AI Agents for Facilitating Social Interactions and Wellbeing

Abstract

Wellbeing AI has been becoming a new trend in individuals' mental health, organizational health, and flourishing our societies. Various applications of wellbeing AI have been introduced to our daily lives. While social relationships within groups are a critical factor for wellbeing, the development of wellbeing AI for social interactions remains relatively scarce. In this paper, we provide an overview of the mediative role of AI-augmented agents for social interactions. First, we discuss the two-dimensional framework for classifying wellbeing AI: individual/group and analysis/intervention. Furthermore, wellbeing AI touches on intervening social relationships between human-human interactions since positive social relationships are key to human wellbeing. This intervention may raise technical and ethical challenges. We discuss opportunities and challenges of the relational approach with wellbeing AI to promote wellbeing in our societies.

Introduction

COVID-19 has revealed the importance of the sense of belongingness and loneliness in mental health of our societies (COVID-19 Mental Disorders Collaborators, 2021). Wellbeing has attracted attention of psychology and public health for improving the mental health of individuals and organizations and has become one of the main targets for public health organizations such as the World Health Organization (WHO; Topp et al. 2015) .

Wellbeing has been studied intensively in the context of psychology (Andrews et al. 1976; Diener et al. 2018; Topp et al. 2015) . In psychology, multiple constructs of wellbeing have been proposed (Dodge et al. 2012). For example, Ryff and Keyes proposed that wellbeing is composed of multiple factors such as autonomy, environmental mastery, personal growth, positive relations with others, purpose in life, and self-acceptance (Ryff and Keyes 1995). Although there are differences in emphasis among psychological theories, positive social relationships have been identified as a crucial factor.

There is a new trend to apply artificial intelligence (AI) to enhance wellbeing due to the development of emotion analysis technologies such as genome-wide analysis, computer ___________________________________ In T. Kido, K. Takadama (Eds.), Proceedings of the AAAI 2022 Spring Symposium “How Fair is Fair? Achieving Wellbeing AI”, Stanford University, Palo Alto, California, USA, March 21–23, 2022. Copyright © 2022 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). vision, and natural language processing (NLP). Multiple services have been introduced to analyze or intervene in mental health by accessing peoples' emotions. For example, analyses by personal genetic data (Fang et al. 2020) , images (Reece et al. 2017), and texts in social media (Chancellor et al. 2020) predict risks and mental conditions including mental disorders. Some applications further intervene in mental conditions based on theories of psychological intervention such as cognitive-behavioral therapies (CBTs; van Agteren et al. 2021). Although social factors are known to be crucial, most AI applications for wellbeing focus on individuals and much less on social groups. Given that we spend most of our time in multiple social groups such as family, workplaces, schools, social clubs, etc., the opportunities and potential impact of such group-targeted AI applications would be enormous. However, AI applications of social groups for wellbeing have attracted little attention.

Here, we present an overview of the emergent role of AIaugmented agents for social interactions. First, we investigate the literature on psychological wellbeing and provide a two-dimensional classification of AI-augmented agents: individual/group and analysis/intervention. The first dimension concerns whether wellbeing AI is used for the analysis or the intervention. The second dimension focuses on whether an AI-augmented agent targets individuals or groups. We point out opportunities for the recently emerging approach, the so-called relational approach, where AIaugmented agents are applied to human-human interactions within groups. Finally, we discuss challenges in the relational approach of AI- augmented agents. We shed light on broader opportunities for AI-augmented agents, and highlight technological and ethical challenges for promoting wellbeing in the real and virtual societies.

Social Construct of Wellbeing

The notion of wellbeing has attracted attention in the context of healthy individual lives and societies. Subjective wellbeing (SWB) has been widely measured as a screening tool for mental disorders based on self- reported questionnaires such as the WHO questionnaire. Several models of SBW have been proposed (Andrews et al. 1976, Dodge et al. 2012, Topp et al. 2015) . SWB is composed of multiple facets comprising two components (Schimmack et al. 2008; Luhmann 2012) : affective and cognitive evaluations of one's life. The affective evaluation measures the emotional experiences of people in daily lives while the cognitive evaluations measure how people evaluate their lives based on their ideals. The affective and cognitive aspects are associated with different scales such as daily emotional experience and life satisfaction, respectively (Diener et al. 2018). Recent studies also suggested that another supplementary factor, harmony in life in a social context is also associated with SWO (Kjell et al. 2016). Harmony in life reflects social and environmental situations and is associated with psychological balance and flexibility in life. Therefore, social factors play a critical role in SWO.

