Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being.

Conversational artificial intelligence (AI), particularly AI-based conversational agents (CAs), is gaining traction in mental health care. Despite their growing usage, there is a scarcity of comprehensive evaluations of their impact on mental health and well-being. This systematic review and meta-analysis aims to fill this gap by synthesizing evidence on the effectiveness of AI-based CAs in improving mental health and factors influencing their effectiveness and user experience. Twelve databases were searched for experimental studies of AI-based CAs' effects on mental illnesses and psychological well-being published before May 26, 2023. Out of 7834 records, 35 eligible studies were identified for systematic review, out of which 15 randomized controlled trials were included for meta-analysis. The meta-analysis revealed that AI-based CAs significantly reduce symptoms of depression (Hedge's g 0.64 [95% CI 0.17-1.12]) and distress (Hedge's g 0.7 [95% CI 0.18-1.22]). These effects were more pronounced in CAs that are multimodal, generative AI-based, integrated with mobile/instant messaging apps, and targeting clinical/subclinical and elderly populations. However, CA-based interventions showed no significant improvement in overall psychological well-being (Hedge's g 0.32 [95% CI -0.13 to 0.78]). User experience with AI-based CAs was largely shaped by the quality of human-AI therapeutic relationships, content engagement, and effective communication. These findings underscore the potential of AI-based CAs in addressing mental health issues. Future research should investigate the underlying mechanisms of their effectiveness, assess long-term effects across various mental health outcomes, and evaluate the safe integration of large language models (LLMs) in mental health care.

Longitudinal associations of social support, everyday social interactions, and mental health during the COVID-19 pandemic.

Main effect models contend that perceived social support benefits mental health in the presence and the absence of stressful events, whereas stress-buffering models contend that perceived social support benefits mental health especially when individuals are facing stressful events. We tested these models of how perceived social support impacts mental health during the COVID-19 pandemic and evaluated whether characteristics of everyday social interactions statistically mediated this association - namely, (a) received support, the visible and deliberate assistance provided by others, and (b) pleasantness, the extent to which an interaction is positive, flows easily, and leads individuals to feel understood and validated. 591 United States adults completed a 3-week ecological momentary assessment protocol sampling characteristics of their everyday social interactions that was used to evaluate between-person average values and within-person daily fluctuations in everyday social interaction characteristics. Global measures of perceived social support and pandemic-related stressors were assessed at baseline. Psychiatric symptoms of depression and anxiety were assessed at baseline, at the end of each day of ecological momentary assessment, and at 3-week follow-up. Consistent with a main effect model, higher baseline perceived social support predicted decreases in psychiatric symptoms at 3-week follow-up (ß = -.09, p = .001). Contrary to a stress-buffering model, we did not find an interaction of pandemic-stressors × perceived social support. The main effect of perceived social support on mental health was mediated by the pleasantness of everyday social interactions, but not by received support in everyday social interactions. We found evidence for both main effects and stress-buffering effects of within-person fluctuations in interaction pleasantness on daily changes in mental health. Results suggest the importance of everyday social interaction characteristics, especially their pleasantness, in linking perceived social support and mental health.

Mental health during the COVID-19 pandemic: Impacts of disease, social isolation, and financial stressors.

BACKGROUND: Although research shows that the Covid-19 pandemic has led to declines in mental health, the existing research has not identified the pathways through which this decline happens. AIMS: The current study identifies the distinct pathways through which COVID-induced stressors (i.e., social distancing, disease risk, and financial stressors) trigger mental distress and examines the causal impact of these stressors on mental distress. METHODS: We combined evidence of objective pandemic-related stressors collected at the county level (e.g., lack of social contact, infection rates, and unemployment rates) with self-reported survey data from over 11.5 million adult respondents in the United States collected daily for eight months. We used mediation analysis to examine the extent to which the objective stressors influenced mental health by influencing individual respondents' behavior and fears. RESULTS: County-level, day-to-day social distancing predicted significantly greater mental distress, both directly and indirectly through its effects on individual social contacts, worries about getting ill, and concerns about finances. Economic hardships were indirectly linked to increased mental distress by elevating people's concerns about their household's finances. Disease threats were both directly linked to mental distress and indirectly through its effects on individual worries about getting ill. Although one might expect that social distancing from people outside the home would have a greater influence on people who live alone, sub-analyses based on household composition do not support this expectation. CONCLUSION: This research provides evidence consistent with the thesis that the COVID-19 pandemic harmed the mental well-being of adults in the United States and identifies specific stressors associated with the pandemic that are responsible for increasing mental distress.

The Channel Matters: Self-disclosure, Reciprocity and Social Support in Online Cancer Support Groups.

