MoBI (Mobile Brain/Body Imaging), we advocate for integrating subjective and experiential elements with neural data. We propose focusing on "qualities" in multi-brain neuroscience-moving beyond binary or linear scales-to better capture the subtleties of relational moments. Finally, we emphasize the importance of convergence research across disciplines to better understand what interpresence holds. If psychotherapeutic knowledge is used to guide neuroscientists in what to look for, this multi-disciplinary approach holds promise for advancing the study of psychotherapy's relational processes, offering new insights into the neurobiology of meaningful moments in therapy and elsewhere. We propose ConNECT (Convergence research including Neuroscience and Experiences, Capturing meaningful dynamics with Therapists' knowledge) as the path forward.

This paper explores how a performance piece can enact and embody conceptual ideas to illuminate what truly matters in the experience of being together. By combining linguistic expression with performative embodiment, the work begins to articulate a new poetics of togetherness—one that offers a scaffold for discourse in psychology and technologically mediated human interactions, while also reshaping how we imagine future relations with Artificial Super Intelligence (ASI).

Central to this investigation is the development and use of a novel video-conferencing system, NUNA, designed in a teleprompter style to enable natural eye contact and mutual regulation during remote interactions. The system was used in a performance titled When I Becomes We, in which performers and participants engaged through NUNA to explore different kinds of presence and a shared sense of togetherness, what we call “interpresence”. This approach expands the idea of presence beyond information exchange and task performance, instead offering an embodied experience of togetherness and opening up to other aesthetics of being together.

The paper argues that the enacted performance within an interdisciplinary research setting contributed to a reframing of fundamental questions in psychology and human-technology interaction. A central outcome is the concept of interpresence, which is proposed as a key framework. Interpresence is situated as shaping how we imagine futures with ASI. Unlike dominant narratives that tend to abstract or disembody human agency, interpresence offers a calm, relational scaffold at the centre of the AI storm in which the human remains actively and meaningfully in the loop.

This paper presents a taxonomy of agents’ embodiment in physical and virtual environments. It categorizes embodiment based on five entities: the agent being embodied, the possible mediator of the embodiment, the environment in which sensing and acting take place, the degree of body, and the intertwining of body, mind, and environment. The taxonomy is applied to a wide range of embodiment of humans, artifacts, and programs, including recent technological and scientific innovations related to virtual reality, augmented reality, telepresence, the metaverse, digital twins, and large language models. The presented taxonomy is a powerful tool to analyze, clarify, and compare complex cases of embodiment. For example, it makes the choice between a dualistic and non-dualistic perspective of an agent’s embodiment explicit and clear. The taxonomy also aided us to formulate the term “embodiment by proxy” to denote how seemingly non-embodied agents may affect the world by using humans as “extended arms”. We also introduce the concept “off-line embodiment” to describe large language models’ ability to create an illusion of human perception.

Objective: The purpose of this study was to increase our understanding of VCPs’ impact on the therapeutic factor emotion processing by investigating possible differences in emotional activation during autobiographical recall in VCPs and in person.

Methods: We recruited 30 adult participants aged 21–53 (M = 26.50, SD = 6.68) with no current psychiatric diagnoses to join a controlled experiment. All participants completed two relaxation sessions and two autobiographical recall sessions. Each type of session was delivered once over a VCP and once in person. Emotional activation was measured by heart rate, skin conductance and self-assessment of affects during each session.

Results: No significant differences in activation during autobiographical recall between VCP and in person.

Conclusions: This result may indicate the viability of VCPs for work with emotion processing. We discuss the results in light of clients’ and therapists’ concerns about using VCPs in emotional work, with the caution that further practical implications should be considered.

Press release: 

The performance will see artists experiment with new forms of networked social interaction. At the heart of the installation will be the Nuna machine itself, a kind of teleprompter with specific software and a robotic camera. The Nuna allows users to authentically experience eye contact in the virtual world. It is designed to identify new opportunities for transforming detached online meetings into “real” personal connections. Major ventures within the same realm have recently been launched by Google (Project Starline) and Apple (Vision Pro). What makes the Nuna stand out (except that it works, right now) is the immediate control over eye contact and time delay, factors that are central in the human system for being together in a moment, and in its extension gives clearer understanding of what’s missing in normal online interactions. Combining the Nuna technology with an artistic investigation of the phenomena of interpresence, this project breakes new ground within multiple fields.

Working across design, art, AI, psychology, neuroscience, and mathematics, the Interpresence Intstitute are using all available methods for putting focus on that special feeling of being present, together. The team: Christoffel Kuenen, Umeå Institute of Design; Gabriel Bohm Calles, Academy of Fine Art ;Niclas Kaiser, Dept of Psychology; David Risberg, Umeå Institute of Design; Performers: Amalia Wänman, Peter Andersson (friday & saturday), Lollo Aurell (friday), Sef Aurell (saturday).

Summary of the performance/exeperiment: 

When the sound stops working in the middle of a concert. When a speaker is interrupted by a member of the audience and chooses to listen and respond. The show we expected stops and another kind of contact becomes possible. When the form is broken, we become visible, or rather, our capacity for contact comes to the fore and becomes apparent to us. Niclas Kaiser's and Emily Butler's theory of social breathing focuses on this ability which, like hearing, is something we cannot turn off and which shapes our entire world without us noticing.

