AI agents are increasingly becoming part of daily life, with advances in large language models (LLMs) enabling people to use them not only for practical tasks but also in increasingly social, relationship-oriented roles. This cross-sectional, between-subjects survey (n=48) examined whether relationships with AI agents show emotional coregulation patterns similar to those in human relationships, based on self-reports from users in friendship or romantic relationships with ChatGPT or Replika. The type of AI agent (Replika or ChatGPT) was treated as an independent variable in the analyses. Both agents offer similar conversational abilities, but Replika is presented as a virtually embodied, human-like companion, whereas ChatGPT is a text-based tool not marketed for companionship. We asked both groups of users to rate their experience with the agent in terms of emotional contagion (sharing emotions), emotional counter-regulation (balancing emotions), and affect change, as well as their perceptions of the agent’s self-disclosure (sharing personal or intimate information) and human-likeness. Participants reported emotional contagion and counter-regulation with both agents. Additionally, conversations with both AI agents significantly increased positive affect and reduced negative affect, with emotional counter-regulation emerging as a key factor in affect improvement across both groups. However, human-likeness did not predict emotional coregulation or affect changes. These findings suggest AI agents can support emotional coregulation, with the agent’s self-disclosure playing a role, but the contribution of virtual embodiment remains unclear.

As social robots gain prominence in supporting older adults’ health and well-being, understanding their effectiveness compared to virtual agents remains critical. This study investigated older adults’ perceptions of a physically embodied robot versus its virtual counterpart when taking on the role of an exercise coach. We recruited 25 healthy older adults, each of whom performed a series of exercises with both the physical NAO robot and its virtual simulation displayed on a computer screen. Participants’ experiences were assessed using the Unified Theory of Acceptance and Use of Technology (UTAUT) and the User Engagement Scale (UES) questionnaires collected after each condition. Results indicated that the Perceived Sociability of the NAO robot was significantly higher in the physically embodied condition compared to the virtual condition. However, no significant differences were found in Anxiety, Attitude, Perceived Enjoyment, Perceived Usefulness, Social Intelligence, or Trust. Similarly, the physically embodied NAO scored higher in Perceived Usability and Aesthetic Elements, but no significant differences were observed in Focused Attention and Reward Factor. These results suggest that physical embodiment could enhance perceptions of sociability and usability, however, it does not necessarily impact all engagement-related factors. Our findings contribute to the design of future socially assistive technologies in eldercare.

Anthropomorphic agents are generally evaluated more positively and trustworthy by human users than agents that are not humanlike. However, subtle mismatches in an agent’s appearance and behavior can lead to perceived uncanniness resulting in a disrupted trust during human-agent interaction. This study investigated the impact of an agent’s appearance and voice mismatch on user perception of the agent and their level of trust during a collaborative decision-making task. In a 2×2 between-subjects experimental design, participants performed an emotion recognition task while receiving recommendations from a virtual agent that either had a humanlike or robotic appearance with either a humanlike or synthesized robotic voice (4 conditions). Trust was measured both subjectively using a questionnaire and behaviorally by evaluating participants’ conformity to the agent’s input in their final decision-making. Results indicated that while the agent’s voice-appearance mismatch affected participants’ perception of anthropomorphism, it was not an influential factor in people’s trusting behavior. We discuss these results in the context of task complexity and make recommendations for future research.

This study aims to estimate an audience's comprehension of the content in a presentation and investigate the relationship between the difficulty of the presentation topics, the quality of explanations, comprehension, and brain activity patterns. Four types of videos with different presentation characteristics were prepared, and brain activity during video viewing was measured using electroencephalography (EEG) sensors. Subsequently, multiple features were generated from the acquired EEG data, and differential evaluations between two types of features were conducted in all six cases. We generated classifiers by features from 16 participants and evaluated them. In the identification of videos with difficult topics explained poorly and videos with easy topics explained well, a maximum accuracy of 71% (21% above the chance level) was recorded. Furthermore, although the identification accuracy between videos with difficult topics explained well and videos with easy topics explained poorly was approximately 60%, applying network analysis to the generated features improved the accuracy up to a maximum of 70%. These results suggest that the meaning of words we usually use ambiguously (difficult/easy, good/bad explanation, and understood/not understood) can be clarified using brain activities.

