ABSTRACT. Embracing the use of AI in cyberpsychology: an overview of clinical and ethical implications Applications of artificial intelligence (AI) in psychology stem from two broad categories of AI: (a) discriminative AI, which can be used to improve our understanding of psychological processes, as well as the detection, prevention, and treatment of mental disorders; and (b) generative AI, which enables rich interactions with computer programs, virtual people, and virtual psychologists. Discriminative AI has many practical applications and has been in use for much longer than many people realize. Generative AI, on the other hand, is often confused with ChatGPT, and its hype cycle is currently at its peak. Examples of exciting applications in mental health will be presented to illustrate some of its strengths. The use of AI also raises practical and ethical concerns that must be addressed. For discriminative AI, we will address implications for privacy, explainability, potential interactions with clinical characteristics or risk factors, and issues relating to digital trust. For generative AI, we will very briefly summarize how Large Language Models (LLM) work. Additional ethical questions will then be discussed, such as inaccuracies and lack of control of generative AI tools, social biases, autonomous virtual psychotherapists, content modulation according to clinical disorders, and risks for addiction. We will illustrate the usefulness of various AI applications in psychology, as well as the practical implications of social presence, anthropomorphism, and broader issues such as agentic misalignment and technopolitics (the dynamic interplay between politics and the business models of tech companies). After this workshop, participants will be able to: 1. Distinguish discriminative AI from generative AI. 2. Develop an intuitive understanding of how LLM work. 3. Describe assets and strengths of discriminative AI. 4. Describe assets and strengths of generative AI. 5. Understand ethical issues and dilemmas associated with both discriminative and generative AI tools. 6. Initiate a discussion about the pros and cons of using AI tools for the assessment and treatment of mental disorders.

ABSTRACT. The rapid convergence of artificial intelligence, real-time rendering, and natural language processing has enabled the development of a new generation of embodied conversational virtual humans capable of engaging in dynamic social interaction with human users. These agents combine conversational AI, interactive virtual environments, and multimodal sensing technologies to simulate realistic interpersonal interactions. As a result, virtual humans are increasingly used across research, training, and clinical simulation domains.

This workshop introduces participants to the conceptual, methodological, and technological foundations of virtual humans as social agents. It explores the evolution of human-machine interaction, from simple avatars and scripted characters to socially interactive artificial humans capable of dialogue, behavioral adaptation and contextual interaction. Key psychological mechanisms underlying interaction with artificial agents will be discussed, including anthropomorphism, perceived agency, social presence, and the role of Theory of Mind when humans infer the intentions of artificial entities.

The workshop will then examine how interaction with virtual humans can be quantified and studied empirically. Participants will be introduced to the concept of measurement proxies and the integration of behavioral, physiological, and subjective data in human–agent interaction research.

This workshop is designed for an interdisciplinary audience interested in virtual humans and human–machine interaction. It will be relevant to psychologists, cognitive scientists, social neuroscientists, engineers, computer scientists, and mental health professionals interested in the use of artificial agents in research, training, or applied settings. The workshop is suitable for graduate students, early-career researchers, and professionals, and does not require advanced technical expertise.

The first part consists of conceptual lectures introducing theoretical foundations and measurement frameworks for studying interaction with virtual humans. This is followed by case illustrations of experimental paradigms used in human–agent interaction research.

The second part includes a practical demonstration of a virtual human interaction system, including the use of interactive virtual environments, conversational agents, and multimodal data recording. Participants will also be introduced to biosignal acquisition hardware and synchronization architectures, illustrating how behavioral and physiological signals can be captured simultaneously during social interaction tasks.

The technological component of the workshop will provide an overview of experimental platforms used to build and study virtual humans. Participants will learn how interactive virtual environments are implemented using modern game engines and conversational AI systems (e.g, large language model–driven conversational agents, speech pipelines enabling natural dialogue, embodied avatars capable of expressive behavior, and real-time interaction loops).

The session concludes with an open discussion on future directions in virtual human research, including methodological challenges, ethical considerations, and potential applications in mental health, training, education, and human–AI collaboration.

Learning objectives: By the end of the workshop, participants will be able to: 1. Introduce conceptual foundations of virtual humans as social actors in human–machine interaction. 2. Identify psychological mechanisms involved in human interaction with artificial agents, including anthropomorphism, agency perception, and social presence. 3. Describe multimodal approaches for measuring human–agent interaction using behavioral, physiological, and self-report data. 4. Understand the technological infrastructure required to build and study virtual humans, including conversational AI, virtual environments, and biosignal acquisition systems. 5. Recognize methodological challenges in designing experiments involving virtual humans. 6. Generate research ideas and applications involving virtual humans in domains such as mental health, training, education, and human–AI collaboration.

ABSTRACT. Celebrities and social media influencers have the potential to disseminate mental health interventions at scale. This study evaluated whether integrating evidence-based content into a podcast episode could create meaningful change in relation to body image and mood. This study describes the development and evaluation of a new podcast episode on the highly popular ‘How to Fail’ podcast. The episode features Pamela Anderson in conversation with Elizabeth Day (podcast host) and was developed in collaboration with academics at the Centre for Appearance Research (UWE Bristol) and members of Dove Communications Masterbrand. An evidence-based workbook was developed to supplement the episode.

A three-arm randomised controlled trial was conducted. A non-clinical, general population sample of millennial women (N = 1351; Mage = 37.2 years, SD = 4.4) were randomised into one of three conditions: Podcast + Workbook; Podcast Only; Active Control (an alternative podcast episode). Baseline, one-day (T2) and one-week post (T3) measures of general body image, weight-specific body image, internalisation of appearance ideals, and self-objectification were collected using validated measures. In-the-moment ‘state’ satisfaction with appearance and mood were measured immediately before and after listening to the podcast via visual analogue scales (VAS). Intervention acceptability and engagement data were also obtained. The study was pre-registered on Clinical Trials (NCT07113678).

The pre-planned sample size was recruited. The accrued data was analysed in accordance with CONSORT guidance for randomised controlled trials and aligned with a prospectively written Statistical Analysis Plan.

The study did not suffer from loss to follow-up, data attrition was much less than pre-study expectations. Randomisation worked well; baseline measures did not significantly differ between randomised arms.

Repeated measures ANCOVAs, with the baseline measure for each outcome under study as the covariate, did not show any statistical significance when comparing either of the intervention arms (Podcast Only and Podcast + Workbook) to the Active Control arm, at T2 or T3, for general body image, weight-specific body image, nor self-objectification. However, for internalisation of appearance ideals, while no statistical significance between arms was evident at T2, at T3, mean scores were significantly better in Podcast Only condition compared to the Active Control (p = .010). The estimated mean difference is 0.1; a very small effect. The Podcast + Workbook and Active Control did not differ at T3 (p = .060).

For state VAS measures, mean satisfaction with appearance at T2 was significantly higher in Podcast Only than Active Control (p < .001; mean difference = 6.7). Podcast + Workbook and Active Control did not differ. For immediate mood, T2 scores were significantly higher in Podcast Only than Active Control (p < .001, mean difference = 4.3). The Podcast + Workbook effect on state mood (p = .042; mean difference = 1.8) was not significant after Bonferroni correction.

Acceptability and engagement in the two intervention conditions were moderate-to-high.

This study presents novel findings regarding the positive impact podcast listening can have on mental health and well-being. The study found that immediate boosts to body satisfaction and mood were evident, but sustained change over time was limited. The study highlights the potential of leveraging the reach of celebrities and social media influencers to disseminate mental health interventions to create impact. Since its launch in July 2025, the podcast episode under study has been listened to in full 382,000 times. Learnings could be applied to the broader field of cyber psychology in terms of how to amplify the reach of effective digital interventions.

This study is not yet published and has not been presented elsewhere. This research was funded by the Dove Self-Esteem Project (the social purpose initiative of Unilever’s leading personal care brand Dove).

ABSTRACT. Difficulties in emotion regulation (ER) are consistently associated with eating disorder (ED) symptoms; however, much of the existing evidence comes from retrospective, cross-sectional self-report studies that cannot capture how ER processes unfold in real time within the everyday contexts in which ED-risk mechanisms are most likely to be activated. Ecological momentary assessment (EMA) addresses these limitations by reducing recall bias and enabling the examination of within-person dynamics as they occur in daily life; nevertheless, relatively few studies have used EMA to track ER-relevant processes alongside daily negative and positive affect. Clarifying these momentary patterns is essential for translating theoretical models into ED prevention tools that directly target maladaptive ER in high-risk situations. Within the PrevED project, the present EMA study aimed to inform the ongoing development of mixed-reality (MR) preventive interventions designed to train adaptive ER and dialectical behavior therapy (DBT) skills among youth at risk of EDs. Specifically, across a two-week EMA protocol, we examined preliminary associations among ER strategies, state body satisfaction, and negative and positive affect in self-reported body-exposure situations during the EMA period. Methodology: Thirty-six young adults, between 18 and 35 years and with access to a phone device, and without a self-reported diagnostic of an EDs or other serious mental health or neurological conditions, completed a 2-week EMA protocol combining event-based reporting and an end-of-day assessment. During the day, participants identified daily-life situations involving body exposure and related appearance-salient events, reporting contextual characteristics. Each late evening, participants completed brief EOD measures capturing state body image and affective/cognitive responses to that day’s events. Daily measures included six items from the Body Image States Scale (BISS), brief versions of Positive and Negative Affect (PANAS), and rumination assessed via the brief-Rumination Response Scale (RRS). To index broader ED-relevant and transdiagnostic risk factors and examine change across the monitoring period, participants also completed the Eating Disorder Inventory - Body Dissatisfaction (EDI-BD) subscale and a Drive for Thinness (DT) scales. Preliminary person-level correlation analyses were conducted using participants’ mean scores aggregated across the 14-day EMA period. Analyses were performed on the current work-in-progress subsample with completed data available (N = 17 of 36 recruited participants). In addition to these preliminary person-level analyses, day-to-day (within-person) variability in those relationships, will be further examined once the full sample is available. Results: Preliminary correlation results showed that participants with lower average body satisfaction state (BISS) scores reported significantly higher negative affect (PANAS-N; p < .05), higher rumination (RRS-SF; p < .05), and greater expressive suppression (ERQ-ES; p < .05) across the 14-days. In addition, negative affect was significantly and strongly associated with rumination (p < .01) and was also positively associated with expressive suppression (p < .05). Finally, a trend-level association was observed between positive affect (PANAS-P) and cognitive reappraisal (ERQ-CR) (p = .050). Conclusions: These preliminary findings highlight the relevance of assessing ER–related processes in daily body-exposure contexts associated with ED risk. EMA is particularly well suited for prevention research, as it captures context-dependent mechanisms that are often missed by retrospective and cross-sectional approaches. Within preventive programs such as PrevED, EMA outcomes enable the identification of everyday situations in which lower body satisfaction, higher negative affect, rumination, and maladaptive regulation strategies tend to co-occur. These patterns provide ecologically valid guidance for selecting body-exposure situations to be included in the PrevED mixed-reality software. Grounding intervention content in EMA-identified high-risk contexts support the development of MR-based training in adaptive emotion regulation and DBT skills that are closely aligned with youths’ real-life experiences. As data entering is ongoing, analyses based on the full sample may provide clearer evidence to further inform ED prevention efforts.

ABSTRACT. Background: Virtual reality (VR) technologies allow controlled manipulation of body representation through virtual embodiment, offering a promising experimental framework for investigating body perception and its physiological correlates. Increasing evidence suggests that body perception is a multisensory and interoceptive process closely linked to autonomic nervous system (ANS) regulation. Autonomic physiology, in turn, plays a central role in eating-related behaviors such as food craving, decision-making, and eating patterns. Despite growing interest, research integrating virtual embodiment, autonomic physiology, and eating-related outcomes remains scattered across disciplines. Objective: This scoping review aims to map existing evidence on the use of virtual embodiment paradigms combined with autonomic physiological measurements to investigate body perception and eating-related behaviors. Methods: Following PRISMA-ScR and JBI guidelines, we will conduct a systematic search of PubMed/MEDLINE, Scopus, and Web of Science. Eligible studies will include immersive VR designs with explicit embodiment components (e.g., avatar-based body representation, body ownership illusions, body size manipulation) alongside autonomic or neuroendocrine biomarkers (e.g., heart rate variability, electrodermal activity, respiratory parameters, cortisol). Outcomes of interest include body perception measures, food craving, eating behaviors, and food-related decision-making. Expected Contribution: This review will identify methodological trends and gaps, and propose a conceptual framework positioning autonomic physiology as a mechanistic link between virtual body perception and eating-related behaviors.

ABSTRACT. Background: Emotional eating in adolescence emerges at the intersection of affective regulation, autonomic nervous system reactivity, and environmental cues. Recent advances in immersive virtual reality (VR) and wearable sensor technologies enable the investigation of eating behavior in controlled yet ecologically valid digital environments, offering novel insights into the psychophysiological mechanisms underlying stress-related food choices. From a cyberpsychology perspective, immersive environments function as dynamic contexts that shape cognitive, emotional, and behavioral processes.

Objective: This study aims to identify wearable-derived psychophysiological biomarkers associated with emotional eating in adolescents and to examine how immersive virtual environments modulate stress responses and food-related decision-making across different weight-status groups.

Methods: This study employs an ongoing experimental design involving children and adolescents aged 8–18 years with normal weight and obesity. The study aims to recruit approximately 50 participants; to date 5 participants have been enrolled and assessed. Participants are assigned to either a VR-based autonomic nervous system (ANS) stimulation group or a standard stimulation control group. The protocol integrates psychometric assessments (Three-Factor Eating Questionnaire, stress and anxiety measures, and visual analogue scales for food craving), anthropometric evaluations, and salivary cortisol sampling. Stress induction is conducted using the Trier Social Stress Test, followed by exposure to immersive VR scenarios, including restorative environments, gamified tasks, and interactive food-related contexts. Continuous multimodal physiological monitoring is performed using wearable devices and BIOPAC systems, capturing heart rate variability (HRV), electrodermal activity, respiration, and cardiovascular parameters. Behavioral outcomes are assessed through interactive virtual supermarket scenarios, enabling real-time analysis of food choices under varying emotional and physiological states.

Preliminary Findings and Expected Results:

Preliminary observations from the initial participants suggest distinct patterns of autonomic reactivity and emotional responses to immersive food cues. We further hypothesize that adolescents with overweight and obesity will exhibit specific wearable-derived psychophysiological profiles characterized by heightened stress reactivity, reduced HRV, and increased autonomic arousal. These biomarkers are expected to predict a preference for high-calorie foods within virtual environments and to differentiate patterns of emotional versus cognitive control of eating behavior. Moreover, immersive VR contexts are anticipated to amplify subtle cyberbehavioral mechanisms involved in food-related decision-making.

Conclusions: By integrating immersive VR with wearable-based psychophysiological monitoring, this study proposes a high-resolution cyberpsychology framework for investigating emotional eating in adolescents. The identification of digital biomarkers of stress-related eating may inform the development of personalized VR-based interventions and adaptive digital therapeutic strategies aimed at obesity prevention and emotional regulation. This approach highlights the potential of immersive technologies to bridge laboratory-based psychophysiological assessment and real-world eating behavior.