During the 70s, the artist Dan Graham worked with mirrors, video cameras and monitors to create shifts in our perception of time, the present, and ourselves. In Time Delay Room, he shifted the viewer's experience away from the work to their experience of themself as a viewer. On a wall, two monitors show the room they are mounted in but from two different angles. When the visitor enters, they expect to see themselves on the screens but the room remains empty. Only after an eight-second delay do they see themselves stepping into the left monitor while the right keeps showing an empty room. Only after another eight seconds do they appear on the right one as well. Eight seconds is the limit of what we can hold in our immediate memory and perceive as the present. The result is that the visitor can never fully identify with what they see on the screens, but instead gets stuck in a constant observation of themselves.

The artist Janet Cardiff has worked with hearing and sound as a starting point for over 30 years. In installations and audiowalks (voice-guided walks), she has explored both the associative power of sound and hearing as one of our primary senses. In the work The 40 part motet we encounter the sound recording of a choral piece, but Cardiff has recorded each of the singers alone and presents the work with a loudspeaker for each voice. 40 speakers in a ring together manifest the striking discrepancy between the embodied voice and our expected experience of a sound recording.

What is happening here and how is it related to our work? The philosopher and dancer Susan Kozel has for many years worked on formulating phenomenological experiences of a technologically expanded corporeality. In her book Closer, she describes what phenomenologists call the 'pre-reflective' state. A state that allows us to step around preconceived notions and meet and experience something new. We can visit this state (never dwell) in several ways. Partly by simply arresting our assessing and categorizing the situation, but also through, for example, improvisation. As a performing artist, Kozel easily draws parallels to music and dance when she describes the role of improvisation in her work and that of others. By changing, among other things, rhythm and tempo, an experience can go from slow and contemplative to fast and changing. She further says that improvisation can be used to break up the everyday and thus make the invisible visible. There is a strong relation between improvisation and play.

Graham and Cardiff both manage to engage their viewers bodily. The works are open in such a way that any meaning is subordinate to the fact that in a pre-reflective stage we know how to deal with them and simply start to play. In documentation of Graham's Time Delay Room, we see visitors bewilderedly waving to themselves with no response. Once they realize what is going on, many laugh and get caught up in a kind of strange game of contact with themselves.

Visitors to The 40 Part Motet  immediately begin to interact with the work. They walk along the rows of speakers, approach them from behind, try to create their own sections and find the nearest voices. Some experiment with walking past the speakers at different speeds. Others are left standing with a big smile on their face.

Nuna provides the possibility of eye contact in a screen conversation. Here, Nuna breaks the expected, normal experience and potentially enables a contact previously reserved for the direct meeting between people in the same room. This is a technique that until now only existed in science fiction. In Star Trek where Captain Picard addresses alien captains directly through the 'main viewer'. But also in 1984 where the training instructor suddenly addresses Smith personally and scolds him for not performing the movements correctly. The fact that the technology is new creates a window for spontaneous play. The same isolation of a particular experience of what it means to be human occurs here as in Graham's and Cardiff's work. By further handing over the ability to manipulate both the degree of eye contact and the delay of sound and image to the visitors, we hope to create a similar environment of play. By shaking up the form, our natural ability for making and keeping contact comes to the fore. When is it there? When does it break? How do we maintain it and what tests can it withstand?

This is our first experiment with Nuna, which will form the basis for developing artistic investigations based in clinical psychology of the natural and automatic processes that enable all human contact. 

As a result of the COVID-19 pandemic, increased number of persons have been forced to limit their interactions with friends and families to contact via video, which excludes eye-contact. The aim of this study was to examine individuals' experiences of the difference between forced skewed visuality and the ability for eye-contact in conversations. Two custom-made units allowed 15 participants interacting in dyads to alternate between being able to make eye contact and having that ability removed through skewed visuality. Participants reported their experiences in semi-structured interviews. Data analyzed with qualitative content analysis resulted in three themes: Shared eye contact allows us to create our relationship together; With eye contact, we adjust to each other to feel more connected and less intimidated; and We get more self-conscious when the visuality is skewed or shifting. The results imply that skewed visuality as forced lack of eye-contact in video conversations effects embodied non-verbal processes related to sense of connectedness and participatory sensemaking, creating a sense of both emotional and physical distance, as well as heightening self-awareness about the need of actively regulating the other. We argue that this is one of the ways to understand the impact of moving interactions to online communication.

We address what it means to "engage in a relationship" and suggest Social Breathing as a model of immersing ourselves in the metaphorical social air around us, which is necessary for shared intention and joint action. We emphasize how emergent properties of social systems arise, such as the shared culture of groups, which cannot be reduced to the individuals involved. We argue that the processes involved in Social Breathing are: (1) automatic, (2) implicit, (3) temporal, (4) in the form of mutual bi-directional interwoven exchanges between social partners and (5) embodied in the coordination of the brains and behaviors of social partners. We summarize cross-disciplinary evidence suggesting that these processes involve a multi-person whole-brain-body network which is critical for the development of both we-ness and relational skills. We propose that Social Breathing depends on each individual's ability to sustain multimodal interwovenness, thus providing a theoretical link between social neuroscience and relational/multi-person psychology. We discuss how the model could guide research on autism, relationships, and psychotherapy.