This paper investigates the efficacy of a passive Brain-Computer Interface (BCI) in enabling a robot tutor to adaptively respond to a user's engagement level in real-time. The BCI system extracted EEG Engagement Index from the user's electroencephalography (EEG) signals as an indicator of engagement during Human-Robot Interaction (HRI). A within-subjects study was conducted in which the robot performed attention-recapturing behavior during a learning task under two conditions; either in an adaptive manner whenever a lapse in the user's engagement level was detected by the BCI system (Adaptive condition) or at random intervals regardless of the user's mental states (Random condition). In both conditions, users completed an information retention test following the interaction. The study found no significant difference in the postinteraction test results or mean EEG Engagement Index values between the Adaptive and Random conditions. However, analysis of 10-sec time windows following robot interventions showed that adaptively timed gestures were significantly more effective in restoring user engagement to optimal level compared to randomly timed gestures. This finding provides evidence for the potential of passive BCIs in improving user experience in pedagogical HRI settings.

The current study investigated how robot tutors influence brain activity during child-robot interaction (CRI) for learning of second language vocabulary. We gathered EEG signals from two groups of children; 1) Robot group (N=21) who listened to a storytelling social robot and learned French words, and 2) Display group (N=20) who listened to the same story in the French language mediated by only a computer screen. To measure learning-induced changes in the brain, functional connectivity analysis was conducted on EEG signals, which quantifies the communication between brain regions during the learning phase. Results showed a significantly higher functional brain connectivity for the Robot group in the theta frequency band, which has been previously associated with language functions in neuroscientific literature. Our results provide neurophysiological evidence for the benefit of robot tutors in second language learning in children.

We introduce an interactive embodied conversational agent for deployment in the healthcare sector. The agent is operated by a software architecture that integrates speech recognition, dialog management, and speech synthesis, and is embodied by a virtual human face developed using photogrammetry techniques. These features together allow for real-time, face-to-face interactions with human users. Although the developed software architecture is domain-independent and highly customizable, the virtual agent will initially be applied to healtcare domain. Here we give an overview of the different components of the architecture.

Social robots have been shown effective in pedagogical settings due to their embodiment and social behavior that can improve a learner’s motivation and engagement. In this study, the impact of a social robot’s motivational gestures in robot-assisted language learning (RALL) was investigated. Twenty-five university students participated in a language learning task tutored by a NAO robot under two conditions (within-subjects design); in one condition the robot provided positive and negative feedback on participant’s performance using both verbal and non-verbal behavior (Gesture condition), in another condition the robot only employed verbal feedback (No-Gesture condition). To assess cognitive engagement and learning in each condition, we collected EEG brain activity from the participants during the interaction and evaluated their word knowledge during an immediate and delayed post-test. No significant difference was found with respect to cognitive engagement as quantified by the EEG Engagement Index during the practice phase. Similarly, the word test results indicated an overall high performance in both conditions, suggesting similar learning gain regardless of the robot’s gestures. These findings do not provide evidence in favor of robot’s motivational gestures during language learning tasks but at the same time indicate challenges with respect to the design of effective social behavior for pedagogical robots.

Brain Computer Interface (BCI) technology offers the possibility to monitor users’ attention and engagement during learning tasks, enabling adaptation of pedagogical strategies for a personalized learning experience. In this paper, we present an EEG-based passive BCI system for real-time evaluation of user engagement during a language learning task. The EEG Engagement Index, which has been previously associated with attention and vigilance, is measured from three frontal electrodes and used in this system as a neural indicator of engagement. To validate our system, we used it in a human-robot interaction (HRI) setting, in which a robot tutor monitored the learner’s brain activity and adapted its tutoring strategy when a lapse in engagement was detected. We discuss the challenges and preliminary results from our pilot study with eight participants.