Social personalities for wellbeing have been widely studied, showing consistent results. A recent meta- analysis, for example, revealed that widely used personality factors (e.g. NEO-PI-R and HEXACO questionnaires) are correlated with several aspects of SWO such as life satisfaction, positive/negative affect, and positive relation with others (Anglim et al.. 2020) . The study especially found that these aspects of SWO are positively correlated with extraversion and conscientiousness although negatively correlated with neuroticism. The sensitivity of SWO, thus, could reflect the personality traits of subjects. It is noteworthy that extraversion, as well as neuroticism and conscientiousness, also influence related factors like depressive symptoms (Hakulinen et al. 2015) . Extraversion is a social indicator for higher positive relationships with others. Meanwhile, neuroticism is a social indicator for less positive relationships related to loneliness (Buecker et al. 2020) . The association between these personalities and wellbeing- related factors supports the idea to promote wellbeing via positive relationships.

It is an important question whether behavioral practice can change social relationships and wellbeing. Multiple attempts showed enhancement of wellbeing as well as associated factors by healthy behaviors such as exercising (Chekroud et al. 2018) and psychological interventions (van Agteren et al. 2021). A cross-sectional study from 1.2 million individuals in the U.S. showed that physical exercising routines such as popular team sports, aerobic, and gym activities decrease up to 22% of mental health burdens compared to the non-exercising group (Chekroud et al. 2018).

Furthermore, different psychological interventions such as behavioral activation interventions (BA), positive psychological intervention (PPI), and mindfulness-based interventions (MBI) also showed small-to-moderate effects on wellbeing (van Agteren et al. 2021). Social interventions alleviated social isolation (Dickens et al. 2011) and loneliness (Masi et al. 2010) . These empirical findings further support that interventions including exercises and psychological intervention for social relationships on wellbeing can promote wellbeing and prevent mental illness.

We spend many hours with family, friends, and colleagues. Subjective wellbeing in groups such as working place is also studied well (Harter et al. 2003; Jain et al. 2009) . Working environments and social networks influence wellbeing and healthy behaviors. Associations between work environment and wellbeing are known (Harter et al. 2003; Bowling et al. 2010) . Life satisfaction and other related factors such as job satisfaction and positive affect are related to wellbeing. Another evidence further showed that mindfulness training had small-to-moderate effects on psychological distress, wellbeing, and sleep (Bartlett et al. 2019) although the influence on work performance could not be concluded due to the insufficiency of pooled data. Internetbased interventions on workers showed small-to-moderate effects on work effectiveness and psychological wellbeing in workplaces (Carolan et al. 2017).

Psychological interventions on social networks, so- called social network interventions, are also effective on wellbeing (Hunter et al. 2019) . This relatively new approach cares for changes in information flow by intensifying, deleting, and transferring social ties (Valente 2012). The social network intervention is expected to enhance the effectiveness of health outcomes such as lower drug use, healthy sex behaviors, stronger social support, and wellbeing.

It is interesting to ask whether this approach is useful for social media and online gameplay. There is a strong public interest in the association of social media use and game playing with mental health. Their potentially harmful influences on mental health have often drawn public attention (Huang 2010; Prescott et al. 2018), but the relationship remains unclear, perhaps due to huge differences in design and concepts within social media (Sakurai et al. 202) and games (Johannes et al. 2021). Communication within online video games such as e- sports can be essential for effective team performance. Effective social intervention may increase not only team performance and wellbeing, but the potential of such social interventions remains clear. Findings on group wellbeing, nonetheless, reveal another potential target of interventions in our societies.

To sum up, existing literature revealed associations of social factors related to genetics, environments, and behaviors with wellbeing. These multiple findings clarify possibilities of interventions of subjective wellbeing as well as group wellbeing.

Types of Wellbeing AI

The effectiveness of interventions on wellbeing triggered expectations to conduct research and development along with a trend of digital therapeutics. Digital therapeutics are

Individual

Individual

Group

Group

Group

Analysis/Intervention

Analysis

Analysis

Intervention

Categories

Genetics

Mental Health

Health Care (Ahmed et al. 2021; Duradoni et al. 2021)

Analysis

Intervention

Intervention

Emotion Analysis (Veltmeijer et al. 2020)

Mental Health

Group Discussion Facilitation

Individual

Intervention

Workplace

Examples Ø Depression Risk (Fang et al. 2020) Ø Subjective Wellbeing (Røysamb et al. 2018) Ø Social Support (Wang et. al. 2017) Ø Emotion Detection (Canedo et al. 2019) Ø Screening Mental Health Status on Social Media (Chancellor et al. 2020) Ø Behavioral Cognitive Therapy (Woebot,

Todaki; Jang et al. 2021) Ø Promotion for Mental Wellbeing (Shim; Ly et al. 2017) Ø Cancer Cares for Young Survivors (Vivibot;

Greer et al. 2019) Ø A Chatbot for Improvement for Sedentary Behaviour and Wellbeing (Welbot; Haile et al.