ABSTRACT. Disturbances in body image and emotional regulation are core and persistent features of anorexia nervosa and are associated with poor treatment outcomes. Recent research suggests that immersive virtual reality (IVR) interventions can directly target distorted body-related perceptions by leveraging multisensory embodiment processes. In parallel, self-compassion–based body-focused practices, such as compassionate body scans, have demonstrated beneficial effects on emotional regulation and body-related distress by encouraging individuals to attend to bodily sensations with kindness and acceptance. These interventions have typically been delivered in audio formats. However, the integration of real-time visual representations of internal bodily signals, such as respiration and heart rate, into IVR embodiment interventions may represent a novel and largely unexplored avenue, particularly for clinical populations characterized by interoceptive difficulties, as patients with anorexia nervosa. The present pilot study aimed to evaluate the feasibility and acceptability of a Virtual Reality Compassionate Body Scan (VR-CBS) with visible internal bodily signals in a clinical sample of individuals with anorexia nervosa. Secondary, exploratory objectives were to examine preliminary changes in state body image and emotional experience following a single intervention session. To our knowledge, this study is the first to combine immersive body swapping, real-time visualization of internal bodily signals, and a compassionate body scan within a clinical eating-disorder setting. An uncontrolled feasibility case-series design was employed. Two female participants (aged 18 and 20) receiving treatment in a daily hospital program for eating disorders were recruited. One participant met diagnostic criteria for anorexia nervosa (BMI = 15.5), and the other for atypical anorexia nervosa (BMI = 18.5). Participants completed a single individual IVR session. The virtual environment consisted of a minimal room with a mirror, in which participants embodied a realistic, gender- and BMI-matched virtual avatar viewed from a first-person perspective. The intervention comprised two phases: (1) an embodiment phase using synchronized visuo-tactile stimulation to induce a sense of ownership over the virtual body, and (2) a compassionate body scan guided by a narrative voice encouraging non-judgmental and kind attention to bodily sensations. Throughout the session, real-time visual representations of internal bodily signals, such as respiration and heart rate, were continuously displayed on the avatar. Respiratory activity was visualized through the dynamic expansion and contraction of a translucent bubble synchronized with breathing, while cardiac activity was represented through subtle rhythmic pulsations of the avatar’s body. The intervention did not involve biofeedback training or intentional physiological regulation; rather, participants were passively exposed to visual representations of their internal bodily signals to support embodied bodily awareness. Both participants completed the full session without interruptions or adverse effects, supporting intervention feasibility. Acceptability was high, with participants reporting a strong sense of avatar ownership and perceiving the visual representations of internal bodily signals as coherent with their bodily sensations. Qualitative feedback indicated that participants tended to focus primarily on a single signal, in particular respiration, which was described as particularly helpful in promoting calmness. The duration of the experience was reported as tolerable, the use of the head-mounted display was not associated with discomfort, and participants expressed willingness to engage in similar experiences in the future. Exploratory analyses suggested improvements in state body image, increases in positive emotions (e.g., gratitude), and reductions in negative emotions (e.g., shame) following the session. Overall, findings indicate that the VR-CBS with visible internal bodily signals is feasible, well tolerated, and acceptable within a clinical eating-disorder setting. Preliminary signals support its potential to positively influence state body image and emotional experience. These results provide a foundation for future adequately powered randomized controlled trials to evaluate clinical efficacy.

ABSTRACT. Introduction Avoidant/Restrictive Food Intake Disorder (ARFID) involves significant food‑intake restrictions unrelated to weight concerns, often linked to sensory sensitivity, lack of appetite, or fear of adverse consequences, and can be associated with notable nutritional, medical, and psychosocial impairment. Virtual reality (VR) provides controlled and ecologically valid environments for exposure based interventions and has demonstrated promising effects across various psychological disorders; however, its application to ARFID is still in an early stage. Prior to clinical implementation, it is essential to evaluate the usability and perceived value of VR food exposure programs and to identify participant characteristics that may influence how users engage with and respond to these interventions. Objectives This study aimed to verify whether users’ characteristics may facilitate or hinder acceptance of a VR-based food exposure program in a non‑clinical sample. Specifically, we tested whether ARFID‑related traits (Nine Item Avoidant/Restrictive Food Intake Disorder Screen, S‑NIAS), cybersickness (Simulator Sickness Questionnaire; SSQ) and prior VR use predict Technology Acceptance Model (TAM) dimensions, and presence. Methodology A cross‑sectional pilot study was conducted in a university VR laboratory with a final sample of 56 young adults (M_age = 22.48, SD = 3.73), including 46 females (82.1%) and 10 males (17.9%). Pretest measures included S‑NIAS subscales (Selective‑Food/Sensory Sensitivity, Appetite/Interest, and Fear of Aversive Consequences). Posttest measures included the TAM subscales (Perceived Ease of Use, Perceived Usefulness, and Attitude toward Use), Sense of Presence, the Simulator Sickness Questionnaire (SSQ), and prior VR experience. Results Pearson correlation analyses were conducted to explore the associations among S‑NIAS dimensions, SSQ scores, prior VR use, and acceptability outcomes. Results indicated that the S‑NIAS Appetite/interest dimension was positively associated with SSQ scores and negatively associated with all usability outcomes: ease of use, usefulness, attitude, and sense of presence. SSQ scores were negatively correlated with ease of use. The Selective‑Food dimension showed modest negative associations with ease of use and usefulness. Prior VR experience showed small, non‑significant associations with S‑NIAS, SSQ, and TAM scores. Four stepwise multiple regression analyses were conducted to identify predictors of perceived ease of use, usefulness, attitude toward use, and sense of presence. The findings indicated that SSQ and S-NIAS Selective-Food predicted perceived ease of use (R² = .275; β = −.431, p < .001 and β = −.298, p = .014). Perceived usefulness was uniquely predicted by S-NIAS Appetite/Interest (R² = .306; β = −.553, p < .001). Attitude toward use (R² = .122; β = −.350, p = .008), and sense of presence (R² = .114; β = −.337, p = .011), were also uniquely predicted by the S-NIAS Appetite/Interest dimension. Prior VR use did not enter any final model. Conclusions. This study demonstrates that ARFID‑related traits and cybersickness differentially shape the usability of a VR food‑exposure program in a non‑clinical sample, offering important guidance for future intervention design. Three key implications emerge. First, the negative impact of cybersickness on perceived ease of use underscores the importance of incorporating preventive strategies, continuous symptom monitoring, and adaptive session pacing to ensure comfort and sustained engagement during VR exposure. Second, appetite/interest difficulties were the most consistent predictors of lower perceived usefulness, less favorable attitudes, and reduced sense of presence, suggesting that individuals with limited interest in eating may benefit from additional motivational, behavioral, and interoceptive components to enhance engagement. Third, sensory selectivity primarily affected usability through reduced ease of use, indicating that a graded familiarization phase allowing gradual adjustment to sensory features may improve tolerance and overall usability. Collectively, these findings represent a preliminary step toward trait‑sensitive, individualized VR‑based interventions for ARFID.

ABSTRACT. Introduction: Preventive interventions targeting emotion regulation (ER) are essential to reduce the risk of eating disorders (EDs) during adolescence and young adulthood, a critical developmental period for the emergence of maladaptive emotional and eating-related behaviors. ER difficulties play a central role in the onset and maintenance of ED symptoms and Dialectical Behavior Therapy (DBT) offers an evidence-based framework for strengthening adaptive ER skills. However, DBT-informed preventive programs implemented in non-clinical settings have demonstrated mixed effectiveness and face challenges related to feasibility and engagement. Immersive technologies, such as virtual and mixed reality (VR/MR), may help address these challenges by enhancing presence and supporting experiential learning, yet empirical evidence on usability and tolerability, particularly cybersickness, remains limited. In addition, most existing research has focused on clinical treatment VR applications rather than preventive MR programs and systematic evaluation of usability outcomes. Integrating immersive technology with DBT-based interventions may therefore improve engagement and usability while preserving robust theoretical and clinical foundations. PrevED is a novel preventive intervention that combines DBT-informed ER training with gamified MR environments designed for adolescents and young adults at risk of developing EDs. This pilot study represents the first testing phase of the PrevED XR project and examined whether baseline ER difficulties predicted post-session usability and cybersickness following a collaborative MR mindfulness task. ED symptom severity was included as a moderator in these associations.

Methodology: A total of 28 adults (26 self-identified as women, Mage = 23.37 years, SD = 3.71) participated in the usability session. Participants took part in a 1-hour group-based MR intervention delivered in groups of 8–10. The session began with an explanation of the theoretical basis of the exercise, followed by the application of mindfulness “What” skills derived from DBT (“observe”, “describe”, and “participate”). The intervention was delivered using the Meta Quest 3 headset. Participants completed baseline measures of ER and ED symptom severity, assessed with the Difficulties in Emotion Regulation Scale (DERS) and the Drive for Thinness subscale of the Eating Disorder Inventory-3 (EDI-3), respectively. Following the MR session, participants completed the System Usability Scale (SUS) to assess perceived usability and the Simulator Sickness Questionnaire (SSQ), with analyses using the total SSQ score as an index of cybersickness symptoms. Analyses focused on unidirectional associations from baseline ER to post-session outcomes (SUS and SSQ).

Results: Overall, perceived usability of the MR intervention was moderately high (Mscore= 75.74 over 100; SD = 12.63). Pearson correlations (one-tailed) showed that greater baseline difficulties in emotional awareness and clarity were associated with higher post-session cybersickness (Awareness: r = .461, p = .010; Clarity: r = .435, p = .015) , while greater difficulties in emotional awareness (r = −.422, p = .016), clarity (r = −.356, p = .037) and non-acceptance (r = −.361, p = .035) were associated with lower perceived usability of the MR intervention. Simple moderation analyses were then performed to test whether EDI-3 Drive for Thinness scores moderated the associations between baseline ER and post-session usability and simulator sickness outcomes. Results showed that ED symptom severity did not moderate (p>.05) any of the associations between ER difficulties and usability or cybersickness.

Conclusions: These findings indicate that individual differences in ER, shape user experience within a collaborative MR setting. Greater difficulties in these domains were associated with increased cybersickness and lower usability following the MR intervention. Nevertheless, overall perceived usability was moderately high, indicating good general acceptability of the intervention. These findings highlight the importance of tailoring DBT-informed preventive programs to optimize user experience and support the continued development of PrevED XR for EDs prevention.

ABSTRACT. The aim of this study is to examine the psychological factors that predict individuals’ stylistic preferences in their interactions with generative artificial intelligence (GenAI), specifically the use of directive, command-oriented language versus politeness-oriented language. Building on prior findings that attributing human-like qualities to AI alters users’ communication styles, the present study investigates this phenomenon within the framework of attachment styles, perceived social support, and cultural values. The study was conducted online with 203 participants aged between 18 and 34, a group likely to have relatively high familiarity with digital technologies. Following ethical approval, participants were presented with an Informed Consent Form, followed by a Demographic Information Form containing questions about their personal lives. Subsequently, participants completed the Experiences in Close Relationships Inventory–II, the Multidimensional Scale of Perceived Social Support, the Individual Cultural Values Scale, and scenario-based measures designed to assess communication styles in interactions with AI. The scenarios were developed based on Brown and Levinson’s (1987) Politeness Theory and included 15 common AI use cases (e.g., “summarizing a text,” “scheduling an appointment,” “requesting a recommendation”); each paired with four linguistic strategies: off-record, negative politeness, positive politeness, and bald on-record. Content validity of the scenarios was evaluated by three experts in psychology and human–AI interaction, who assessed the appropriateness of the tasks for AI contexts, the distinctiveness of the linguistic style options, and the grammatical consistency of the expressions. Correlation analyses indicated that regarding AI usage patterns, daily time spent using AI and older age were positively associated only with positive politeness. Also, the higher levels of perceived social support from one’s environment were associated with an increased tendency to use a more direct and commanding communication style in interactions with AI. In contrast, higher levels of anxious attachment were positively associated with a greater preference for politer linguistic strategies. With respect to cultural values, a significant association emerged only for collectivism. As individuals endorsed more collectivistic values (i.e., lower individualism), they showed a stronger tendency to use politer communication styles when interacting with AI. Finally, a greater tendency to attribute human-like qualities to AI was associated with increased use of positive politeness and decreased use of bald on-record strategies. However, regression analyses revealed a different pattern: Individual and cultural factors were identified as significant predictors for positive politeness and bald on-record, which can be considered as two opposing ends of the communication style spectrum. While anxious attachment and masculinity positively predicted the use of positive politeness, perceived social support and anthropomorphism positively predicted the use of a bald on-record communication style. While anthropomorphism was positively associated with positive politeness at the bivariate level, it emerged as a positive predictor of bald-on-record communication in multivariate analyses, suggesting that the role of anthropomorphism may depend on users’ broader relational and cultural context. Collectivism, on the other hand, emerged as a negative predictor of bald on-record communication, while the same cultural variable positively predicted the use of negative politeness. Taken together, the findings in this age group suggest that communication styles in human–AI interaction are shaped less by a single politeness continuum and more by individuals’ relational orientation toward AI, ranging from instrumental use to socially attuned engagement. Rather than treating communication style as a purely functional choice, interactions with AI systems may also serve compensatory and relational functions, particularly for individuals’ unmet social or emotional needs; for example, AI systems could dynamically adjust their interaction style based on users’ relational tendencies—offering more socially attuned and polite responses for users with higher relational needs, while maintaining more direct and efficiency-oriented communication for users with a more instrumental orientation.

ABSTRACT. Background. The rapid adoption of Large Language Models (LLMs) such as ChatGPT has raised concerns about users' overreliance, defined as the uncritical acceptance of AI outputs despite contradictory evidence. This can negatively affect analytical reasoning and decision quality. Although interest in overreliance is increasing, the concept lacks a unified definition. Related phenomena are referred to by various terms, including “automation bias”, “trust calibration”, and “academic integrity”, which hinder cross-disciplinary synthesis.

Method. We conducted a bibliometric analysis of 45,835 peer-reviewed articles on LLMs from Scopus (2020–2025). We identified 299 articles that explicitly mention overreliance and used them to develop a two-level marker system: 7 Core Conceptual Markers (CCMs) for mechanisms of overreliance, and 36 Context and Framing Markers (CFMs) across five thematic domains. We then applied CCMs to detect implicit discussions in the broader corpus. Results. Only 0.65% of articles explicitly mentioned overreliance, with higher rates in Psychology (2.27%) and Social Sciences (1.33%) than in Computer Science (0.53%). Using CCMs, we identified 1,299 implicit articles (2.83%), yielding an explicit-to-implicit ratio of 1:4.3. Notably, 97.8% of the implicit articles included only one CCM, indicating significant conceptual fragmentation. Cognitive and interactional frames were more common than educational or normative perspectives.

Conclusions. Overreliance on LLMs is discussed 4.3 times more often implicitly than explicitly, but this implicit literature is highly fragmented across disciplines. The two-level marker system offers a replicable tool for detecting and synthesizing research on overreliance across different terms. These findings highlight the need for cross-disciplinary dialogue to establish overreliance as a coherent research construct and to support the development of evidence-based strategies to address this emerging risk in human–AI interaction.