With forty-six Action Units (AUs) forming the building blocks in the Facial Action Coding System (FACS), millions of facial configurations can be formed. Most research has focused on a subset of combinations to determine the link between facial configurations and emotions. Despite the value of this research for psychological and computational reasons, it is not clear what the most common combinations of AUs are to form the most commonly expressed facial configurations. We used three diverse corpora with human coded facial action units for a computational analysis. The analysis demonstrated that the largest portion of facial behavior consists of the absence of AU activations, yielding only one specific facial configuration, that of the neutral face. These results are important for cognitive scientists, computer graphics designers and virtual human developers alike. They suggest that only a relatively small number of AU combinations are initially needed for the creation of natural facial behavior in Embodied Conversational Agents (ECAs).

Mindfulness is the state of paying attention to the present moment on purpose and meditation is the technique to obtain this state. This study aims to develop a robot assistant that facilitates mindfulness training by means of a Brain-Computer Interface (BCI) system. To achieve this goal, we collected EEG signals from two groups of subjects engaging in a meditative vs. non-meditative human-robot interaction (HRI) and evaluated cerebral hemispheric asymmetry, which is recognized as a well-defined indicator of emotional states. Moreover, using self-reported affective states, we strived to explain asymmetry changes based on pre- and post-experiment mood alterations. We found that unlike earlier meditation studies, the fronto-central activations in alpha and theta frequency bands were not influenced by robot-guided mindfulness practice, however there was a significantly greater right-sided activity in the occipital gamma band of Meditation group, which is attributed to increased sensory awareness and open monitoring. In addition, there was a significant main effect of Time on participant's self-reported affect, indicating an improved mood after interaction with the robot regardless of the interaction type. Our results suggest that EEG responses during robot-guided meditation hold promise in real-time detection and neurofeedback of mindful state to the user, however the experienced neurophysiological changes may differ based on the meditation practice and recruited tools. This study is the first to report EEG changes during mindfulness practice with a robot. We believe that our findings driven from an ecologically valid setting, can be used in development of future BCI systems that are integrated with social robots for health applications.

Brain-computer interfaces (BCI) decode the electrophysiological signals from the brain into an action that is carried out by a computer or robotic device. Motor-imagery BCIs (MI-BCI) rely on the user's imagination of bodily movements, however not all users can generate the brain activity needed to control MI-BCI. This difference in MI-BCI performance among novice users could be due to their cognitive abilities. In this study, the impact of spatial abilities and visuospatial memory on MI-BCI performance is investigated. Fifty-four novice users participated in a MI-BCI task and two cognitive tests. The impact of spatial abilities and visuospatial memory on BCI task error rate in three feedback sessions was measured. Our results showed that spatial abilities, as assessed by the Mental Rotation Test, were not related to MI-BCI performance, however visuospatial memory, assessed by the design organization test, was higher in high aptitude users. Our findings can contribute to optimization of MI-BCI training paradigms through participant screening and cognitive skill training.

It is crucial that naturally-looking Embodied Conversational Agents (ECAs) display various verbal and non-verbal behaviors, including facial expressions. The generation of credible facial expressions has been approached by means of different methods, yet remains difficult because of the availability of naturalistic data. To infuse more variability into the facial expressions of ECAs, we proposed a model that considered temporal dynamic of facial behaviors as a countable-state Markov process. Once trained, the model was able to output new sequences of facial expressions from an existing dataset containing facial videos with Action Unit (AU) encodings. The approach was validated by having computer software and humans identify facial emotion from video. Half of the videos employed newly generated sequences of facial expressions using the model while the other half contained sequences selected directly from the original dataset. We found no statistically significant evidence that the newly generated facial expression sequences could be differentiated from the original ones, demonstrating that the model was able to generate new facial expression data that were indistinguishable from the original data. Our proposed approach could be used to expand the amount of labelled facial expression data in order to create new training sets for machine learning methods.

People's responses to a hypnosis intervention is diverse and unpredictable. A system that predicts user's level of susceptibility from their electroencephalography (EEG) signals can be helpful in clinical hypnotherapy sessions. In this paper, we extracted differential entropy (DE) of the recorded EEGs from two groups of subjects with high and low hypnotic susceptibility and built a support vector machine on these DE features for the classification of susceptibility trait. Moreover, we proposed a clustering-based feature refinement strategy to improve the estimation of such trait. Results showed a high classification performance in detection of subjects' level of susceptibility before and during hypnosis. Our results suggest the usefulness of this classifier in development of future Bel systems applied in the domain of therapy and healthcare.