2020) Ø Images (Tan et al. 2017) Ø Sounds (Franzoni et al. 2020) Ø Videos (Sánchez et al. 2020) Ø Social Media (Gong et al. 2019) Ø A Chatbot for Positive Messaging (Sunny; Narain et al. 2020) Ø Group Discussion Facilitation such as GlahBlahBot (Shin et al.. 2020) , Micbot (Tennent et al.. 2019) , Groupfeedbot (Kim et al. 2020) , Vulnerable-Robots (Traeger et al. 2020) evidence-based therapeutic interventions with software programs to cure and prevent medical disorders. There is also a trend to apply AIs to such evidence-based interventions in mental health (D’Alfonso 2020) . In this section, we summarize and explain the types of AI-augmented applications such as robots, avatars, and bots for wellbeing. By doing so, we clarify currently active approaches in wellbeing AI.

We categorize AI applications for wellbeing with two-dimensional axes: analysis/intervention and individual/group (Table 1.). The first dimension means whether the aim of digital wellbeing is analysis or intervention. Many potential applications focus on monitoring emotional states or related factors associated with wellbeing through genes (Fang et al. 2020) , images (Canedo et al. 2019) , and texts (Chancellor et al. 2020) from social media.

Other applications target the individual status of wellbeing-related factors through apps. Meanwhile, positive social networks are a crucial social basis for groups as well as individuals. There are only a few applications of wellbeing targeting groups from this perspective (Narain et al. 2020) . For example, a Facebook messenger chatbot, Sunny, is meant to promote social interactions and wellbeing within groups (Narain et al. 2020) . We belong to multiple social groups in different contexts such as schools, workplaces, sports clubs, and our houses. Scopes of group wellbeing should be also broad. There are some studies on group wellbeing and proposals to promote such group wellbeing from AI-augmented agents such as robots (Kim et al. 2020; Shin et al. 2020; Tennent et al. 2019; Traeger et al. 2020) . However, group wellbeing targeted by AI applications is relatively understudied and may bring new opportunities for the promotion of wellbeing.

In summary, most of the applications in wellbeing AI focus on analysis and psychological interventions of individual wellbeing through mobile devices. Meanwhile, group wellbeing is relatively under-explored but may have a huge impact on our societies.

The Relational Approach for Group Wellbeing with AI

In this section, we overview a relational approach for group wellbeing with the literature of analysis and interventions on human-human interactions with AI agents. By doing so, we outline opportunities of wellbeing AI for group wellbeing. We then discuss types of the relational approach: analysis of group dynamics and social connectedness. Finally, the challenges of the relational approach will be discussed. Literature Review on Analysis and Interventions for Social Groups with AI Detection and intervention of group wellbeing with AI are not well studied. However, related studies on automated group emotion and artificial agents for social interactions have been active recently.

Automated group-level emotion recognition has been studied recently since 2012 (Veltmeijer et al. 2020; Table. 1) . Group emotion is not a simple sum of individual emotions in a group. Instead, automated emotion recognition needs to track unique group emotion dynamics. A user survey has been developed as a proxy of such group emotion (Dhall et al. 2015) . Multiple studies predict group emotional labels based on various datasets from images (Tan et al. 2017), videos (Sánchez et al. 2019) , and social media (Gong et al. 2019) . Such studies target different sizes and states of seated, standing, and dynamic groups. Veltmeijer et al. pointed out three technical challenges (Veltmeijer et al. 2020) . First, group size changes. Second, subgroup emotions in a larger group can be different. Third, group emotion can also change. Although methods are under development, automated emotion detections for groups are perhaps applicable to group wellbeing detection. Several types of researches, applications, and commercial products for interactions with AI have been introduced in various situations such as education, hospitals, games, workplaces, social media, banks, online dating, sports, tourism, etc. These agents are expected to increase learning speed, team performance, successful dating matches, or satisfaction during traveling.

Not only AI without agents but also AI-augmented agents are widely used in our societies. We define such artificial agents as three types: robots, social bots, and avatars. AIaugmented agents are commonly used for cooperative purposes for interactions between human and artificial agents. Human-robot studies are commonly done to understand the capability of robots (Sheridan, 2016) and how humans recognize robots (Chae et al. 2016; Lucas et al. 2014). Social bots have also been studied for communications through apps and social media (Assenmacher et al. 2020) . Avatarhuman interactions are further studied in the context of remote learning although humans control such avatars in most studies (Chae et al. 2016). These studies aim for interactions between human and artificial agents.