People with health concerns go to online health support groups to obtain help and advice. To do so, they frequently disclose personal details, many times in public. Although research in non-health settings suggests that people self-disclose less in public than in private, this pattern may not apply to health support groups where people want to get relevant help. Our work examines how the use of private and public channels influences members' self-disclosure in an online cancer support group, and how channels moderate the influence of self-disclosure on reciprocity and receiving support. By automatically measuring people's self-disclosure at scale, we found that members of cancer support groups revealed more negative self-disclosure in the public channels compared to the private channels. Although one's self-disclosure leads others to self-disclose and to provide support, these effects were generally stronger in the private channel. These channel effects probably occur because the public channels are the primary venue for support exchange, while the private channels are mainly used for follow-up conversations. We discuss theoretical and practical implications of our work.

Seekers, Providers, Welcomers, and Storytellers: Modeling Social Roles in Online Health Communities.

Participants in online communities often enact different roles when participating in their communities. For example, some in cancer support communities specialize in providing disease-related information or socializing new members. This work clusters the behavioral patterns of users of a cancer support community into specific functional roles. Based on a series of quantitative and qualitative evaluations, this research identified eleven roles that members occupy, such as welcomer and story sharer. We investigated role dynamics, including how roles change over members' lifecycles, and how roles predict long-term participation in the community. We found that members frequently change roles over their history, from ones that seek resources to ones offering help, while the distribution of roles is stable over the community's history. Adopting certain roles early on predicts members' continued participation in the community. Our methodology will be useful for facilitating better use of members' skills and interests in support of community-building efforts.

Scaling up analogical innovation with crowds and AI.

Analogy-the ability to find and apply deep structural patterns across domains-has been fundamental to human innovation in science and technology. Today there is a growing opportunity to accelerate innovation by moving analogy out of a single person's mind and distributing it across many information processors, both human and machine. Doing so has the potential to overcome cognitive fixation, scale to large idea repositories, and support complex problems with multiple constraints. Here we lay out a perspective on the future of scalable analogical innovation and first steps using crowds and artificial intelligence (AI) to augment creativity that quantitatively demonstrate the promise of the approach, as well as core challenges critical to realizing this vision.

Self-disclosure and Channel Difference in Online Health Support Groups.

Online health support groups are places for people to compare themselves with others and obtain informational and emotional support about their disease. To do so, they generally need to reveal private information about themselves and in many support sites, they can do this in public or private channels. However, we know little about how the publicness of the channels in health support groups influence the amount of self-disclosure people provide. Our work examines the extent members self-disclose in the private and public channels of an online cancer support group. We first built machine learning models to automatically identify the amount of positive and negative self-disclosure in messages exchanged in this community, with adequate validity (r>0.70). In contrast to findings from non-health-related sites, our results show that people generally self-disclose more in the public channel than the private one and are especially likely to reveal their negative thoughts and feelings publicly. We discuss theoretical and practical implications of our work.

Commitment of Newcomers and Old-timers to Online Health Support Communities.

For online communities to be successful, they must retain an adequate number of members who contribute to the community. The amount and type of communication members receive can play an important role in generating and sustaining members' commitment to it. However, the communication that members find valuable may change with their tenure in the community. This paper examines how the communication members receive in an health-support community influences their commitment and how this influence changes with their tenure in the community. Commitment was operationalized with three measures: self-reported attachment, continued participation in the community, and responding to others. Results show that receiving communication was generally associated with increased commitment across the three measures, with its impact increasing with members' tenure. However, the average amount of informational and emotional support members received per message was associated with decreased commitment. Results have implications for interventions to encourage members' commitment to their communities throughout their history in the community.

Eliciting and receiving online support: using computer-aided content analysis to examine the dynamics of online social support.