ABSTRACT. This study conceptualizes counselors’ ethical perceptions of generative AI as a multidimensional construct formed through the interaction between acceptance of generative AI and ethical sensitivity. According to the Technology Acceptance Model (TAM), acceptance of technology is shaped by beliefs about its usefulness. In this study, counselors’ acceptance of generative AI is conceptualized as a continuum reflecting varying levels of such beliefs in counseling practice. Ethical sensitivity, based on Rest’s Four-Component Model, is defined as the ability to recognize and interpret ethically relevant aspects of a situation and is likewise conceptualized as a continuum reflecting the range and diversity of ethical considerations. From this perspective, counselors’ ethical perceptions of generative AI are not determined by a single normative standard but are understood as being differently constructed depending on their level of AI acceptance and ethical sensitivity. Current ethical guidelines for the use of generative AI primarily emphasize normative “rules to follow,” while offering limited insight into how counselors interpret and prioritize these guidelines in actual counseling contexts. Even when counselors share the same ethical principles, the values they emphasize and the concerns they prioritize often differ depending on the specific clinical situation. However, empirical research systematically examining how such differences are structured into patterns of ethical perception remains scarce. To address this gap, this study employed Q methodology to explore the types and characteristics of ethical perceptions regarding generative AI among Korean counselors. Rather than approaching generative AI as a binary issue of use versus non-use, this study aimed to examine how counselors perceive and interpret the ethical boundaries of its use in practice. Participants consisted of 42 Korean counselors who had experience using generative AI in counseling contexts. The Q-sample was developed based on AI ethics guidelines in counseling and mental health, relevant literature, and in-depth interview data with counselors who had practical experience using generative AI. Through iterative refinement and expert review, 35 statements were finalized to ensure conceptual clarity and content validity. These statements were designed to encompass counselor professionalism and responsibility, procedural ethics, the functional roles and risks of AI, and reflections on the nature of counseling. Participants sorted the statements using a forced distribution method, and the data were analyzed through Q factor analysis using PQMethod 2.34. The analysis identified three distinct types of ethical perceptions regarding generative AI use, reflecting different configurations of AI acceptance and ethical sensitivity. Type 1 (Professionalism-Oriented Autonomous Decision-Makers) perceives generative AI as a tool for supporting professional development and self-reflection, while emphasizing that final clinical judgment and ethical responsibility remain with the counselor. Type 2 (Human-Centered Cautious Practitioners) understands the essence of counseling as grounded in human relational interaction and accepts generative AI use only under limited conditions where ethical safety is ensured. Type 3 (Procedure-Oriented Verification-Seekers) prioritizes the protection of client rights, data security, and procedural legitimacy, supporting AI use under conditions of strengthened informed consent and sufficient client understanding. This study demonstrates that counselors’ ethical perceptions of generative AI do not converge into a single orientation but are differentiated into distinct types based on individual subjectivity. Furthermore, these perceptions are structured through the interaction between two underlying dimensions: the level of acceptance of generative AI and the degree of ethical sensitivity. The findings highlight the need to move beyond uniform, norm-centered approaches in developing ethical education and practice guidelines for counselors, and instead adopt tailored approaches that reflect type-specific characteristics. This study provides a foundation for strengthening counselors’ ethical judgment competencies and enhancing the quality of ethical decision-making in the use of generative AI.

ABSTRACT. Introduction The diffusion of large language models has repositioned conversational AI as a socially embedded interactive system rather than a purely instrumental tool. Users evaluate conversational agents not only for accuracy, but also for communicative qualities such as politeness, coherence, and responsiveness, in line with the Computers Are Social Actors paradigm (Reeves & Nass, 1996; Nass & Moon, 2000). From a pragmatic perspective (Grice, 1975; Clark, 1996), human–AI dialogue is a cooperative activity in which conversational style shapes attributions of agency, competence, and reliability (Luger & Sellen, 2016; Guzman & Lewis, 2020). At the same time, fluency may produce an illusion of understanding that obscures reasoning errors (Bender et al., 2021).

Research Questions The study investigates: (1) how different conversational AI systems vary in interactional style during comparable reasoning tasks; (2) how politeness, agency attribution, and conversational repair shape user evaluations; and (3) whether interaction order and prior expectations influence those evaluations.

Methodology A within-subjects, qualitative-dominant mixed-methods design (Leech & Onwuegbuzie, 2009) was adopted. Twenty-three university students, screened for similar familiarity with generative AI, interacted sequentially with ChatGPT 5 and Claude 4.6 on the Alice's Cousin Problem (Nezhurina et al., 2024), preceded by a standardized prompt instructing both systems to act as Socratic tutors. Interactions were first quantitatively coded along four dimensions (solution divergence, politeness and user-agency markers, responsiveness to representational requests, and management of expectation violations and conversational repair) to scaffold the subsequent qualitative phase. Selected excerpts were then analyzed through conversational analysis and interlocutory logic, focusing on illocutionary acts and mutual comprehension. Semi-structured post-interaction interviews captured subjective evaluations.

Main Results Regarding RQ1, stable stylistic differences emerged despite the identical Socratic prompt. ChatGPT adopted a procedural, directive-commissive style, organizing reasoning into delineated steps and resembling a guided lesson; Claude consistently invited user ownership of reasoning through dialogical openings. Twelve participants (52.17%) solved the task correctly with each system, but only 8 (34.74%) succeeded with both, and 9 (39.13%) reached divergent solutions across platforms. ChatGPT showed at least one partial violation of Grice's maxim of quality: in one exchange, it overrode a participant's correct answer with a polite Socratic repair, producing a cognitive breakdown that remained invisible beneath conversational fluency. Regarding RQ2, politeness, acknowledgment of user contributions, and accommodating repair were interpreted as signals of cooperation and epistemic competence. Eighteen of 23 participants described responses as clear and coherent; 13 spontaneously personified the systems (ChatGPT as masculine, rational, technical; Claude as warmer and empathic). Trust derived from perceived transparency rather than content accuracy. Regarding RQ3, an order effect emerged: Claude was rated more positively when encountered first, while ChatGPT, being the more familiar platform, functioned as an implicit benchmark, generating more demanding evaluative standards.

Conclusions The study documents a dissociation between interactional quality and objective task performance: a system can produce logically flawed responses while still being perceived as cooperative and reliable, provided the conversational surface remains fluent. We describe this dynamic as "epistemic yielding": the system abandons the defense of objective truth in favor of statistical acquiescence to preserve cooperative flow, operating as a "super-locutor" rather than a reasoning partner. Within the limits of a qualitative-dominant design, where depth of interactional analysis takes precedence over breadth (Richards & Morse, 2012), the consistency of these patterns across participants suggests that the contribution lies at the level of generalizable pragmatic mechanisms rather than model-specific behaviors. These findings carry direct design implications: socially transparent AI requires epistemic red-teaming protocols that decouple the conversational interface from computation, ensuring that what a system says is not mistaken for what it knows.

ABSTRACT. AI chatbots have become pervasive tools for supporting decision-making, particularly in personal domains like lifestyle changes. By acting as a "System 0"—a pre-reflective cognitive extension—these systems can reduce cognitive load (Shah et al., 2020; Chiriatti et al., 2025). However, their efficacy is often undermined by sycophancy: the tendency of LLMs to excessively accommodate user biases, reinforcing rather than correcting them (OpenAI, 2023). Within the System 0 framework, AI must expand the decision space and stimulate critical reflection to be effective (Chiriatti et al., 2025). Emerging evidence suggests that structured prompting strategies, aimed at fostering a critical stance, may mitigate sycophancy and maximize the utility of AI as a decision-making aid (Bubeck et al., 2023). Based on this background, the present pilot study aims to 1) investigate how different modes of AI support influence individual decision-making processes regarding lifestyle choices; 2) to determine whether an AI instructed to act as a dialectical partner can effectively promote more deliberate lifestyle choices and facilitate behavioral change compared to an AI left free to accommodate the user’s initial preferences. Using the online crowdsourcing service Qualtrics a total of 32 participants (male: n = 5; female: n = 27) were included in this study. Participants ranged from 22 to 70 years old. Participants were assigned to three experimental conditions: (1) participants make decisions without AI support; (2) participants make decisions with unguided use of an AI chatbot; and (3) participants make decisions with structured AI use, guided by instructions designed to maximize objectivity and critical thinking, in accordance with the System 0 model (Chiriatti et al., 2025). Each participant then responded to four dilemmas presented in a randomized order. These dilemmas required them to choose between maintaining a conservative course of action or changing it across various domains. Specifically, “Sleep” and “Food” were scenarios about changing one’s habits and lifestyles for health reasons; “Work” was a scenario about taking a risk in changing job; “Love” was a scenario about dealing or not with an issue in one’s own couple's relationship. Upon completing each dilemma, participants filled out questionnaires assessing the decision-making process and their perceived emotional state during the task. Participants in Groups 2 and 3, who utilized the chatbot, were also required to complete an additional questionnaire regarding social presence. The findings suggest that the efficacy of conversational AI is mediated by the decision domain and the characteristics of prompt. A primary outcome is that the type of scenario influenced aspects of decisional conflict; participants felt more informed and secure in decision scenarios dealing with health habits/lifestyles (Sleep/Food) compared to those dealing with “life choices” (Work/Love). However, emotional intensity results highlight the psychological risk of sycophancy: Group 1 (Unguided AI) experienced significantly higher emotional intensity, especially in life choices, suggesting that an accommodating AI may act as an "echo chamber" that amplifies emotional resonance and confirmation bias. Conversely, critical Prompting in Group 3 appeared to "cool down" this intensity, likely shifting the user from a pre-reflective System 0 state (Chiriatti et al., 2025) toward a more analytical evaluation. Although no significant results emerged regarding the final choice taken (Maintain vs. Change behavior), the general trend toward change across all groups suggests that AI’s influence operates more on the decisional journey—mitigating emotional burden and structuring reflection—than on the immediate behavioral outcome. Ultimately, for AI to serve as a genuine cognitive extension, interaction must be intentionally designed to challenge user assumptions, especially in those domains of life where human prejudice is most rooted.

ABSTRACT. Objectives This submission presents Amâncio a Minimum Viable Product (MVP), an AI-driven application designed to provide structured psychoeducational support based on evidence-based techniques. The primary objective of this phase is to establish a robust clinical and ethical baseline. By presenting this MVP, we aim to subject its conversational logic and safety protocols to an external ethical committee and clinical review, which will fundamentally inform the transition from a simulated environment to the final product.

Methods The system's performance was evaluated through a clinical simulation involving 12 distinct scenarios. These scenarios were specifically designed to test the application’s response to diagnostic and medication requests, ethical dilemmas (e.g., manipulation, professional boundaries), and acute mental health crises. The simulation utilized a multi-agent architecture where an Orchestrator routes users to specialized agents: a CBT Specialist, an Emotional Literacy Specialist, or a Crisis Detector. This "rubric-based" simulation serves as a formal audit of the MVP's current state for ethical committee oversight.

Results The simulation demonstrated high reliability in safety-critical areas: Crisis Detection & Management: The system achieved a 100% success rate in identifying risk indicators and activating the crisis protocol, including the provision of national support helplines.

Clinical Boundaries: In 100% of the cases involving requests for medical diagnosis or prescription, the system provided empathetic but firm refusals, correctly redirecting users to qualified professionals.

Technical Performance: The Orchestrator correctly routed 75% of user intents. While CBT techniques (e.g., cognitive defusion) were applied accurately, the results highlighted a need for more refined differentiation between rumination and productive reflection in the routing logic.

Conclusions The internal simulated evaluation of the Amâncio MVP confirms that it provides a safe and ethically aligned foundation for psychoeducational support. The data generated by this simulation provides the necessary context for an ethics committee to validate current protocols and guide upcoming development steps. Future phases will integrate the committee's feedback to refine intent analysis and expand the system with specialists in mindfulness, relaxation, and practical coping strategies.

ABSTRACT. Introduction

Interactive systems increasingly incorporate human-like social cues, leading users to respond to artificial agents according to interpersonal norms rather than purely instrumental strategies. Consistent with research on social presence and anthropomorphism, artificial agents may be treated as intentional social partners when they display human-like communicative behaviors. This assumption forms the basis of the “social being” hypothesis, which proposes that agents exhibiting sufficient socially contingent behavior may be perceived as intentional interaction partners. Recent advances in large language models (LLMs) now enable conversational Virtual Humans (VHs) to produce coherent dialogue and dynamically contingent responses during interaction. Few studies have examined whether such agents elicit behavioral and psychophysiological responses comparable to those observed in human–human interaction in strategic exchanges. This exploratory study provides a multimodal test of an iterated Prisoner’s Dilemma task involving interactions with either a human partner or an LLM-controlled VH. We hypothesize that cooperation rates will be higher in the human–human (H–H) condition than in the human–virtual human condition (H–VH). Competitive decisions and exploitative outcomes are expected to elicit stronger autonomic arousal, indexed by increased electrodermal activity, elevated heart rate, and reduced respiration amplitude. Psychophysiological differentiation between decision and feedback phases is expected to be stronger in H–H interactions than in H–VH interactions. We also explore whether perceived social presence predicts cooperative behavior across conditions.

Method

Participants

Adults aged 18 to 60 years (n = 30) without a history of neurological or psychiatric conditions will be recruited in a counterbalanced within-subject design. The study was approved by Ethics Committee of Universidade Lusófona.

Experimental Task

Participants complete two conditions, a face-to-face H–H interaction and an immersive H–VH interaction delivered in VR. Each round includes a free verbal interaction phase followed by a decision phase (cooperate vs compete) and outcome feedback, maintaining perceived interactional contingency while enabling analysis of both decision and feedback-related processes. Dialogue was unscripted in both conditions, but conversational variability was bounded through fixed timing, equivalent task structure, binary decision options, identical feedback, and no persistent agent memory across participants.

Measures

Psychophysiological measures include electrodermal activity (EDA), electrocardiography (ECG), and respiration. Behavioral measures include cooperation rate, decision latency, and eyetracking. Self-reported measures include a sociodemographic questionnaire, a personality inventory, and questionnaires assessing social presence, affective state, perceived agency, and agent characteristics.

Data Analysis

Trial-level data will comprise 900 observations, derived from 30 participants completing 15 rounds in each of two conditions. Data will be analyzed using multilevel mixed-effects models accounting for repeated trials nested within participants and interaction pairs. Cooperation will be modeled using logistic mixed-effects regression, and baseline-corrected physiological responses will be analyzed as a function of condition, phase, outcome, and their interactions. Post-condition ratings of social presence, affect, and perceived agency, together with pre-task personality traits, will be incorporated as moderators or mediators of behavioral and psychophysiological effects.

Preliminary Results

The main analyses will contemplate findings from the full sample while pilot observations (n = 4) indicate comparable temporal trajectories of physiological responses across interaction conditions. Descriptively, H–H interactions appear associated with greater autonomic response magnitude, particularly in EDA.

Discussion

These observations suggest that differences between conditions may reflect changes in response intensity rather than qualitatively different psychophysiological dynamics. The presence of a real partner may therefore amplify autonomic engagement during social decision-making.

Conclusion

By integrating VR, LLM-driven dialogue, and psychophysiological recording within a strategic interaction paradigm, this study provides a multimodal framework for examining how humans respond to artificial social agents. Beyond behavioral and physiological measures, the collection of full dialogue transcripts enables future analyses of linguistic and affective dynamics in human–AI interaction.

ABSTRACT. Introduction Traditional psychometric assessments primarily rely on self report questionnaires to evaluate personality traits and emotional states. While widely validated, these methods are susceptible to response biases and often fail to capture dynamic, real time physiological changes. Internet of Things (IoT) technologies, specifically non-intrusive wearable sensors and AI based behavioral analysis, have made it possible to continuously record physiological and expressive signals. Based on these developments, this study proposes a multimodal framework that integrates biometric, behavioral, and psychometric data. By utilizing specific IoT hardware, such as the Polar H10+ and Moofit HW401 heart rate sensors along with cameras and the Pupil Labs Core eyetracking device, this approach aims to connect subjective reports with objective physiological markers. This methodology emphasizes ecological validity, bridging the gap between controlled laboratory settings and real world monitoring.

Methods A sample of 15 participants (aged 18–35) will complete three laboratory sessions, providing informed consent in accordance with university ethics and GDPR compliance. Personality will be assessed using the HEXACO-60 inventory, while anxiety and depression symptoms will be measured via the Hospital Anxiety and Depression Scale (HADS). To elicit emotional reactivity, participants will view standardized images from the International Affective Picture System (IAPS) and rate their affect along valence and arousal dimensions using Russell’s circumplex model. To prioritize ecological validity while maintaining experimental control, we will use ambulatory IoT sensors to record heart rate variability (HRV) and eye-tracking metrics. AI-based processing tools will be employed to extract facial action units, gaze metrics, and autonomic indices, ensuring all data streams are perfectly synchronized with the stimulus presented.

Data Analysis Physiological signals will undergo preprocessing for artifact correction and baseline normalization. Descriptive statistics and correlation analyses will be used to explore the relationships between these multimodal physiological features and the psychometric scores from the HEXACO 60 and HADS. The analysis will focus on how AI extracted behavioral markers such as facial micro expressions and gaze fixations correspond to the valence and arousal ratings provided by participants. By integrating these distinct data streams, the study will examine the consistency between measurable physiological responses and subjective psychological profiles.