An emergent application of artificial agents as social mediators is expected to promote social interactions between humans and prevent problematic behaviors within a group (Chita-Tegmark, 2020; Dafoe et al. 2021; Rahwan et al. 2020) . However, much fewer studies target social groups for wellbeing. Some recent studies worked on discussion facilitation with artificial agents (Kim et al. 2020; Traeger et al. 2020; Tennent et al. 2019) and wellbeing promotion (Narain et al. 2020) . The group intervened in a social group to induce conversations and engagement on problem-solving. One study with a messenger chatbot, Sunny, worked on group wellbeing by sending positive messages to 4 member groups and had positive effects on wellbeing (Narain et al. 2020) . These studies are mostly limited in discussion facilitation, but potential applications of social groups can be more extensive in different fields such as houses, schools, hospitals, caregivers, sports, workplaces, social media, tourism, where social groups are formed. For example, artificial agents could work for team engagements in sports by giving analysis or feedback based on their performance. The AI-augmented agents could also work on the mediation of conflicts between members in workplaces as well as enhancement of discussion facilitation. In doing so, we may expect artificial agents to promote wellbeing.

It is also critical to ask whether we explicitly need such artificial agents. It may be sufficient to have AIs without agents such as recommendation systems for e-commerce. One benefit of artificial agents could be related to attentional engagements by agents (Chae et al. 2016; Lucas et al. 2014; Mollahosseini et al. 2018; Spicer et al. 2021) . Multiple studies showed artificial agents enhance engagements (Oertel et al. 2020) . This attentional engagement can be augmented by the appearance of artificial agents (Li et al. 2010; Bente et al. 2008) . Several pieces of evidence also revealed that the appearance of artificial agents influences human trust of the agents and induces similar human behaviors to humans by the agents (Caruana et al. 2017; Lucas et al. 2014). It is uncertain whether artificial agents work best in all situations, but they may exert stronger influences than just non-agentive AIs via emulating human-like interactions.

In sum, previous studies on the analysis of emotional detection for social groups and intervention of social interactions are active. However, these analyses and interventions on social interactions have not yet merged, and few studies focus on group wellbeing.

Communicating with Social Human Groups via Artificial Agents The mediative role for human-human interactions with artificial agents has not been well studied. Potential opportunities of such artificial agents are more extensive than current opportunities for individual wellbeing. However, the mediative role of AI in group wellbeing, the so-called relational approach, has not been explicitly explored. Here, we clarify two types of a relational approach to social groups. By doing so, we prompt the development of relational approaches for group wellbeing.

One type of the relational approach is to analyze group dynamics itself from conversations or their behaviors (Figure.1.A). The previous studies on automated emotion detection target analyzing such group dynamics by facial expressions and conversations (Kim et al. 2020; Narain et al. 2020, Tennent et al. 2019; Traeger et al. 2020) . A robot agent study also targets group performance such as total conversation time (vulnerable-robots; Traeger et al. 2020) . This approach focuses on average or wholistic group dynamics not considering the relationship among members in a group.

Another type of the relational approach is to analyze oneto-one member interactions in detail (Figure.1.B). We also directly contact a person in management not by groups. For example, when certain group members engage in group discussion, another member familiar with one of the members may have insight. In this case, you as a mediator want to ask the member to promote his or her engagement in the discussion. Such a role can be served by an artificial agent. This type should be computationally intensive since N-to-N human interactions are analyzed based on methods such as computer vision and natural language processing (Poria et al. 2019) . Along with such development, computation power increases these days, so current computation power could be sufficient for human- human interactions within a few members.

It is interesting to ask whether these two types of the relational approaches can be integrated as computational methods like social network analysis (Gesell et al. 2013) and network controllability (Liu et al. 2011) . One potential key field is related to network analysis considering both each social connection and network organization. In neuroscience, analysis and intervention of whole-brain state with regionregion interactions are actively studied (Tang and Bassett 2018). Such network analysis further would bring an integrative perspective of social interactions and group dynamics. These approaches should be enriched by the development of the two types of the relational approach for group wellbeing. The distinction could be tentative but should be useful to work on wellbeing AI from a view of social interactions.

Challenges on Relational Approach and Potential Ethical Issues We discuss three types of challenges on designing and managing wellbeing AI for the human-human social interactions based on the relational approach. 1) Changes related to fairness issues of computation and authority from the viewpoint of different cultural contexts, conflict of interest, and structure of benefits. 2) Challenges related to the privacy of human-human interactions from the viewpoint of ownership and autonomy of communications. 3) Challenges related to usefulness from the viewpoint of users of accessibility and safety. These challenges must be overcome for the effective and successful introduction and management of wellbeing AI.

Fairness of computation is raised as an important issue in AI research. Fairness in AI research is composed of three perspectives: fairness, conflict of interest, and respect of different communities. First, each social connection within a group should be fairly considered. Asymmetrical social connections may cause issues within a group. Next, the introduction of wellbeing AI by administrators should be fairly considered for users. Wellbeing AI can be expected by administrators to enhance engagements of users in workplaces, social media and games. Such increased engagements have the potential to deteriorate life satisfaction causing burnout symptoms in the long term. Long-term wellbeing for users, then, should be considered. Third, different cultures of communities should be respected. Perception of wellbeing is known to differ in different communities and populations (Lai et al. 2013) . This example may reflect individual traits based on experience, personality, and genetics. Computation of social connections should not only consider a specific type of individual traits but also multiple perceptions.

Privacy of human-human interactions is another crucial issue. Multiple issues have been raised by previous actions by companies on controlling and using the private data of users. In this regard, privacy, autonomy, and ownership of social interactions should be considered. First, excessive access and storage of private data should not be permissible. Communications are performed with image, auditory, and text information but storing, analyzing, and providing to a third party should depend on the permission of users. Such actions potentially causing disadvantages to users should not further be taken. Second, whether actions by wellbeing AI are excessive should be considered. Such actions may cause behavioral constraints for users. Such interventions to corruption of autonomy in groups should not be permissible. Third, appropriate interventions on social connections should be considered. Related to autonomy preservation, some interventions may be permissible depending on group characteristics, but others not. This is perhaps related to a discussion of moral agency in AI where what AI agents are allowed to perform. Members within groups should determine which type of analysis and interventions is permissible.

Whether interventions to users are appropriate or not is critical. One reason is designing wellbeing environments can be more important than introducing wellbeing AI. For example, the relationship between employees and wellbeing is dependent on working environments. In other words, it might not be important to have such wellbeing AI if the working environment is not designed to promote wellbeing. This idea might be aligned with wellbeing in the importance of game design rather than the importance of introducing wellbeing AI. Appropriateness of wellbeing should be considered from usability, understanding, and the public interest of users. The stability of wellbeing AI is the priority to be considered. Attempts to have wellbeing AI is still under exploration. Real applications might not often be welcomed in social contexts. What factors determine such usability for users should be investigated. Second, understanding users is important. Related to usability, the mismatch between users and applications might be associated with misunderstanding of users by administrators. Third, a perspective of public interest is needed. This is a third-party view of wellbeing AI. Even though users and administrators gain benefits from applications of wellbeing AI, the relational approach may have a huge harmful impact on the public interest. Such appropriateness should be considered too.

Multiple challenges including three types of perspectives exist for designing and managing the relational approach of wellbeing AI since such approach is implicitly under development. Nonetheless, the relational approach of wellbeing AI has huge room to benefit our societies.

Conclusion

In this paper, we introduced the notion of AI-supported wellbeing in the era of digital worlds and presented an overview of the relational approach to promoting positive social interactions by analyzing and managing human-human interactions for the introduction of AI-supported wellbeing in the era of digital worlds. First, we described psychological research on wellbeing based on personality, genetics, and behavioral and cognitive interventions, and concluded that social relationships are crucial for wellbeing. We, then, identified an unexplored category of wellbeing AI and group wellbeing. Group wellbeing through telecommunications is especially critical since the expansion of telecommunications may cause psychological issues such as distress and loneliness which are reported during COVID-19.

By reviewing previous literature of interventions on social networks with a robot and virtual agents, we further introduced the relational approach, which analyzes and mediates human-human interactions with artificial agents such as chatbots and robot agents. The relational approach mediates human-human social interactions in the real or digital world to promote wellbeing and other factors such as team performance. Finally, we discussed potential challenges of design and usage of the relational approach in wellbeing AI to establish its successful support of human social networks. We shed light on the mediative roles of AI-augmented agents to benefit human mental health and wellbeing in real and digital environments. By doing so, we expect a broader understanding and further development of group wellbeing.

Acknowledgments

We acknowledge Toshifumi Sasaki in Osaka University for discussion. Research

Li, D.; Rau, P.L.P.; and Li, Y. 2010. A Cross-cultural Study: Effect of Robot Appearance and Task. International Journal of Social Robotics 2: 175-86. doi.org/10.1007/s12369-010-0056-9 Sakurai, R. et al. 2021. Who is Mentally Healthy? Mental health profiles of Japanese social networking service users with a focus Sánchez, F.L. et al. 2020. Revisiting Crowd Behaviour Analysis through Deep Learning: Taxonomy, anomaly detection, crowd emotions, datasets, opportunities and prospects. Information Fusion 64: 318-35. doi.org/10.1016/j. inffus.2020 .07.008 Schimmack, U. 2008. The Structure of Subjective Wellbeing. In The science of subjective wellbeing, edited by M. Eid & R.J. Larsen, 97-123. New York: Guilford Press.