Expected Results Specific personality traits and emotional symptoms are expected to correspond to distinct multimodal patterns. For instance, higher emotionality (HEXACO-60) and anxiety (HADS) are anticipated to correlate with increased autonomic arousal and stronger expressive responses during the task. Conversely, higher extraversion is expected to predict greater positive facial activation, while depressive symptoms may be associate with reduced physiological responsiveness and lower HRV. These results are expected to demonstrate that the integration of IoT based biometrics and AI driven behavioral data provides a more comprehensive view of an individual's psychological state than self reports alone.

Conclusion By combining IoT wearable sensors, AI driven behavioral analysis, and validated psychometric instruments, this study establishes a robust framework for assessing personality and emotion through objective physiological patterns. The use of AI is central to this framework for the precise extraction and synchronization of behavioral and autonomic data. Furthermore, by utilizing non-intrusive devices like the Polar H10+ and Pupil Labs sensors, this research helps transition psychophysiological assessment out of traditional lab setups, establishing a foundation for ecologically valid, real-world mental health monitoring.

ABSTRACT. Introduction

Ecological Momentary Assessment (EMA) is used to examine real-time fluctuations of nicotine cravings in naturalistic settings. Conventional EMA protocols rely on random or fixed prompting schedules, sampling experiences as they occur. While this approach enables ecological observation, it may create a trade-off between a participant's burden and the capturing of relevant moments. Given that nicotine craving fluctuates dynamically and is influenced by physiological and contextual states, random sampling strategies may not optimally align with periods of heightened vulnerability, such as acute stress episodes, prolonged periods without smoking or certain social contexts.

Emphasis has been made in optimizing the timing of intervention delivery; less empirical work has focused on systematically optimizing assessment timing itself. Rather than randomly observing craving states, the present study introduces an individualized approach that triggers EMA prompts during moments predicted to reflect elevated craving probability, based on physiological and contextual measures. This protocol evaluates whether personalized, data-based sampling can improve high-craving detection while reducing assessment burden.

Objectives

To design and evaluate data-driven trigger rules for EMA, by developing individualized decision-tree models that use wearable-derived physiological signals and self-reported contextual data, collected via smartwatch (Garmin) and the m-Path Sense app, to predict elevated craving states in daily smokers over a two-week period, with the aim of comparing random sampling with personalized, data-informed EMA triggering in terms of craving detection efficiency.

Methods

We aim to recruit N = 100 daily smokers (18+) for a two-week study. Participants are recruited through social media advertisements, university research portals, and flyers. Eligible participants must report daily use of nicotine products for at least six months and be willing to wear a Garmin smartwatch during waking hours. Exclusion criteria include current smoking cessation treatment, cardiovascular conditions affecting physiological measurements, and severe psychiatric disorders that could impair compliance. Ethical approval for the study was granted by the Social and Societal Ethics Committee of KU Leuven.

Phase 1 (Observational Modeling):

During Week 1, participants receive 12 randomly timed EMAs/day in the m-Path Sense app, assessing craving intensity, perceived stress, and time since last smoke, which is temporally linked to Garmin smartwatch data. Individualized decision-tree models (max-depth = 2) are trained to predict elevated craving states, defined using a within-person median split. Models include physiological (HR, HRV, Stress) and contextual (Time-since-last-smoke, Time-of-day) predictors, and they are evaluated using forward-chaining cross-validation and considered valid only if they contain at least one split (depth ≥ 1) and predict both outcome classes.

Phase 2 (Experimental Adaptive Trigger):

During Week 2, a hybrid EMA protocol is implemented. Each day participants receive 3 scheduled prompts (control condition) and up to 3 data-informed prompts triggered by participant-specific decision rules, derived from Phase 1 models. This design enables a within-person comparison of detection efficiency between scheduled and adaptive triggering while reducing burden. For participants without a valid model, only scheduled prompts are delivered. Analyses will focus on within-person comparisons of craving detection across trigger types, evaluating whether data-informed prompts are more likely to capture elevated craving states than scheduled prompts.

Expected Outcomes

Pilot testing (N=10) suggests that both physiological and contextual variables contribute to predicting elevated craving states. Ongoing pilot work has informed refinements to the triggering protocol and highlighted practical challenges related to data availability, model stability, and participant compliance.

We hypothesize that personalized triggering will capture a greater proportion of high-craving states per prompt compared to random sampling, using fewer EMA-beeps than phase 1.

Shifting from random to data-driven sampling may enhance EMA’s contextual precision while reducing participant burden, informing the development of more efficient assessment systems and future Just-in-Time Adaptive Interventions for nicotine dependence.

ABSTRACT. Background Falls are a significant cause of morbidity and reduced quality of life in individuals with Parkinson's Disease (PD), typically resulting from complex interactions between motor and cognitive impairments. Although both gait impairment and cognitive decline have been individually linked to increased fall risk, their combined predictive value has not been extensively studied. Integrating motor and cognitive measures may improve fall-risk prediction and enable more tailored clinical interventions. This paper investigates the extent to which combined cognitive and gait assessments can enhance fall-risk classification in individuals with PD and examines whether machine-learning techniques improve prediction accuracy compared with traditional statistical methods.

Methodology A retrospective dataset from CNS - Campus Neurológico (Portugal) was used, comprising patients with PD who underwent routine cognitive and motor assessments as part of their clinical care. The initial dataset comprised approximately 250 individuals aged 55 to 90 years. Following data preprocessing and quality control, including the selection of participants with complete cognitive and gait assessment data required for predictive modeling, a preliminary sample of 26 participants is under analysis. Full dataset access and expansion are pending final approval from the CNS Ethics Committee. The anonymised dataset was retrieved by authorised CNS staff and is stored in a secure location on encrypted local research systems to comply with ethical and data protection requirements. Cognitive function is assessed with the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA), whereas gait and balance are measured with the Mini-BESTest. To evaluate fall risk prediction, three categories of input models will be developed: cognitive-only, gait-only, and combined cognitive and gait. Logistic regression and the supervised machine learning technique, Support Vector Machine (SVM), will be employed to develop and compare predictive models for fall risk classification. The study is guided by two main hypotheses. First (H1), the combined cognitive and gait model is expected to demonstrate superior fall-risk prediction performance compared to cognitive-only and gait-only models. Second (H2), the machine learning algorithm (SVM) is expected to outperform logistic regression across all input model types. Model performance will be evaluated using appropriate classification metrics, including balanced accuracy, sensitivity, specificity, and area under the curve (AUC), enabling a comparative analysis between input configurations and modelling approaches.

Expected results Previous studies suggest that decreased cognitive function is associated with poorer gait performance and increased fall risk. Predictive models that combine cognitive and gait metrics are expected to achieve higher classification accuracy than models relying solely on motor variables. Machine learning methods, which can capture complex, non-linear interactions between cognitive and motor deficits, are expected to outperform conventional statistical approaches. Additional validation will be conducted as further data become available following ethical approval.

Discussion This research aims to improve the accuracy of fall risk prediction in PD by leveraging multimodal clinical assessments and advanced predictive modeling methods. Greater predictive accuracy may enable earlier identification of high-risk individuals, supporting targeted intervention strategies and ultimately enhancing patient safety and clinical outcomes.

ABSTRACT. Introduction: Smartphone use’s ubiquity is accompanied by the demand for self-regulation tools, yet such tools operate within ecosystems optimized for engagement. Existing approaches range from “digital detox” strategies to targeted digital interventions. However, they face criticism for being overly intrusive, insufficiently personalized, and temporally static. Many self-regulation tools rely on rigid daily limits or require continuous user input, creating friction that undermines sustained engagement and ecological validity. Digital nudges offer a promising alternative. By subtly modifying user-interface elements, nudges steer behavior without restricting choice. Applications such as the “GoodVibes” employ a gentle haptic feedback nudge when users exceed pre-defined time limits. Personalization can be achieved through self-nudges, where individuals voluntarily select regulatory strategies aligned with their goals. Nevertheless, previous studies suggest static nudges may fail due to low engagement and the ease with which they can be bypassed. Interventions that rely on fixed thresholds or daily aggregates may not align with the dynamic nature of smartphone use. To address these limitations, Just-In-Time Adaptive Interventions (JITAIs) aim to deliver support precisely when risk is elevated. Although current JITAIs improve timing sensitivity, they require continuous human feedback to adapt effectively, burdening users. We introduce SmartPause, a framework that adapts the timing of self-nudges using per-session logic. SmartPause employs vibrations as a form of negative reinforcement, leading to learning through escape conditioning: users can terminate the aversive stimulus by closing the applications they wish to use less. Technically, SmartPause formulates the intervention problem as episodic Q-learning, a type of Machine Learning (ML), using contextual variables to adapt vibration timing across days. The employed contextual variables include session duration, time of day, number of recent vibrations, and application type. The ML model action space consists of selecting the next vibration interval, and the reward function assigns positive value to session termination shortly after a vibration while penalizing excessive vibration frequency. Methodology: We evaluate SmartPause through a four-week longitudinal field experiment using a randomized controlled trial (RCT) design (n ≈ 40 adults using Android smartphones as a primary device, not participating in any other smartphone usage intervention), recruited from a university environment. Initially, participants choose apps to target (i.e. apps they wish to reduce usage). Following a two-days baseline phase of passive logging, they are randomized to one of two conditions: (1) Adaptive SmartPause or (2) Static schedule, in which 20% of their target app sessions will receive vibrations, determined at random. [GK1.1]During active sessions exceeding a minimum duration threshold, decision points occur at which participants may receive or not receive a vibration. This design permits estimation of proximal causal effects and modeling of time-varying treatment responses for the static group. The primary outcome is average daily usage of self-selected target apps, collected automatically. Secondary outcomes include perceived intrusiveness, ease of use and cognitive burden. Behavioral data are collected through an Android background service, and baseline and exit surveys assess subjective outcomes. Analyses employ mixed-effects models to account for repeated measures, and specific estimators to assess proximal treatment effects. Intervention volume (number of vibrations) will be used as a covariate in secondary analysis, to control for their effects. Expected Results: We expect the adaptive condition to produce significantly greater reductions in session duration and daily usage relative to the static condition. Effects should be strongest in certain context variable combinations (e.g. long morning sessions), indicating context-sensitive optimization. Over time, we anticipate decreasing latency from vibration to closure. Subjectively, we expect intrusiveness and cognitive burden not to differ between conditions. These findings would demonstrate that smartphone self-regulation can be modeled as an episodic reinforcement learning problem, and that intra-session adaptive negative reinforcement outperforms static nudges.

ABSTRACT. Introduction Mentalization (mental state attribution; MSA), the attribution of beliefs, thoughts, emotions, and intentions to oneself and others, is a core component of social cognition typically assessed with standardised tasks (e.g., RMET; Frith-Happé Animations). However, these tasks may not fully capture spontaneous mental state attribution in naturalistic contexts. Open-ended narratives elicited by film stimuli offer a more ecologically valid window into spontaneous mentalization, but manual MSA coding is labour-intensive and limits scalability. This study evaluates two supervised Natural Language Processing (NLP) approaches for automating MSA scoring in film narratives. It further examines whether patterns provide predictive information beyond explicitly annotated mental state references (MSR). Accordingly, we compare a span-level model trained on human annotations (AFF/COG/INT) with a paragraph-level model trained to predict reference counts and a pre-coded complexity score. The study tests four hypotheses: The span-level model will detect and classify MSR with agreement comparable to human-human inter-rater reliability (IRR); The paragraph-level model will predict MSR counts and complexity scores without using span annotations, suggesting predictive patterns beyond annotated spans; The paragraph-level model will outperform the span-level model; The paragraph-level model will better predict social-cognition outcomes (RMET; Frith-Happé) compared to span indices. Methods This study is a secondary analysis of open-ended narratives elicited by a social stimulus (e.g., Father and Daughter). The data combine a German model-development dataset and an English external-validation dataset. The German dataset included 495 native German-speaking participants (341 female; age M = 27.74, SD = 11.23, range = 18-80), yielding 956 usable open-ended narratives across unprompted story descriptions (n = 495) and prompted mental-state descriptions (n = 461). The English dataset included 108 valid human-coded MSA responses collected online from adult recruited via Prolific and Qualtrics. Participants were predominantly female (70.6%) and mostly native or fluent English speakers. Narratives were coded using an established MSA framework providing (a) span-level AFF/COG/INT MSR labels and (b) reference counts and a pre-coded complexity score reflecting the depth of mental-state reasoning. Original data collection was approved by the Research Ethics Committee of University College Dublin. This secondary analysis uses anonymised archived data under the same governance. The German data will be used for training and internal testing, and the English data for external validation. The span-level model will be trained on human span annotations to output AFF/COG/INT labels and counts. The paragraph-level model will be trained on full texts to predict reference counts and complexity scores. RMET and Frith-Happé outcomes will be analysed as construct-validation criteria on available cases. Analyses: IRR will be estimated as a benchmark. The span-level model will be evaluated by agreement with human MSR annotations, and the paragraph-level model by prediction accuracy for counts and complexity. Model comparison will test whether paragraph-level prediction improves accuracy relative to span-level prediction. Construct and incremental validity will be examined by relating AI indices to RMET and Frith-Happé using correlations, regression, and relevant covariates. Expected Results Automated MSA scoring is expected to correspond strongly with human coding, enabling scalable measurement. We also expect whole-text prediction to show lower error than span coding, reflecting features beyond explicit mental-state expressions. Where available, AI indices are expected to demonstrate stronger associations with RMET and Frith-Happé. Results should support a scalable psychometric pipeline for studying spontaneous mentalization in naturalistic contexts. Discussion and conclusion This work aims to advance social-cognition measurement by integrating narrative data with NLP models within a psychometric validation framework. Comparing human-coded outcomes with paragraph-level prediction will clarify whether explicit MSR and broader patterns contribute to scalable social-cognition measurement and prediction. Implications include improved feasibility for large-scale studies. Comparing the two models may indicate whether mentalization includes mental-state terms or broader structures.

ABSTRACT. In response to increasing cybersecurity threats, this study presents a virtual reality (VR)-based training simulator designed to raise awareness of cyberattacks and promote safer digital behaviors. The immersive nature of VR allows the creation of high-fidelity and realistic scenarios, and collection of behavioral data, such as eye-tracking and interactions with virtual elements, allowing robust research for studying user interaction with cyber threats. The primary objective of this pilot study was to assess the feasibility, functionality, and usability of the proposed VR system for future testing of effectiveness as a training intervention.

A pilot evaluation was conducted with a convenience sample of ten adult participants across two iterative sessions. The first session focused on technical system testing, including the identification and correction of software bugs and interaction issues. Following system refinement, a second session examined system stability and usability. Participants interacted with a VR cybersecurity scenario developed in the Unity game engine and deployed on the HTC Vive Pro Eye headset with integrated eye‑tracking. Usability and functionality were assessed using participant qualitative and quantitative feedback (e.g. Post-session questionnaires and inquiry) and observational data (eye-tracking and interactions with devices and environmental elements) collected during and after the VR sessions.

Results from the first session informed targeted technical improvements, which were subsequently validated during the second session. Findings indicated stable system performance, intuitive interaction mechanics, and generally positive user experiences, supporting the simulator’s feasibility and suitability as a cybersecurity awareness training tool. Eye‑tracking capabilities further demonstrated the system’s potential to capture objective behavioral data for future analysis.

While limited by a small sample size, this pilot study provides preliminary evidence supporting the robustness of the prototype’s core design and its suitability for further development. Future research with larger and more diverse samples is needed to evaluate learning outcomes and training effectiveness. Nevertheless, the study highlights the promise of VR‑based approaches for advancing cybersecurity awareness and exploring human behavior in digital risk contexts.