References

Andrews , F.M. ; Withey , S.B. 1976 . Measuring Global Wellbeing .

Anglim , J. ; Horwood, S. ; and Smilie , L.D. 2020 . Predicting Psychological and Subjective Wellbeing from Personality: A MetaAnalysis . Psychological Bulletin 146 ( 4 ): 279 - 323 .

doi.org/10.1007/978-1- 4684 -2253- 5 _ 3 .

Assenmacher , D. et al.. 2020. Demystifying Social Bots: On the Intelligence of Automated Social Media Actors. Social Media + Society . doi.org/10.1177/2056305120939264 Bente, G. et al. 2008 . Avatar- Mediated Networking: Increasing Social Presence and Interpersonal Trust in Net-Based Collaborations . Human Communication Research 34 ( 2 ): 287 - 318 .

doi.org/10.1111/j.1468- 2958 . 2008 . 00322 .x Bartlett, L. et al. 2019 . A Systematic Review and Meta-Analysis of Workplace Mindfulness Training Randomized Controlled Trials .

Journal of Occupational Jealth Psychology 24 ( 1 ): 108 - 126 .

doi.org/10.1037/ocp0000146 Bowling, N.A. ; Eschleman , K.J.; and Wang , Q. 2011 . A Meta-Analytic Examination of the Relationship Between Job Satisfaction and Subjective Well-being . Journal of Occupational and Organizational Psychology 83 ( 4 ): 915 - 34 .

doi.org/10.1348/096317909X478557 Buecker, S. et al. 2020 . Loneliness and the Big Five Personality Traits: A Meta-analysis . European Journal of Personality 34 ( 1 ): 8 - 28 . doi.org/10.1002/per.2229 Chae, S.W. ; Lee , K.C. ; Seo , and Y.W. 2016 . Exploring the Effect of Avatar Trust on Learners' Perceived Participation Intentions in an e-Learning Environment . International Journal of Human-Computer Interaction 32 ( 5 ): 373 - 93 .

The Lancet 398 ( 10312 ): 1700 - 12 . doi.org/10.1016/S0140- 6736 ( 21 ) 02143 - 7 Dafoe , A. et al. 2021 . Cooperative AI : Machines Must Learn to Find Common Ground . Nature 593 : 33 - 36 .

doi.org/10.1038/d41586-021-01170-0 D 'Alfonso , S. 2020 . AI in Mental Health . Current Opinion Psychology 36 : 112 - 7 . doi.org/10.1016/j.copsyc. 2020 . 04 .005 Dhall, A. et al. 2015 . The More the Merrier: Analysing the Affect of a Group of People in Images . 11th IEEE International Conference and Workshops on Automatic Face and Gesture Recognition (FG).

Dickens , A.P. et al. 2011 . Interventions Targeting Social Isolation in Older People: a Systematic Review . BMC Public Health 11 ( 647 ).

doi.org/10.1186/ 1471 -2458-11-647 Diener , E.; Oishi , S. ; and Tay , L. 2018 . Advances in Subjective Well-Being Research . Nature Human Behaviour 2 , 253 - 60 .

doi.org/10.1038/s41562-018-0307-6 Duradoni, M. et al. 2021 . Robotic Psychology: A PRISMA Systematic Review on Social-Robot-Based Interventions in Psychological Domains . J 4 ( 4 ): 664 - 97 . doi.org/10.3390/j4040048 Dodge, R. et al. 2012 . The Challenge of Defining Wellbeing . International Journal of Wellbeing 2 ( 3 ): 222 - 35 .

doi:10 .5502/ijw.v2i3.

Fang , Y. et al. 2020 . Genomic Prediction of Depression Risk and Resilience under Stress . Nature Humam Behaviour 4 : 111 - 8 .

doi.org/10.1038/s41562-019-0759-3 Franzoni , V.; Biondi , G.; and Milani , A. 2020 . Emotional Sounds of Crowds: Spectrogram-Based Analysis Using Deep Learning .