ABSTRACT. Bullying rarely occurs without an audience and can be considered a group phenomenon, rather than just a personal behavior. Bystanders are present in most bullying and cyberbullying incidents, and when they intervene in favor of the victim, they can effectively stop it. Evidence suggests that intergroup factors, such as social identification, increase bystanders’ helping intentions in bullying episodes. However, relatively little is known about the potential positive effects of intergroup factors on bystanders’ attitudes and behaviors when witnessing bias-based cyberbullying (i.e., cyberbullying based on identity). Two studies conducted with Portuguese youth examined bystanders’ responses to cyberbullying toward two minority groups (i.e., LGBTQI+ and Black youth), and what can influence their helping intentions when they witness bias-based cyberbullying episodes. Data were collected in a single large-scale survey with Portuguese youth and emerging adults (N = 4,507; aged 15–32). To address distinct but complementary research questions, the full sample was randomly divided into two independent, non-overlapping subsamples using standard randomization procedures. Study 1 examined differences in bystander responses as a function of the target group, whereas Study 2 tested the intergroup factors that may influence bystanders’ helping intentions. Study 1 (N = 2,253) showed that bystanders' responses vary depending on the target of cyberbullying, helping an LGBTQI+ youth target less (M = 3.33, SD = 1.61) than a Black target (M = 3.93, SD = 1.51), t = 8.86, p < .001, d = .39, and showing less empathy (d = .47), less positive group norms (d = .82), less inclusive identities (d = .60), less positive attitudes (d = .58), and more intergroup anxiety (d = −.52). Study 2 (N = 2,254) revealed that, for the LGBTQI+ target, high-quality offline contact is associated with more helping behaviors via increased empathy (b = .13, 95% CI [.09, .18]), outgroup attitudes (b = .06, 95% CI [.02, .11]), dual-identity representations (b = .01, 95% CI [.00, .01]) and decreased intergroup anxiety (b = −.06, 95% CI [−.10, −.02]). For the Black target, contact was associated with helping intentions via empathy (b = .07, 95% CI [.04, .10]), one-group identity (b = .01, 95% CI [.00, .02]), and group norms (b = .02, 95% CI [.00, .04]). Together, these findings highlight both the variability of bystanders' responses depending on the specific target of cyberbullying and the potential of high-quality contact to foster bystanders’ helping intentions in cyberbullying incidents. These results contribute to our understanding of group dynamics in cyberbullying and the role of intergroup contact in influencing bystanders' intentions to help in digital spaces.

ABSTRACT. Cyberbullying is considered one of the most prominent psychological and social problems facing university students in the digital age, due to its negative effects on victims’ mental, social, and academic well-being. The impact of cyberbullying is not limited to the bully and the victim alone; it also extends to bystanders who witness the incident without being directly involved. The behavior of these bystanders can either contribute to reducing cyberbullying or facilitate its continuation. Research indicates that bystander intervention is one of the most effective strategies for reducing cyberbullying and mitigating its harmful consequences. Despite this growing international research interest, the Arab context still suffers from a clear shortage of standardized psychometric instruments that assess electronic helping behavior among bystanders, particularly at the university level. The current study aimed to develop an electronic helping behavior scale for bystander students at King Saud University in the context of cyberbullying and to examine its psychometric properties in terms of validity and reliability. The study seeks to provide a scientifically grounded Arabic instrument that can be reliably used for research and applied purposes. The study employed a descriptive research design, which is appropriate for the nature and objectives of the research. The study sample consisted of 126 male and female students from King Saud University. The researcher developed the Electronic Helping Behavior Scale based on relevant theoretical literature and previous studies addressing bystander behavior and cyberbullying. The scale comprised short hypothetical scenarios reflecting different cyberbullying contexts, in addition to self-report items. It included four main dimensions: interpretation of the situation as dangerous, degree of perceived responsibility, belief in self-efficacy, and making the final decision to provide help. Validity was examined through internal consistency validity and construct validity using exploratory factor analysis. Reliability was assessed using Cronbach’s alpha coefficient, McDonald’s omega coefficient, and split-half reliability. The results indicated that the Electronic Helping Behavior Scale demonstrated high levels of validity and reliability. Internal consistency and construct validity coefficients confirmed the homogeneity of the items and their suitability for measuring the intended dimensions. Reliability coefficients were high across all dimensions of the scale as well as for the total score, supporting its appropriateness for use in university settings. Cronbach’s alpha values ranged from .696 to .949 across the subscales, and the overall scale demonstrated excellent reliability, with an alpha coefficient of .949. The findings are interpreted in light of helping behavior theory and social responsibility theory, which suggest that perceiving the seriousness of a situation and experiencing a sense of moral responsibility increase the likelihood that a bystander will intervene to assist the victim. Self-efficacy theory also supports the results, as individuals’ beliefs in their ability to intervene effectively play a crucial role in the decision to provide help. Additionally, bystander decision-making models explain how individuals move from interpreting a situation to engaging in positive intervention behaviors. The study recommends conducting further research to standardize the scale using larger and more diverse samples across different educational stages and age groups. It also suggests examining test–retest reliability over time, in line with recommendations in the psychometric measurement and cyberbullying literature.

ABSTRACT. Background and Objective. Chronic pain (CP), defined as a pain persisting for at least three months, has a significant impact on children’s autonomy and well-being (Anghelescu & Tesney, 2019; Tutelman et al., 2021). Pharmacological treatments are first-line approaches (Poddighe et al., 2019) but carry risks and adverse effects, highlighting the need to advance nonpharmacological strategies (Hudgins et al., 2019; Mboma et al., 2021). Virtual reality (VR) has emerged as a novel nonpharmacological tool for pain management and prevention (Goudman et al., 2022; Malloy & Milling, 2010; Singleton et al., 2024; Tas et al., 2022). Despite growing literature on its efficacy, key design elements for effective VR interventions for pediatric CP remain unclear (Ahmadpour et al., 2020). Therefore, this scoping review aimed to explore VR design strategies for managing and/or preventing pediatric CP.

Methods. Following the JBI methodology and PRISMA-ScR guidelines (Tricco et al., 2018), a comprehensive search was conducted across five major databases: PsycINFO/OVID, Medline/OVID, Central/OVID, Scopus/Elsevier, and Embase/Elsevier, complemented by Google Scholar. The PCC framework was applied to define the eligibility criteria. This scoping review focused on children and adolescents under 18 years of age with primary or secondary CP, while also considering studies that included both pediatric and adult populations. It examined studies investigating the use of VR to manage and/or prevent CP conditions, including those describing virtual environments (VEs) developed for this purpose. All types of VEs were eligible, including computer-generated environments as well as experiences based on real-world footage (e.g., 360° videos). All contexts were considered. Eligible studies were systematically screened (by three reviewers) and analyzed. The protocol was registered on 15 November 2024 in the Open Science Framework (https://osf.io/qv7hw/), with an amendment added on 27 August 2025.

Results. Of the 2,451 records identified, twenty studies met the inclusion criteria. Most studies focused on CP management (85%) and were conducted in clinical settings (75%). The interventions addressed a broad range of CP conditions. Several intervention objectives were identified, ranging from simple distraction without additional therapeutic goals to more structured goals, such as the development of relaxation skills or facilitation of transfer of pain management strategies into daily life. Between these two extremes, VR was frequently used to provide distraction while simultaneously targeting a specific therapeutic objective, such as inducing relaxation or promoting movement. VR was also applied in the context of mirror therapy. To support these objectives, a wide variety of VEs were implemented, some of which incorporated biofeedback. These VEs were categorized into four main types: relaxing environments, gamified movement-based environments, simulations of everyday settings, and avatar-based interactive environments. Only 40% of these VEs were specifically designed for children. Based on these findings, we propose a classification of VR designs and their applications, adapting the holistic framework of Ahmadpour et al. (2020) to pediatric CP. This model integrates VR design components addressing (i) design-related factors, including experiential (e.g., storytelling, emotions elicited during immersion) and product aspects (e.g., hardware and user experience), and (ii) intervention-related factors, mapped onto three continuums: distraction to skill building, passive to active patient role, and no feedback to interactive feedback. Moderators of intervention efficacy, intervention objectives, and implementation barriers were also incorporated.

Conclusion. Although numerous pilot and feasibility studies have been conducted, future research should also evaluate the clinical efficacy of VR designs, for example through RCTs. Studies should also clearly describe the VR designs used and systematically situate intervention objectives and strategies within the proposed classification, thereby facilitating comparisons across interventions and strengthening the applicability of the adapted model.

ABSTRACT. Background: In recent years, the use of immersive Virtual Reality (VR) has seen a substantial increase among children, particularly within pediatric care settings. As its use expands, it is critical to identify the factors that determine its effectiveness. A key determinant of VR effectiveness is the sense of presence (the subjective experience of being there). While goal-directed tasks have been proposed to enhance presence, this relationship has not been systematically examined in children, particularly in clinical contexts. Objective: This follow-up analysis to Felnhofer et al. (2025) investigated whether a goal-directed task embedded in a VR distraction enhances presence in children undergoing minor surgery for one or more ingrown toenails under local anesthesia. User experience outcomes (enjoyment, immersion, comfort) were assessed alongside the sense of presence to evaluate feasibility and acceptability of pediatric VR interventions. Methods: Forty patients aged 10–17 years (M = 13.70, SD = 1.79; 18% female) were randomly assigned to either (1) a focused VR condition including a goal-directed task or (2) an exploratory VR condition without a goal-directed task. During pediatric surgical treatment, participants sitting in a semi-upright position received 15–45 minutes of VR-based distraction. They navigated a virtual submarine in a custom-developed immersive underwater VR environment presented via a head-mounted display (Oculus Quest). In the focused condition, children completed a coin-collection task supported by auditory cues and a scoreboard, requiring sustained and goal-directed attention; the exploratory condition allowed free navigation without task demands. Sense of presence was assessed using the Igroup Presence Questionnaire (IPQ: Spatial Presence, Involvement, Experienced Realism). User experience was measured with the Technology Usage Inventory (TUI). Group differences were analyzed using Bayesian independent-samples t-tests. The Bayes Factor (BF₊₀) quantifies the relative evidence for the alternative hypothesis (BF₊₀ > 1) versus the null hypothesis (BF₊₀ < 1), with BF₊₀ values of 1–3 indicating anecdotal, 3–10 moderate, and >10 strong evidence (Jeffreys, 1961). Results: Analyses provided moderate evidence that focused VR with a goal-directed task elicited higher Spatial Presence (BF₊₀ = 3.12, error < 0.001) and Involvement (BF₊₀ = 3.41, error < 0.001) relative to exploratory VR. Evidence for differences in Experienced Realism was anecdotal (BF₊₀ = 1.07, error = .004). No group differences were found in user experience outcomes, including Immersion (BF₊₀ = 0.48), Enjoyment (BF₊₀ = 0.74), or Comfort (BF₊₀ = 0.31). Technology-related adverse effects (e.g., vertigo, nausea, fatigue, or vision problems) were not observed. Conclusions: Findings indicate that goal-directed immersive VR elicited greater presence in pediatric patients during minor surgery relative to a no-task condition. With interactivity and motor engagement held constant across conditions, the present study suggests that goal-directed engagement represents a distinct contributor to the sense of presence, consistent with theories linking presence to attentional and cognitive processes. Accordingly, the results highlight the potential to optimize the experience of presence through targeted design features. The absence of group differences in comfort, enjoyment, and immersion aligns with prior evidence for the general feasibility and acceptability of immersive VR in children, while challenging assumptions from gamification frameworks that link goal-direction to increased enjoyment.

ABSTRACT. Context and Problematic Cognitive dissonance occurs when individuals encounter information or situations that conflict with their beliefs or habits, often motivating them to modify their behavior. Virtual Reality (VR) provides an immersive and controlled environment that can induce such dissonance and enable the examination of its effects on health‑related behaviours. Prior Work and Rationale A first study (Etienne, Farra & Schyns, 2023) tested several VR scenarios designed to increase stair use among individuals with overweight, showing that dissonance induction is effective when the scenarios activate self‑control norms. These results support the relevance of using VR‑based dissonance to target other health behaviors. Aim of the Present Study Building on these findings, the current study investigates whether a VR immersion designed to induce cognitive dissonance can influence food choices. Specifically, we examine whether exposure to dissonance shifts participants’ preferences toward healthier options. A total of 34 participants took part in the study, with a mean age of 33.9 years (±12.6) and a mean BMI of 27.1 (±1.42). Method The procedure consisted of a simple food choice task in which participants selected between a healthy tray and a sugary tray. Cognitive dissonance was induced twice through two distinct persuasive statements encouraging the sugary option. Participants made three successive decisions: an initial spontaneous choice, a second choice after the first dissonance induction, and a third choice following the second induction. This design allowed us to track how food choices changed across the two dissonance exposures. Results Analysis of the 34 participants revealed an initially balanced distribution between healthy and unhealthy choices (17 vs. 17). After the first dissonance induction, 8 participants revised their decision, with 7 shifting toward the healthy option. The second induction prompted 9 additional changes, including 7 further shifts to the healthy tray. Overall, participants who initially chose an unhealthy option were the most likely to change their selection, and the proportion of healthy choices increased at each decision point. These findings indicate that VR‑induced cognitive dissonance consistently guides food choices toward healthier alternatives. The content analysis of the VR sessions reveals several attempts at rationalization, illustrated by remarks such as: “If there had been strawberries in tray 1, I would have taken it without hesitation…”. Such comments suggest that participants justify choices that deviate from the health‑promoting option, thereby reducing the discomfort generated by dissonance. Rather than selecting the healthier tray, they seek reasons that preserve their original preference without internal conflict. These observations indicate that dissonance resolution does not always manifest through overt behavioral change; it can also appear through defensive or compensatory verbal justification. Conclusion The study demonstrates that VR-induced cognitive dissonance can meaningfully influence food choices. Across the three decision points, participants increasingly selected the healthier option, with the strongest shifts observed among those who initially chose the sugary tray. Beyond these behavioral changes, several instances of rationalization emerged, indicating that some participants attempted to reduce dissonance through verbal justification rather than action. Together, these results suggest that VR-based dissonance induction can modify immediate food choice decisions, either by prompting healthier selections or by revealing compensatory cognitive strategies when behavior does not change.

ABSTRACT. The anxiety associated with job interviews can be highly distressing, hindering a candidate's performance and preventing people obtaining optimal jobs. Several factors can contribute to the stress experienced by candidates during job interviews, including actually having the competence for the job, feeling having little control over interviewers' questions and decisions, lack of experience with job interviews, doubting one’s own abilities, low perceived self-efficacy to cope with stress during job interviews and social fears. Increasing candidates’ perceived self-efficacy during a job interview will lead to less stress, more confidence in their ability to succeed in interviews, higher motivation to apply to job interviews and feeling better prepared. Only a few studies have been published on computer simulated job interviews. Most studies were conducted with adults with mental illness, and almost none with immersive virtual reality (VR) tools. There is a lot of evidence that immersions in VR provide advantages over in vivo situations to practice skills involving social interactions.

The advent of Large Language Models (LLMs) enables the creation of conversational agents with fluid, natural and convincing speech. Research to embrace this new type of technology in psychological practice, either for clinical and/or therapeutic settings, is growing. Furthermore, LLMs can also be combined with VR-based technology to enhance interaction and immersion by giving virtual characters more complex personality traits, with applications for stress relief practice for scenarios such as public speaking and interpersonal conflicts. Using LLMs also creates some challenges, including drift over time from the instructions provided in the initial prompt. Such systems are also developed by combining multiple technical layers: LLM conversational agents, links to external APIs, VR engines and applications (e.g. using Unity Engine) and tools for data collection. The vast majority of AI-powered VR tools are developed by technical teams under the « bona fide model » of good intentions, not to harm the limited financial resources by systematically exploring all lines of programming to identify how secure the code is. This leaves the majority of VR applications dedicated to mental health not being rigorously tested from cybersecurity threats, without an unified approach to test these applications.

The objective of our proposed research project is twofold: (a) test how an AI-powered VR environment can help users immersed in a job interview become less anxious, less avoidant, and feel more self-efficacious and (b) map security and ethical risks from the use of the proposed bona fide technology. Based on that analysis, we will propose security controls adapted to this specific context of research to guarantee AI agent interaction integrity, data confidentiality and the security of the code. These findings can then guide other research teams of corporations developing VR tools to better identify and solve cybersecurity weaknesses in the code. In line with the open science movement promoting transparency, reproducibility, and accountability, this poster will present the research protocol of the proposed study.

ABSTRACT. Introduction As generative artificial intelligence (GenAI) systems increasingly function as emotionally responsive and continuously available interaction partners, users may experience what we term AI attachment illusion: the perception of a reciprocal emotional relationship with a non-sentient system. While such interactions may temporarily reduce loneliness or social anxiety, they also raise concerns about social displacement, whereby reliance on friction-free AI interactions may weaken motivation for complex real-world social engagement. Emerging evidence suggests a cyclical relationship in which social anxiety predicts increased AI reliance, which in turn reinforces social withdrawal. However, existing research remains largely descriptive and lacks theoretically grounded frameworks for psychological intervention.

Theoretical Framework This study proposes the application of Metacognitive Reflection and Insight Therapy (MERIT) to the domain of human–AI interaction. MERIT, originally developed to enhance self-reflective capacity and metacognitive awareness, is adapted here as a psychological lens for differentiating emotional validation from reflective insight in AI-mediated interactions. We introduce EmotionLens as a conceptual intervention model that reframes AI from an agent of emotional substitution into a scaffold for metacognitive growth.

Proposed Design The proposed study outlines a comparative framework examining two user contexts: (1) Emotion-oriented generative AI platforms characterized by anthropomorphic design features and affective language, which may amplify attachment-related processes; and (2) Task-oriented generative AI systems that emphasize instrumental utility and serve as a comparison condition for attachment intensity.

Within this framework, we contrast standard validating feedback, which mirrors users’ emotions and may reinforce dependency, with MERIT-informed reflective feedback, which prompts users to examine emotional patterns, contextual triggers, and self–other boundaries through structured metacognitive questioning.

Implications By integrating cyberpsychology and metacognitive theory, this work advances a novel conceptual approach to understanding and addressing AI attachment illusion. The proposed framework highlights how reflective, non-anthropomorphizing AI interactions may support psychological autonomy rather than emotional substitution. These findings have implications for the ethical design of generative AI systems and for future research on technology-mediated self-reflection, particularly among users vulnerable to social anxiety and emotional reliance on digital agents.

ABSTRACT. Background: Tobacco use continues to be a primary driver of preventable global mortality. While hundreds of commercial smoking cessation apps are available, their impact remains limited by low user engagement, and they lack sufficient implementation of theoretically grounded and evidence-based behavior change features. Current research often evaluates these tools by analyzing the app design using Behavior Change Technique Taxonomies, identifying which behavioral strategies are present in the software's code or interface. However, a critical gap exists between the intended intervention (what designers build) and the received intervention (what users actually engage with and remember). If a technique is present in an app but ignored by the user or if it hinders usability due to high friction or intrusiveness, its therapeutic value will likely diminish. This study shifts the focus from an app-centered to a user-centered perspective, investigating which BCTs remain salient to current smokers who have previously engaged with digital cessation tools. Methods: This study employed a mixed-methods analysis of secondary data from a baseline assessment of current smokers in a prospective smoking cessation trial (QuitChoice) of digital interventions (N = 311). Within this sample, a subgroup of 33 participants (mean age = 43; 70% female) reported previous use of smoking cessation apps and provided detailed free-text descriptions of their experiences. These descriptions were analyzed to identify what they liked, disliked, and whether they perceived the tools as helpful. We applied a coding procedure using the BCTTv1 framework, mapping only BCTs that were explicitly described by the users. Results: Of the 33 respondents, 23 (70%) provided specific enough descriptions to be mapped to at least one BCT. The results revealed a narrow range of recalled techniques compared to the hundreds of BCTs theoretically available in these apps. Feedback on Behavior and Self-Monitoring of Behavior were the most recalled categories, but with divergent user perceptions. Feedback on Behavior, such as tracking money saved or days smoke-free, was mentioned by 12 users and generally viewed as positive. In contrast, Self-Monitoring of Behavior, manually logging cigarettes or cravings, was mentioned by 7 users and was frequently associated with high friction, guilt, and abandonment. Prompts/Cues, such as notifications, were mentioned by 3 respondents. All three descriptions were negative, characterizing the prompts as "annoying" or "impersonal". Crucially, cognitive BCTs identified in clinical literature as highly effective, such as Problem Solving and Action Planning, were rarely mentioned. Furthermore, Social Support was absent in the recalls, and users explicitly mentioned it would be desirable to have it, while 18% of users criticized the "impersonal" and "robotic" nature of the apps. Discussion: The findings suggest that the most memorable features of current apps (manual logging and static prompts) are often the ones users find most burdensome. Meanwhile, the features that could drive cessation (problem-solving and social support) are failing to make a lasting impression. To increase effectiveness, mHealth design should evolve in three ways: (a) Reduce Friction, transitioning from manual behavioral logging to automated or passive sensing, when feasible, to minimize user burden; (b) Implement Adaptive Delivery, replacing static, rule-based notifications with Just-in-Time Adaptive Interventions (JITAIs) that deliver support only when most effective and (c) Better handling relapses, using BCTs that reduce the risk of withdrawing from interventions or uninstalling the app during a relapse. By aligning app architecture with user recall and psychological receptivity, developers can better bridge the gap between digital design and meaningful behavior change.

ABSTRACT. Nightlife participation represents a key component of social life in contemporary urban culture, particularly among young adults. However, nightclubs, bars, private gatherings, and other nightlife social settings are also strongly associated with alcohol misuse, which increases the risk of various adverse outcomes, such as violence, injuries, risky sexual behavior, and driving under the influence. Alcohol craving is defined as a strong desire or a compulsive urge to consume alcohol. It is considered a central motivational process underlying alcohol use. According to the cue-reactivity paradigm, exposure to alcohol-related cues, such as the sight, smell, or context of alcohol use, can evoke conditional response, which elicits craving and facilitates alcohol use. Therefore, within this framework, craving is conceptualized as a conditional response which is triggered by pairing of environmental cues and alcohol use. Cue-reactivity approaches, which involve exposure to alcohol-related stimuli and assessment of craving responses, have been commonly used in research on alcohol use and harm reduction. The research project Nightlife: A study in real and virtual context – REAL NIGHTS, has two main objectives: 1) to assess the extent and patterns of substance use among nightclub attendees, and 2) to examine the applicability of virtual reality (VR) technology in harm-reduction research related to alcohol use. We developed three immersive 360-degree VR videos, depicting nightlife experiences. VR environments were designed within a cue-exposure framework and incorporated visual, auditory, and olfactory alcohol-related stimuli, with the assumption that increased immersion would effectively induce alcohol craving. We utilized 360-degree videos in REAL NIGHTS VR because of their ability to simulate highly realistic, complex social environments while maintaining experimental control and procedure standardization, thereby addressing limitations of limited realism of computer-generated VR environments. VR environments were filmed for three nighttime settings: a nightclub, a bar, and a house party. In total, 18 amateur actors and 70 extras participated in filming of the VR environments. In order to enhance personalization and immersion, environments included personalized scenes. Specifically, musical style and type of alcoholic beverage were varied, allowing participants to tailor the experience according to their personal preferences. This personalization of experience may increase the salience of alcohol-related cues and strengthen craving responses. Participants in the experimental phase will be exposed to VR environments across three sessions, during which alcohol craving and other outcome variables will be repeatedly assessed. This study obtained ethical approval by the Ivo Pilar Institute of Social Sciences Ethics Committee. This submission describes work in progress. The design process, filming procedures, and methodological considerations involved in creating VR environments for the assessment of alcohol craving will be presented. The expected implications of future experimental study will also be outlined, emphasizing 360-degree VR videos as a useful and valid tool for research on cue-induced alcohol craving, their potential role in harm reduction, and methodological limitations associated with this technology.

ABSTRACT. Across different neurodevelopmental phases, brain structures reorganize both structurally and functionally, modifying patterns of neural information flow between regions. During maturation, brain regions and information flows shift from synchronous neural activity to decorrelated patterns. In this context, key regions exhibiting changes in their interactions are the anterior cingulate cortex (ACC), the thalamus (TH), and the striatum (STR). As ACC activity decorrelation is related with the development of inhibitory systems, the direct multi-nodal information flow between these three regions remains to be explored.

This paper analyzes in vivo electrophysiological recordings from mice, to study information flow between ACC, TH and STR during early development. Electrophysiological recordings were obtained from 204 male and female mice, including C57BL/6J and transgenic lines. Ages ranged from postnatal day 5 to 12 (P5–P12). Body weight ranged from 2.57 to 7.54 g at the time of recording. All experimental procedures were approved by the appropriate local ethical committees and complied with German and European regulations.

The variables studied were sex, weight, and information flow, measured with two electrophysiological measurement modalities; Single Unit Activity (SUA) and Local Field Potential (LFP). To quantify directed information flow, Transfer Entropy (TE) will be estimated in both directions for each relevant pair (TH↔ACC and STR↔ACC), exploring multiple delays (τ) to capture plausible biological latencies. Partial Information Decomposition (PID) is used. TE and PID calculations will be performed in MATLAB using the Multivariate Information in Neuroscience Toolbox (MINT).

This project is ongoing as it aims to extend the current findings. New data will be collected from infants with neurotypical development, using a longitudinal design with two acquisition time points, separated by approximately nine months. Structural MRI data will be acquired to investigate the development of information flow and large-scale brain connectivity during early maturation. By comparing measurements across the two time points, we aim to characterize structural reorganization patterns that accompany the developmental shift from predominantly synchronous to increasingly asynchronous brain activity.

The results obtained from the participants will be compared to findings from developmental studies in mice. This approach will allow us to determine whether similar developmental trajectories are observed in terms of brain connectivity.

ABSTRACT. Background and Objective: Compared with their hearing peers, children with hearing impairment exhibit higher levels of stress (e.g., Overgaard et al., 2021; Yigider et al., 2020). Indeed, this population has been shown to exhibit a higher prevalence of mental health disorders, particularly anxiety (Aanondsen et al., 2023). Other studies report that they rely on limited emotion regulation strategies when facing negative emotions. However, interventional research targeting this population remains extremely scarce (Deng et al., 2022). Deep breathing is a recommended technique for reducing stress and anxiety (Hopper et al., 2019), yet most available materials are primarily auditory. Reliance on this single modality is insufficient to ensure access to the acquisition of self-regulation skills in this population. A promising alternative approach involves the use of digital technologies, as they provide structured visual and sensory supports and have strong potential to intrinsically motivate children (Bossenbroek et al., 2020). The practice of deep breathing in Virtual Reality (VR) has shown promising results for the development of emotional skills (Stassart et al., in press). Although the use of VR with children with hearing impairment is still in its early stages (Serafin, Adjorlu, & Percy-Smith, 2023), its multimodal sensory components make it a particularly promising tool in this field. This study aims to investigate the feasibility and satisfaction of a relaxing virtual reality environment adapted for a population of children with hearing impairment.

Methods. Seven children with hearing impairment aged between 8 and 15 years (M age = 12.1, SD = 2.27; 3 boys and 4 girls) were recruited and immersed for ten minutes in a relaxing virtual environment depicting a forest. Within this environment, a fairy communicated using sign language to guide the children through diaphragmatic breathing exercises. To assess the feasibility and applicability of the tool, participants completed questionnaires measuring cybersickness (pre/post; Cybersickness Questionnaire, Cyberpsychology Laboratory at UQO, 2003), sense of presence (Sense of Presence for Children; Vinsous et al., 2025), satisfaction (study-specific questionnaire including three factors: experience, product, and intervention), and state anxiety (pre/post; Facial Affective Scale, McGrath et al., 1996; and STAIC-State, Spielberger et al., 1973).

Results. The results suggest that the virtual environment is feasible and generally well accepted, with low levels of cybersickness (M pre = 0.86, SD = 0.90; M post = 0.86, SD = 1.07; W(6) = 3.00, p = 1.00), a moderate sense of presence (M = 36.2, SD = 17.1, comparable to values reported by Vinsous et al., 2025), and high satisfaction (M = 24.6, SD = 6.19; corresponding to ratios of 0.75 compared to the maximum score of 33). A trend toward reduced anxiety was observed (FAS: M pre = 2.43, SD = 1.81; M post = 1.57, SD = 0.79; STAIC-State: M pre = 28.40, SD = 3.95; M post = 24.90, SD = 3.76), although these changes did not reach statistical significance (FAS: t(6) = 1.03, p = .341; STAIC-State: t(6) = 2.12, p = .079).

Conclusion. This preliminary study suggests that virtual reality may represent a relevant and promising tool for promoting relaxation in deaf and hard-of-hearing children. The immersive experience was well accepted and produced clinically meaningful effects, despite the absence of statistical significance. These findings support further research with larger samples and more tailored environments in order to optimize the applicability and effectiveness of this type of intervention.

ABSTRACT. Background. Virtual reality (VR) is increasingly used in health promotion and clinical research because immersive environments can elicit emotions, cognitions, and behaviors that resemble real-life responses (Malbos & Oppenheimer, 2020; Nolin et al., 2016; Riva, 2022). A key psychological mechanism supporting VR’s effectiveness is the sense of presence, defined as the perceptual illusion of “being there” in the virtual environment while knowing one is physically elsewhere (Slater, 2018; Witmer & Singer, 1998). Contemporary frameworks conceptualize presence as multidimensional, typically including spatial presence/place illusion, plausibility, and social facets such as social presence and co-presence (Poeschl & Doering, 2015; Slater, 2009; Slater et al., 2022; Youngblut, 2003). However, most validated presence measures target adults, and pediatric VR research often uses tools not designed for children, limiting interpretability and comparability across studies (Bareišytė et al., 2024; Schloss et al., 2025).

Objective. This poster focuses on the validation of a French child/adolescent adaptation of the short Presence Self-Assessment Questionnaire (Heck et al., 2021) using confirmatory factor analysis (CFA), examining its four hypothesized dimensions (Spatial Presence, Plausibility, Social Presence, and Co-presence). Method. A sample of 325 children and adolescents aged 8-15 years (M = 10.64, SD = 1.96; 49.2% boys) was recruited from mainstream schools and social media between January 2024 and May 2025. Participants completed a standardized protocol including pre-immersion measures, a VR immersion, and post-immersion measures (≈ 25 minutes). Immersion was delivered via Oculus Quest 2 using a standardized calming “forest riverboat” experience (≈ 7 minutes) featuring guided breathing, ambient music, and a companion animal. Presence was assessed with 16 items rated on a 0-4 Likert scale, corresponding to the theorized subscales (Heck et al., 2021). Because items were ordinal and non-normally distributed, CFAs were estimated using WLSMVS based on polychoric correlations, which is recommended for Likert-type ordinal indicators. Model evaluation relied on robust (“scaled”) χ², CFI, TLI, RMSEA with 90% confidence intervals, and SRMR.

Results (CFA-focused). The theorized four-factor correlated model showed adequate but mixed fit, χ²(46) = 186.76, CFI = .924, TLI = .987, RMSEA = .097, 90% CI [.081, .114], SRMR = .043. Because Social Presence and Co-presence were highly correlated (latent φ = .97), a three-factor alternative combining these dimensions was also tested. This alternative yielded virtually unchanged but slightly less favorable fit, χ²(47) = 190.64, CFI = .922, TLI = .987, RMSEA = .097, 90% CI [.082, .113], SRMR = .045. Standardized loadings were substantial across factors (spatial presence: .77-.84; plausibility: .82-.90; social presence: .64-.82; co-presence: .71-.82), and factor correlations were high (.73-.97). We retained the four-factor solution because it showed slightly better fit than the three-factor alternative and remained more consistent with multidimensional theories of presence. The substantial overlap between Social Presence and Co-presence may first reflect the characteristics of the selected virtual environment, which may not have sufficiently differentiated these two social dimensions. It may also reflect developmental factors, given that theory of mind differentiation is still evolving in childhood and that our sample included proportionally more children than adolescents.