Multimedia Tools and Applications 79 : 36063 - 75 .

doi.org/10.1007/s11042-020-09428 -x Gesell , S.B.; Barkin , S.L. ; and Valente, T.W. Social Network Diagnostics: a Tool for Monitoring Group Interventions . Implement Science 8 : 116 . doi.org/10.1186/ 1748 -5908-8-116 Gong , V.X. et. el. 2019 . Estimate Sentiment of Crowds from Social Media during City Events . Transportation Research Record 2673 ( 11 ): 836 - 50 . doi.org/10.1177/0361198119846461 Greer, S. et al. 2019 . Use of the Chatbot “Vivibot” to Deliver Positive Psychology Skills and Promote Well-Being Among Young People After Cancer Treatment: Randomized Controlled Feasibility Trial . JMIR Mhealth Uhealth 7 ( 10 ): doi.org/10.2196/15018 Haile, C. et al. 2020 . Pilot Testing of a Nudge-Based Digital Intervention (Welbot) to Improve Sedentary Behaviour and Wellbeing in the Workplace . International journal of environmental research and public health 17 ( 16 ): 5763 . doi.org/10.3390/ijerph17165763 Harter, J.K. ; Schmidt , F.L. ; and Keyes , C.L.M. 2003 . Well-being in the Workplace and its Relationship to Business Putcomes: A Review of the Gallup Studies. Flourishing: Positive psychology and the life well-lived, edited by C.L.M. Keyes & J. Haidt , 205 - 24 .

Hakulinen , C. et al. 2015 . Personality and Depressive Symptoms: Individual Participant Meta-Analysis of 10 Cohort Studies . Depression and Anxiety 32 ( 7 ): 461 - 70 . doi.org/10.1002/da.22376 Huang, C. 2010 . Internet Use and Psychological Well-being: a Meta-Analysis. Cyberpsychology, behavior and social networking 13(3 ): 241 - 9 . /doi.org/10.1089/cyber. 2009 .0217 Hunter, R.F. et al. 2019 . Social Network Interventions for Health Behaviours and Outcomes: A Systematic Review and Meta-analysis . PLoS medicine 16(9): e1002890. doi.org/10 .1371/journal.pmed.1002890 Jain, A.K. ; Giga , S.I.; and Cooper , C.L. 2009 . Employee Wellbeing, Control and Organizational Commitment . Leadership & Organization Development Journal 30 ( 3 ): 256 - 73 .

doi.org/10.1108/01437730910949535 Jang, S. et al. 2021 . Mobile App-based Chatbot to Deliver Cognitive Behavioral Therapy and Psychoeducation for Adults with Attention Deficit: A Development and Feasibility/Usability Study .

International journal of medical informatics 150 : 104440 .

doi.org/10.1016/j.ijmedinf. 2021 .104440 Johannes, N.; Vuorre , M. ; and Przybylski , A.K. 2021 . Video Game Play is Positively Correlated with Well-being . Royal Society Open Science 8 ( 2 ). doi.org/10.1098/rsos.202049 Kjell, O.N.E. et al. 2016 . The Harmony in Life Scale Complements the Satisfaction with Life Scale: Expanding the Conceptualization of the Cognitive Component of Subjective Well-Being . Social Indicators Research 126 : 893 - 919 . doi.org/10.1007/s11205-015- 0903-z Kim , S. et al. 2020 . Bot in the Bunch: Facilitating Group Chat Discussion by Improving Efficiency and Participation with a Chatbot.

In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems. Honolulu, HI, April 25 -30.

Lai , L.C.H. ; Cummins, R.A. ; and Lau A.L.D. 2013 . Cross-Cultural Difference in Subjective Wellbeing: Cultural Response Bias as an Explanation. Social Indicators doi .org/10.1007/s11205-012-0164-z Lucas, G.M. ; Gratch , J. ; King , A. ; and Morency , L. 2014 . It's only a computer: Virtual humans increase willingness to disclose . Computers in Human Behavior 37 : 94 - 100 .

doi.org/10.1016/j.chb. 2014 . 04 .043 Luhmann, M. et al. 2012 . Time Frames and the Distinction between Affective and Cognitive Well-being . Journal of research in personality 46 (4) : 431 - 41 . doi.org/10.1016/j.jrp. 2012 . 04 .004 Ly, K.H. ; Ly , A.M. ; and Andersson, G. A Fully Automated Conversational Agent for Promoting Mental Well-being: A Pilot RCT Using Mixed Methods . Internet Interventions 10 : 39 - 46 .

doi.org/10.1016/j.invent. 2017 . 10 .002 Masi, C.M. et al. 2011 . A Meta-Analysis of Interventions to Reduce Loneliness. Personality and social psychology review: an official journal of the Society for Personality and Social Psychology 15 ( 3 ) 219 - 66 . doi.org/10.1177/1088868310377394 Mollahosseini, A. et al. Role of Embodiment and Presence in Human Perception of Robots' Facial Cues . International Journal of Human-Computer Studies 116 : 25 - 39 .

doi.org/10.1016/j.ijhcs. 2018 . 04 .005 Narain, J. et al. Promoting Wellbeing with Sunny, a Chatbot that Facilitates Positive Messages within Social Groups . In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems . New York: Association for Computing Machinery.