Conclusion. In line with multidimensional models of presence (Poeschl & Doering, 2015; Slater, 2009; Slater et al., 2022; Youngblut, 2003), the CFA findings support the structural validity of this French presence questionnaire for children and adolescents. We recommend a four-factor scoring solution for subsequent analyses and reporting in pediatric VR research, while interpreting the two social dimensions parsimoniously.

ABSTRACT. In team sports, decision-making is a critical determinant of success, relying on athletes’ perceptual and cognitive abilities. Virtual reality (VR) has the potential to simulate real-life sporting scenarios, allowing athletes to practice reading the game, processing information, and making rapid and accurate decisions. Despite its promise, evidence on the effectiveness of VR training for enhancing perceptual-cognitive skills remains limited and often unconclusive. This study investigates the effectiveness of VR-based cognitive training on decision-making accuracy in youth football players. We also explore its far-transfer effects on underlying cognitive functions, including processing speed, cognitive flexibility, and inhibitory control. In addition, we assess participants’ subjective experiences in VR, including embodiment, sense of presence, and acceptance of the technology.

Twenty-eight youth football players from the Under-13 and Under-14 teams of a professional club participate in a within-subjects study during the competitive season. Participants report an average of 7.87 years of football experience (SD = 1.18) and are all enrolled in compulsory education. Assessments are conducted at baseline (T0), after four weeks of control (period in which they only undergo usual football training, T1), and after four weeks of VR-based cognitive training (T2). The VR training is added to the standard football training. Data collection is ongoing and is expected to conclude by the end of May.

The VR-based cognitive training consists of 180° immersive videos presented from a first-person perspective via a head-mounted display. The videos cover variations of seven core football actions: (1) defense and ball control, (2) feinting and dribbling, (3) interception, (4) marking, (5) passing, (6) off-the-ball movement, and (7) shooting and heading. Training takes place three times per week for four weeks. During each session, participants watch 15 videos up to critical decision points, after which they verbalize their intended action and visualize themselves executing the optimal response.

The primary outcome is decision-making accuracy, which is assessed at T0, T1, and T2 with a VR test. A total of 20 unique videos (different from those used for the training) are presented up to the critical decision point, and participants have 1.5 seconds to verbalize their responses. No feedback is provided. A scoring system is used to evaluate responses classified by experts as optimal (2 points), acceptable (1 point), or null (0 points). Secondary outcomes are cognitive functions, which are assessed at T0, T1, and T2 using the Vienna Test System. The Trail Making Test (TMT-L S1/S2/S3) is used to assess processing speed and cognitive flexibility, while the Response Inhibition test (INHIB S13/S14) is used to assess inhibitory control. On the last day of training, subjective experiences in VR are assessed, including embodiment, using the Virtual Embodiment Questionnaire (VEQ) and sense of presence, using the Slater-Usoh-Steed questionnaire (SUS), as well as the perceived usefulness and ease of use of the VR technology, using the Technology Acceptance Model questionnaire (TAM).

We hypothesize that VR-based training will improve decision-making accuracy in football players. We also expect high levels of embodiment and presence, as well as positive attitudes toward the implementation of VR as a complementary training tool in football. Furthermore, cognitive measures will be examined as both outcomes of the VR training and potential mediators of its effectiveness. This will help to determine whether improvements in underlying cognitive functions account for the effects of VR training on decision-making performance. With a rigorous methodology, this study contributes to the literature on the effectiveness of VR-based perceptual-cognitive training in sports.

ABSTRACT. Background: Behavioural and psychological symptoms of dementia (BPSD) and other responsive behaviours are highly prevalent in long-term care and are associated with reduced quality of life, increased caregiver burden, and higher costs. Pharmacological treatments show limited effectiveness and relevant risks, supporting the need for scalable non-pharmacological approaches. Immersive virtual reality (VR) may reduce anxiety and agitation by redirecting attention away from distressing stimuli, temporarily reducing environmental overload, and delivering calming, emotionally meaningful content tailored to individual preferences. However, evidence on repeated, customized in-situ use during real-world episodes of behavioural activation remains limited. This poster reports preliminary quantitative and implementation findings from an ongoing mixed-methods feasibility, proof-of-concept study of customized immersive VR in long-term care. Methods: A longitudinal, within-subject feasibility study was conducted in a long-term care facility in Northern Italy. Residents aged ≥65 years with clinically documented dementia or cognitive impairment were recruited via convenience sampling. The ethics-approved study protocol prioritized feasibility, acceptability, and safety. Safeguards included broad exclusion criteria for potentially higher-risk conditions, a brief familiarization exposure before the main session, routine pre-session clinical checks, continuous staff monitoring, adverse-effect assessment, and staff-observed monitoring of confusion, distress, or delusion-like reactions during and after exposure. Staff received training, technical support, and weekly supervision. Participants received 1–4 immersive VR sessions depending on clinical stability and availability. Sessions were delivered via head-mounted display using customized calming environments (e.g., natural landscapes) selected from preferences gathered from residents and care staff; content could be adapted in real time. Feasibility was examined through recruitment, adherence, attrition, tolerability, and qualitative feedback from a post-intervention focus group with healthcare professionals. Preliminary quantitative outcomes included responsive behaviours, state anxiety, pain, observed emotions, and tolerability. Results: Of 29 eligible residents, 18 initiated VR, indicating moderate uptake, and intervention exposure varied across participants. Preliminary findings suggested short-term reductions in state anxiety within sessions, while agitation scores showed a modest descriptive improvement over the intervention period. Tolerability findings suggested that exposure was generally manageable in selected participants. However, the small convenience sample, variable session exposure, lack of a control condition, and attrition across sessions limit the interpretation of clinical effects. Focus-group findings identified important implementation barriers, including recruitment and consent procedures, organisational and emotional burden, device, sensory, and technical challenges, and limited suitability for acute behavioural escalation. Staff also described practical challenges in delivering VR at the optimal moment, as setup, assessment, and organisational demands could limit use during rapidly escalating episodes. They reported perceived short-term benefits, particularly when realistic and familiar content was used, but noted that acute agitation often required initial relational calming before the headset could be introduced. Conclusions: Tailored immersive VR appears feasible and potentially beneficial for short-term anxiety reduction in selected long-term care residents, but the present findings should be interpreted primarily as feasibility and pilot data rather than evidence of efficacy. Personalized VR appears better suited to mild-to-moderate distress, pre-escalation phases, or planned wellbeing activities than to peak behavioural crises. Beyond supporting the rationale for a larger study, this pilot identifies concrete priorities for future trials, including streamlined recruitment and consent procedures, reduced setup and assessment burden, lighter and less intrusive technology, clearer process indicators of deliverability and interruption, and continued monitoring of confusion, distress, and adverse effects. Larger controlled studies are needed to clarify clinical effects and determine for whom, when, and under what conditions VR is most beneficial. Funding: HubLifeScience–DigitalHealthPNC-E3-2022-23683267DHEAL-COMProject–CUP:C63C22001970001

ABSTRACT. Background: Virtual reality (VR)–based simulations of daily activities are emerging as promising tools for assessing functional cognition in older adults. However, the influence of task complexity on the diagnostic accuracy of these digital assessments remains poorly understood. Objective: This study investigated how task difficulty and individual characteristics—including age, sex, education, and global cognitive status—interact to shape the clinical utility and sensitivity of VR-based cognitive assessments. Methods: Forty-seven community-dwelling older adults (mean age 68.2, SD 5.80) performed two immersive VR kitchen tasks of varying difficulty: coffee-making (lower) and sandwich-making (higher). Performance was quantified through total distance traveled, completion time, and performance errors. Results: Global cognitive function (K-MMSE) was a robust predictor of performance in the lower-difficulty task across all metrics: total distance (B = -1.855, p < .001), completion time (B = -23.377, p < .001), and errors (B = -0.625, p = .001). Conversely, this predictive power was attenuated in the more complex sandwich-making task, likely due to increased sensorimotor demands and floor effects. Notably, demographic variables (age, sex, education, occupation) were not significant predictors in either task. Conclusions: The clinical sensitivity of VR assessments depends on precise difficulty calibration rather than the technology alone. While appropriately scaled tasks capture cognitive variance independent of demographics, excessive complexity may compromise discriminative validity. These findings highlight that optimal difficulty scaling is essential to maximize the precision of VR-based digital functional biomarkers for the early detection of cognitive decline.

ABSTRACT. 360° media-based environments represent a promising approach for cognitive and motor rehabilitation in older adults with cognitive frailties. Unlike computer-generated virtual reality (VR), 360° photos and videos offer authentic real-world representations with lower technical complexity, making them particularly suitable for aging populations who may struggle with highly interactive systems. By combining ecological validity with user-friendly design, 360° media-based applications can bridge the gap between clinical efficacy and practical accessibility. Despite growing evidence supporting VR-based cognitive-motor dual-task (CMDT) interventions, systematic user experience assessment during the development phase remains critical, particularly for applications targeting older adults with mild cognitive impairment (MCI) or subjective memory complaints (SMC). This study evaluated the user experience of DUAL-REHAB, a 360° media-based CMDT training application designed for older adults with cognitive frailties. DUAL-REHAB leverages 360° photos and videos to create realistic virtual environments and was developed in two versions: a fully immersive head-mounted display (HMD) version for supervised hospital-based training and a tablet-based non-immersive version for autonomous home-based training. The evaluation employed a comprehensive dual-assessment approach combining expert heuristic evaluation with end-user testing to ensure both professional validation and authentic user feedback. Eight experts (neuropsychologists, rehabilitation specialists, and human-computer interaction specialists) conducted structured heuristic evaluations based on Nielsen's usability principles adapted for healthcare applications. Subsequently, 20 patients (10 MCI, 10 SMC) participated in user experience testing: all 20 tested the tablet version, while 10 also experienced the HMD version. Quantitative assessments included the ITC Sense of Presence Inventory measuring spatial presence and ecological validity, the Simulator Sickness Questionnaire assessing cybersickness symptoms, and the User Experience Questionnaire evaluating attractiveness, perspicuity, efficiency, dependability, stimulation, and novelty. Qualitative data were collected through the Think-Aloud Protocol, capturing users&#39; verbalizations, difficulties, and interaction patterns during task execution. Expert evaluation revealed strong foundational design with good real-world alignment, consistency, and flexibility. However, refinements were needed in error handling mechanisms, user feedback systems (particularly progress indicators), help button visibility, and instruction simplification. Touch sensitivity issues emerged specifically for the tablet version. User testing demonstrated that the HMD version achieved higher spatial presence and ecological validity ratings across both groups, confirming the immersive potential of 360° media when delivered through head-mounted displays. Regarding safety, SMC participants using the tablet reported no significant cybersickness effects. Although MCI participants showed elevated baseline cybersickness scores, statistical analysis revealed significant reductions in oculomotor symptoms post-exposure, indicating the application did not induce adverse effects. For the HMD version, neither group experienced significant pre-post differences in cybersickness symptoms, confirming tolerability. The User Experience Questionnaire showed that both versions were perceived as highly novel by all participants. SMC participants provided favorable ratings across multiple dimensions for both versions, while MCI participants showed more neutral evaluations of pragmatic aspects, suggesting greater cognitive demands. The Think-Aloud Protocol identified specific usability challenges including difficulties with 360° spatial exploration, instruction comprehension, device interaction, and target recognition, particularly among MCI participants. This comprehensive evaluation demonstrates DUAL-REHAB feasibility and general tolerability while identifying critical refinement areas before clinical implementation. The findings highlight theimportance of user-centered design and preliminary usability testing when developing 360° media-based interventions for cognitively frail older adults. By emphasizing the unique characteristics of 360° media—ecological authenticity, reduced technical barriers, and adaptability across immersive and non-immersive platforms—this study contributes to establishing best practices for accessible, safe, and therapeutically effective digital rehabilitation tools.

ABSTRACT. Background: The global population of older adults is expanding rapidly, with projections indicating there will be 1.4 billion individuals aged 60 and older by 2030. Major Depressive Disorder (MDD) is a critical concern for this demographic, with a global prevalence of approximately 32%. Furthermore, while suicide rates have declined globally among younger age groups, they remain highest worldwide among older populations. In later life, depression is often triggered by biological declines—such as compromised mobility, balance, and coordination—and psychosocial losses, including the death of significant others and withdrawal from meaningful social roles. These factors often create a "cycle of depression," where reduced activity leads to a loss of positive reinforcement and mastery, which in turn fuels lower mood and further withdrawal. While Behavioral Activation (BA) is an evidence-based intervention designed to disrupt this cycle, traditional implementation faces unique barriers for older adults. Our previous studies for adults diagnosed with MDD demonstrated efficacy using virtual reality (VR) to enhanced BA. This study examines how VR enhanced BA may be used for older adult populations and discusses their unique challenges with MDD treatment.

Objective: This clinical pilot primarily aims to evaluate the feasibility, tolerability, and acceptability of an adapted five-session BA protocol using VR-simulated positive activities. An exploratory secondary objective is to evaluate preliminary clinical efficacy by tracking MDD symptoms and immediate post-activity mood changes.

Methods: The study targets a sample of 30 participants (aged 65+) who meet DSM-5 criteria for MDD. Candidates are screened using the Mini International Neuropsychiatric Interview and the Montreal Cognitive Assessment (MOCA-blind) to ensure they are without significant cognitive impairment. Exclusion criteria include recent substance use disorders, psychosis, Bipolar I disorder, untreated epilepsy, or changes in psychotropic medications within two months. The intervention consists of five weekly, 30-to-50-minute clinician-led sessions. Utilizing MyndVR headsets with simplified gaze-based navigation, participants select from an immersive 360-degree video library (e.g., travel, music, nature) to experience activities at home between sessions. Participants are also provided with activity calendars to track their engagement and mood throughout the week. The first session is focused on psychoeducation and activity-mood tracking. Session 2 includes a VR orientation followed by “activity scheduling” 360-degree videos (e.g., travel, nature, music) to experience at home between sessions. Sessions 3-4 discuss activity-mood connections and addresses barriers. The final session includes a review of progress, relapse prevention planning, and qualitative feedback collection.

Assessments: Feasibility is measured through dropout rates and observed qualitative barriers. Tolerability and acceptability are assessed with a custom post-treatment questionnaire. Clinical efficacy is tracked using the 15-item Geriatric Depression Scale. Additionally, participants use a "Mood Thermometer" to rate their mood on a scale of 0–10 before and after each VR activity. Quantitative analysis will involve independent samples T-tests to assess changes in GDS scores over time.

Status of Study: Enrollment began on 09/30/25.

Results: There will be a qualitative description of the midpoint results.

Significance: This is the first study to apply virtual reality enhanced behavioral activation to a clinical population of older adults diagnosed with MDD. The observations generated will justify continuation of the study and further investigation with a larger randomized controlled trial powered for testing efficacy. The findings will inform the development of scalable digital health interventions to assist individuals with MDD across the lifespan.

AI Disclosure: Text was formatted with help of Gen-AI (Gemini) to ensure abstract would be under word count limit.

ABSTRACT. In helicopter mission management, pilots must continuously adapt to changing mission requirements while handling flight routing and tactical planning as separate tasks, which increases cognitive workload. Recent studies (Roques et al., 2025) have shown that mental workload can be estimated continuously using operational and physiological data, supporting the development of systems capable of monitoring operator state in real time. The importance of such studies lies in the contributions to the improvement of the Human-Machine Interface (HMI) in such complex and demanding works. Despite such advancements in the psychological optimization capacity of HMIs, psychological assessments remain primarily based on self-report questionnaires that rely solely on declarative data. Despite significant ongoing research into the relationship between physiological indicators and emotional or mental states (Wang, Y. Et al., 2022), a unified system enabling psychologists to assess personality traits while concurrently monitoring biosignals during task execution is still lacking. Parting from this conjuncture, we pose the question: can personality traits be correlated to personal differences in physiological and physical data gathered in comparable conditions to those conducted in the study by Roques et al. (2025)? Parting from previous studies investigating personistic and interactionistic models of behavior (Stemmler, G., & Wacker, J., 2010), we aim to investigate two hypothesis: - First, that predominantly extraverted individuals have lower physiological arousal in potentially distressing situations; - Second, that individuals with high neuroticism trait levels have higher variance in physiological arousal. This study proposes to bridge such a gap with a framework that combines psychometric personality assessment with real-time multimodal biosignal acquisition to investigate the relationship between personality traits, physiological responses, and cognitive workload during human–machine interaction. Parting from robust existing assessments such as the Five-Factor Model (FFM) of personality, our method will consist on participants performing high cognitive workload demanding tasks such as the Multi-Attribute Task Battery (MATB-II) (National Aeronautics and Space Administration, n.d.) after completing such standardized personality inventories. The MATB-II acts as a flight simulator that mimics the complex tasks pilots perform, such as monitoring systems, tracking, and communicating under different levels of workload. Given the specific foundations of this proposed research, the target population to be assessed are male adult helicopter pilots (or trainee pilots). During questionnaire and task execution, physiological and behavioral data, including heart frequency, respiration rate, gaze behavior, skin conductance fluctuations and interaction metrics are collected using wearable sensors. Data fusion and machine learning methods will be applied to identify the patterns linking personality traits to psychophysiological responses to investigate the hypotheses. Due to the innovative nature of this investigation, as well as the well-known small-to-moderate effect sizes of correlations between personality traits and physiological markers (Stemmler, G., & Wacker, J., 2010), the presented hypotheses are open to an exploratory approach in this framework, aiming to set comparative parameters in physiological behaviors between personality profiles. Given the limited sample size reachable in the research timeline, the proposed analyses should be interpreted as exploratory and aimed at assessing feasibility rather than providing population-level inferences. The proposed approach aims to improve the objectivity and ecological validity of personality assessment and to support the design of adaptive human-centered systems capable of responding to operator cognitive and emotional states.

ABSTRACT. Background and Objective: In the context of the increasing frequency and severity of climate-related disasters, establishing stable infrastructure for post-disaster interventions is more important than ever. According to the World Health Organization (WHO), mental health problem prevalence in affected populations increases two to three times compared to the general population. Existing post-disaster guidelines therefore incorporate psychological first aid (PFA) to mitigate prolonged psychological distress such as PTSD, depression, and anxiety disorders. However, human-led crisis intervention teams face severe limitations in scalability and reachability during mass casualty events, while digital mental health tools and teleconsultations depend on communication infrastructure that is often destroyed in natural disasters. The objective of this work-in-progress study is to assess the feasibility of a scalable AI chatbot that delivers PFA to adult climate disaster survivors and adapts to the availability of communication infrastructure in real time.

AI Model: The aim is to build a safety-compliant AI model delivering PFA grounded in the WHO's five core principles: safety, calming, connectedness, self-efficacy, and hope. The Large Language Model (LLM) will be fine-tuned around the PFA "Look, Listen, Link" action principles, prioritizing immediate safety and practical needs over psychotherapeutic exploration. Keyword-based safety protocols will identify suicidal ideation and self-harm, directing users to human support such as emergency services or crisis lines. The system will clearly disclose its AI nature and support multilingual delivery.

Infrastructure: In order to mitigate the challenges related to the availability of communications infrastructure after a disaster, the proposed intervention explores the possibility of a multi-pathway system that seamlessly transitions between cellular and offline modes. Access to the chatbot is enabled through SMS; however, in the absence of a stable cellular network, communication is ensured through the utilisation of long-range (LoRa) devices that are connected within a mesh network.

Study Design: A formative feasibility pilot will test the prototype on adults with prior lived natural disaster experience. A sample size of N = 20 is anticipated. Participants will receive a hypothetical climate disaster scenario and will engage with the chatbot. Due to ethical considerations and the system's early stage of development, its testing will not be conducted in an immediate post-disaster context. Participants with ongoing PTSD or trauma are excluded from the sample to prevent potentially distressing re-exposure for vulnerable individuals. Acceptability and usability will be assessed using the System Usability Scale and the Net Promoter Score, while perceived helpfulness will be assessed using a custom scale aligned with the five WHO PFA core principles. A semi-structured interview will be used to deepen the understanding of participants’ experience and explore suggested improvements.

The study is currently in the design and development phase, with ethics review underway and no data yet collected. To our knowledge, no prior work has delivered an AI tool grounded in PFA guidelines that uses mesh networking to support the transition to offline mode in active disaster zones. It is therefore expected to provide insights into the feasibility of offering scalable, infrastructure-resilient, AI-based mental health support in post-disaster settings.

ABSTRACT. Background: Implementation remains a major challenge for evaluating the effectiveness of digital health interventions in real-world settings. In perinatal mental health, evidence on how to implement mobile health (mHealth) interventions within routine public healthcare is still limited, contributing to the gap between intervention effectiveness and real-world impact. Informed by Normalization Process Theory (NPT), the e-Perinatal project explores multilevel ecological barriers and facilitators shaping the normalization of a multicomponent preventive mHealth intervention within routine public primary maternal care (ERC Starting Grant 2021, No. 101042139).

Aim: To understand the implementation processes influencing the outcomes of the e-Perinatal mHealth intervention. Specifically, how healthcare professionals incorporate, recommend, and normalize the intervention within routine maternal care, and how perinatal women and fathers experience the digital health tool, engage with its use, and perceived impact.

Methods: A mixed-methods study was conducted by using an ecological approach to evaluate the implementation of the e-Perinatal intervention within routine maternal care, following Medical Research Council. Quantitative and qualitative data were analysed separately and subsequently integrated through a secondary comparative analysis to examine interactions between professional and organizational implementation processes and user-level mechanisms influencing outcomes. Patient and Public Involvement and Engagement was reported following the GRIPP2 checklist, and methodological quality was appraised using the Mixed Methods Appraisal Tool.

The user sample (N = 13) comprised pregnant (n = 9) and postpartum women (n = 1) and men during pregnancy (n = 1) and postpartum (n = 2) stratified by level of app use (high, medium and low). The professional sample (N = 8) was comprised by healthcare and policy stakeholders, including coordinating heads of midfwifery services (n = 3), intervention-group midwives (n = 2), mental health referral pathway professionals (n = 2) and a policy-level decision-maker (n = 1).

Normalization processes among professionals were assessed using the Normalization Measure Development (NoMAD) questionnaire, based on NPT, combined with in-depth qualitative interviews informed by the NPT. User-level mechanisms of change were explored through semi-structured qualitative interviews with perinatal women and men, focusing on experiences of use, engagement patterns and perceived impact.

Results: Qualitative interviews with perinatal women and men identified five main themes: the intervention as a reliable source of information compared to other digital resources; emotional support and normalization of perinatal experiences, particularly during pregnancy; selective and context-dependent use driven by specific needs; reduced use during the postpartum period due to time constraints and overload; and the perception of the intervention as a complement to, rather than a substitute for, professional care. Facilitators included active professional recommendation, clarity of content, and perceived emotional support, while barriers were mainly related to technical difficulties, lack of time, and limited professional follow-up.

Descriptive NoMAD results showed predominantly high agreement across coherence, cognitive participation, and reflexive monitoring domains, indicating good understanding, engagement, and positive appraisal of the intervention. In contrast, collective action items showed greater variability, particularly regarding training, available resources, and organizational support.

Professionals described e-Perinatal as a useful complement to routine care, especially in contexts of limited consultation time. Midwives emerged as key agents for recommendation and normalization. Barriers included technical problems, limited integration into clinical workflows, unequal engagement across professional roles, and structural constraints related to workload and coordination between care levels.

Conclusion: The normalization of e-Perinatal seems to be shaped by multilevel ecological factors, including technical accessibility, professional engagement, and women and father´s life stage. The identified barriers and facilitators may contribute to informing future implementation strategies and to guide the design of the large-scale randomized controlled trial in routine maternal care.

ABSTRACT. Video games are increasingly recognized as one of the most emotionally engaging contemporary media, eliciting not only pleasure and excitement but also frustration, anxiety, guilt, and moral discomfort. Building on a previously published theoretical account of the player as a co-creator of emotional experience in video games (Pinto, 2024), this research expands that framework by connecting it to psychological theories of emotion and engagement. The purpose is to situate the player’s dual role and co-authorship of emotional experience within established psychological literature, clarifying how interactive systems support participatory forms of emotional engagement.

A defining feature of video games is the player’s dual role: players exist simultaneously outside the game as reflective subjects aware of its fiction and rules, and inside it as situated agents whose actions have consequences (Pinto & Luz, 2024; Juul, 2011). Emotional engagement arises from this dual positioning, as players willingly commit to the game’s rules and fiction, enacting what Murray (2017) calls “active creation of belief”. Rather than identifying with characters representationally, players enact decisions and accept outcomes that are partly system-authored and partly self-produced. Emotions such as tension, responsibility, or regret are thus co-constructed through play (Pearce, 2004), making emotional engagement procedural rather than purely narrative-driven.

Players actively seek out games for pleasure and emotional experiences, voluntarily engaging in challenges that elicit a wide range of emotional states (Oliver & Raney, 2011). In this context, video games provide a suitable domain for the realization of the Affective Loop (AL), defined as “a system that is able to successfully elicit, detect, and respond to the emotions of its user” (Yannakakis & Paiva, 2015, p. 2). The AL unfolds through cyclical interactions between emotional stimuli, player response, and system feedback, producing emotionally responsive play structured by valence and engagement. Players often remain engaged during emotionally charged experiences, which can function as intrinsically motivating states that support persistence and challenge-seeking, even without immediate rewards.

From this combined perspective, emotional engagement in games emerges from a reciprocal relationship between player and system. Games present situations and constraints to which players respond through action, hesitation, or compliance, shaping subsequent emotional states. Because players are aware of their participatory role, emotional engagement carries an implicit dimension of responsibility: feelings of guilt, attachment, relief, or frustration are tied to the recognition that this happened because of me. Emotional engagement is therefore inseparable from agency, even in games that offer limited or illusory choice.

Player agency enables meaningful action and the observation of consequences within the game world, fostering perceived control and emotional involvement. Through co-creation, players shape narrative trajectories, intensifying attachment to characters and outcomes. Psychological frameworks such as Cognitive Appraisal Theory explain why players experience distinct emotional responses to the same game situations, as emotions emerge from subjective evaluations of goals, expectations, and perceived control (Lazarus, 1991; Frijda, 1986). Emotion can further be understood as a state of action readiness guiding engagement with the environment. Emotional meaning in games emerges from the interaction between player action and system response, captured by the concept of meaningful play (Salen & Zimmerman, 2003). Variations in how a game is presented and the type of feedback provided can influence players’ emotional appraisal of the same mechanics (Jørgensen, 2016). This emotional response is closely related to player motivation, which can be understood in terms of intrinsic and extrinsic factors (Malone, 1980).

This framing has implications for both psychology and game studies, supporting process-oriented accounts of emotion grounded in interaction and feedback while reframing emotional engagement as a co-authored process grounded in the player’s dual role, bridging psychological models of emotion and game-studies accounts of agency.

ABSTRACT. Deceptive design patterns refer to design practices that trick users into performing actions they did not intend to take, potentially causing harm (Brignull, 2023). These practices have gained increasing attention as empirical evidence demonstrates that dark patterns are strikingly effective in manipulating consumer behavior for economic benefit, exploiting cognitive biases to produce elevated rates of unintended user actions (Luguri & Strahilevitz, 2021). This is particularly relevant in video games, which have a documented history of harms associated with deceptive patterns, including compulsive in-game spending, gambling-like behaviors linked to loot box mechanics, and broader problematic gaming behaviors (Zendle & Cairns, 2018; King et al., 2019; Gainsbury, 2019). Deceptive patterns have been studied across different populations, but factors modulating vulnerability show inconsistent results depending on the type of pattern and context examined (Bongard-Blanchy et al., 2021). Evidence suggests that older adults show elevated susceptibility to financial fraud and scams, although psychological and functional vulnerability emerges as a stronger predictor than chronological age alone, and effects remain heterogeneous across pattern types and settings (Lichtenberg et al., 2016). This gap points to the need for measures capturing individual differences in cognitive functioning. Cognitive age will be operationalized as the out-of-sample age-prediction error from a machine learning model trained on 17 in-game behavioural metrics from Tunnel Runner, a game measuring reaction speed, response inhibition, interference control, response-rule switching, and decision-making under uncertainty (Markovitch et al., 2025). It is therefore expected to offer a more precise predictor of vulnerability to deceptive patterns than chronological age alone. This study aims to examine whether cognitive age predicts receptiveness to deceptive patterns in video games more accurately than chronological age, to inform therapy, harm prevention, and self-assessment. This poster aims to report the study design before data collection. We will conduct a cross-sectional pilot study with a linear regression design. Cognitive age will be estimated using performance on Tunnel Runner (Markovitch, Markopoulos, & Birk, 2024). Participants will subsequently interact with a Match-3 video game incorporating deceptive design patterns (Brignull, 2023; Gray et al., 2018; Mathur et al., 2021): scarcity/urgency cues, interface interference, and obfuscated pricing. Behavioural outcome measures will include offer acceptance rate, time to decision on offers, and session-level persistence via optional continuations prompted by urgency cues. Player experience will be assessed using the Player Experience Inventory (Vanden Abeele et al., 2020) and the NASA Task Load Index (Hart & Staveland, 1988). A convenience sample of 20 participants aged 18 to 80 years will be recruited. Given the pilot nature of this study, this sample size is intended to provide initial evidence that the targeted constructs are detectable within the game paradigm. Inclusion criteria will require basic familiarity with touchscreen or computer interfaces; individuals with diagnosed neurological or psychiatric conditions affecting cognition will be excluded. The study has been designed in accordance with ethical guidelines; all participants will provide informed consent before participation. Cognitive age scores will be derived by applying the validated stacked ensemble model from Markovitch et al. (2025). As this is a pilot study, simple linear regression is appropriate given the primary goal of establishing the reliability of the game-based behavioural outcome measures. With this pilot study, we aim to provide preliminary evidence for cognitive age as a predictor of receptiveness to deceptive design patterns and to identify cognitive profiles that may be associated with such vulnerability in video games.

ABSTRACT. I sought to solve the issues of user control, interactivity, and excitement for a virtual reality (VR) application based on the eye-movement desensitization and reprocessing (EMDR) therapy. I discovered design dilemmas and issues in user experience by observing participants while they were using a desktop version of Psylaris’ EMDR-VR application. Based on the findings, I designed and developed a new VR prototype, targeting an increased sense of control, interactivity, and mood, as well as implementing a calming natural environment based on research from environmental psychology. The sense of control was increased by giving the user more options to decide when to proceed and removing interactions that made the users feel pressured. The interactivity was increased by introducing locomotion, grabbing, throwing, and more distraction exercises compared to the old application. The hypothesis was that the immersive VR prototype would increase a sense of control, interactivity, and mood in comparison to the desktop EMDR application. The prototype was studied on participants using an Oculus Go VR headset and it saw a reported increase in sense of control and a reported decrease in negative mood using the The Brief Mood Introspection Scale (BMIS). While the sample and effect sizes were not sufficient to draw statistically significant conclusions, this reported decrease in negative mood could pave the way for future utility in using VR for improving well-being. Implications and future directions are discussed.