doi.org/10.1145/3334480.3383062 Oertel, C. et al. Engagement in Human-Agent Interaction: An Overview . Frontiers in Robotics and AI 7 : 92 .

doi.org/10.3389/frobt. 2020 .00092 Poria, S. et al. 2018 . MELD: A Multimodal Multi-Party Dataset for Emotion Recognition in Conversations . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics.

doi.org/ 10.18653/v1/ P19 -1050 Prescott , A.T.; Sargent , J.D. ; and Hull , J.G. 2018 . Metaanalysis of the Relationship between Violent Video Game Play and Physical Aggression Over Time . Proceedings of the National Academy of Sciences of the United States of America 115 ( 40 ): 9882 - 8 .

doi.org/10.1073/pnas.1611617114 Rahwan, I.; Crandall, J.W. ; and Bonnefon , J. 2020 . Intelligent Machines as Social Catalysts . Proceedings of the National Academy of Sciences of the United States of America 117 ( 14 ): 7555 - 7 .

doi.org/10.1073/pnas.2002744117 Reece, A.G. ; and Danforth , C.M. 2017 . Instagram Photos Reveal Predictive Markers of Depression . EPJ Data Science 6 ( 15 ).

doi.org/10.1140/epjds/s13688-017-0110-z Røysamb , E.; and Nes, R.B. 2019 . The Role of Genetics in Subjective Well-being . Nature Human Behaviour 3 ( 3 ).

doi.org/10.1038/s41562-018-0494-1 Ryff, C.D.; and Keyes , C.L. 1995 . The Structure of Psychological Well-being Revisited . Journal of personality and social psychology 69 (4) : 719 - 27 . doi.org/10.1037//0022- 3514 . 69 .4.719 Sheridan, T. 2016 . Human-Robot Interaction: Status and Challenges . Human Factors 58 ( 4 ): 525 - 32 .

doi.org/10.1177/0018720816644364 Spicer, C. ; Khwaounjoo , P. ; and Cakmak, Y.Z. 2021 . Human and Human-Interfaced AI Interactions: Modulation of Human Male Autonomic Nervous System via Pupil Mimicry . Sensors (Basel 21 ( 4 ): 1028 . doi.og/10.3390/s21041028 Shin, D. et al. 2021 . BlahBlahBot: Facilitating Conversation between Strangers using a Chatbot with ML-infused Personalized Topic Suggestion . In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems . New York: Association for Computing Machinery.

doi.org/10.1145/3411763.3451771 Tan, L. et al. 2017 . Group Emotion Recognition with Individual Facial Emotion CNNs and Global Image based CNNs . In Proceedings of the 19th ACM International Conference on Multimodal Interaction . New York: Association for Computing Machinery.

doi.org/10.1145/3136755.3143008 Tennent, H.; Shen , S. ; and Jung M. 2019 . Micbot: A Peripheral Robotic Object to Shape Conversational Dynamics and Team Performance . In Proceedings of ACM/IEEE International Conference on Human-Robot Interaction (HRI) . New York: Association for Computing Machinery. doi.org/10.1109/HRI. 2019 .8673013 Tang, E.; and Bassett , D.S. 2018 . Colloquium: Control of Dynamics in Brain Networks . Reviews of Modern Physics 90 : 031003 .

doi.org/10.1103/RevModPhys.90.031003 Topp, C.W. et al. 2015 . The WHO-5 Well-Being Index: a Systematic Review of the Literature . Psychother Psychosom 84 ( 3 ): 167 - 76 . doi.org/10.1159/000376585 Traeger, M.L. et al. 2020 . Vulnerable Robots Positively Shape Human Conversational Dynamics in a Human-Robot Team . Proceedings of the National Academy of Sciences of the United States of America 117 ( 12 ): 6370 - 5 . doi.org/10.1073/pnas.1910402117 Valente, T.W. 2012 . Network Interventions. Science 337 ( 6090 ): 49 - 53 . doi.org/10.1126/science.1217330 van Agteren , J. et al. 2021 . A Systematic Review and Meta-Analysis of Psychological Interventions to Improve Mental Wellbeing .

Nature Human Behaviour 5 : 631 - 52 . doi.org/10.1038/s41562- 021-01093-w Veltmeijer , E.A.; Gerritsen , C; and Hindriks , K. 2021 . Automatic Emotion Recognition for Groups: a Review . IEEE Transactions on Affective Computing. doi.org/10 .1109/TAFFC. 2021 .3065726 Wang, R.A.H. et al. 2017 . Social Support and Mental Health in Late Adolescence are Correlated for Genetic, as well as Environmental, Reasons . Scientific Reports 7 : 13088 .

doi.org/10.1038/s41598-017-13449-2