BACKGROUND: Although many people with serious diseases participate in online support communities, little research has investigated how participants elicit and provide social support on these sites. OBJECTIVE: The first goal was to propose and test a model of the dynamic process through which participants in online support communities elicit and provide emotional and informational support. The second was to demonstrate the value of computer coding of conversational data using machine learning techniques (1) by replicating results derived from human-coded data about how people elicit support and (2) by answering questions that are intractable with small samples of human-coded data, namely how exposure to different types of social support predicts continued participation in online support communities. The third was to provide a detailed description of these machine learning techniques to enable other researchers to perform large-scale data analysis in these communities. METHODS: Communication among approximately 90,000 registered users of an online cancer support community was analyzed. The corpus comprised 1,562,459 messages organized into 68,158 discussion threads. Amazon Mechanical Turk workers coded (1) 1000 thread-starting messages on 5 attributes (positive and negative emotional self-disclosure, positive and negative informational self-disclosure, questions) and (2) 1000 replies on emotional and informational support. Their judgments were used to train machine learning models that automatically estimated the amount of these 7 attributes in the messages. Across attributes, the average Pearson correlation between human-based judgments and computer-based judgments was .65. RESULTS: Part 1 used human-coded data to investigate relationships between (1) 4 kinds of self-disclosure and question asking in thread-starting posts and (2) the amount of emotional and informational support in the first reply. Self-disclosure about negative emotions (beta=.24, P<.001), negative events (beta=.25, P<.001), and positive events (beta=.10, P=.02) increased emotional support. However, asking questions depressed emotional support (beta=-.21, P<.001). In contrast, asking questions increased informational support (beta=.38, P<.001), whereas positive informational self-disclosure depressed it (beta=-.09, P=.003). Self-disclosure led to the perception of emotional needs, which elicited emotional support, whereas asking questions led to the perception of informational needs, which elicited informational support. Part 2 used machine-coded data to replicate these results. Part 3 analyzed the machine-coded data and showed that exposure to more emotional support predicted staying in the group longer 33% (hazard ratio=0.67, P<.001), whereas exposure to more informational support predicted leaving the group sooner (hazard ratio=1.05, P<.001). CONCLUSIONS: Self-disclosure is effective in eliciting emotional support, whereas question asking is effective in eliciting informational support. Moreover, perceptions that people desire particular kinds of support influence the support they receive. Finally, the type of support people receive affects the likelihood of their staying in or leaving the group. These results demonstrate the utility of machine learning methods for investigating the dynamics of social support exchange in online support communities.

Effectiveness of shared leadership in Wikipedia.

OBJECTIVE: The objective of the paper is to understand leadership in an online community, specifically, Wikipedia. BACKGROUND: Wikipedia successfully aggregates millions of volunteers' efforts to create the largest encyclopedia in human history. Without formal employment contracts and monetary incentives, one significant question for Wikipedia is how it organizes individual members with differing goals, experience, and commitment to achieve a collective outcome. Rather than focusing on the role of the small set of people occupying a core leadership position, we propose a shared leadership model to explain the leadership in Wikipedia. Members mutually influence one another by exercising leadership behaviors, including rewarding, regulating, directing, and socializing one another. METHOD: We conducted a two-phase study to investigate how distinct types of leadership behaviors (transactional, aversive, directive, and person-focused), the legitimacy of the people who deliver the leadership, and the experience of the people who receive the leadership influence the effectiveness of shared leadership in Wikipedia. RESULTS: Our results highlight the importance of shared leadership in Wikipedia and identify trade-offs in the effectiveness of different types of leadership behaviors. Aversive and directive leadership increased contribution to the focal task, whereas transactional and person-focused leadership increased general motivation. We also found important differences in how newcomers and experienced members responded to leadership behaviors from peers. APPLICATION: These findings extend shared leadership theories, contribute new insight into the important underlying mechanisms in Wikipedia, and have implications for practitioners who wish to design more effective and successful online communities.

Effects of internet use on health and depression: a longitudinal study.

BACKGROUND: The rapid expansion of the Internet has increased the ease with which the public can obtain medical information. Most research on the utility of the Internet for health purposes has evaluated the quality of the information itself or examined its impact on clinical populations. Little is known about the consequences of its use by the general population. OBJECTIVE: Is use of the Internet by the general population for health purposes associated with a subsequent change in psychological well-being and health? Are the effects different for healthy versus ill individuals? Does the impact of using the Internet for health purposes differ from the impact of other types of Internet use? METHODS: Data come from a national US panel survey of 740 individuals conducted from 2000 to 2002. Across three surveys, respondents described their use of the Internet for different purposes, indicated whether they had any of 13 serious illnesses (or were taking care of someone with a serious illness), and reported their depression. In the initial and final surveys they also reported on their physical health. Lagged dependent variable regression analysis was used to predict changes in depression and general health reported on a later survey from frequency of different types of Internet use at an earlier period, holding constant prior depression and general health, respectively. Statistical interactions tested whether uses of the Internet predicted depression and general health differently for people who initially differed on their general health, chronic illness, and caregiver status. RESULTS: Health-related Internet use was associated with small but reliable increases in depression (ie, increasing use of the Internet for health purposes from 3 to 5 days per week to once a day was associated with .11 standard deviations more symptoms of depression, P = .002). In contrast, using the Internet for communication with friends and family was associated with small but reliable decreases in depression (ie, increasing use of the Internet for communication with friends and family purposes from 3 to 5 days per week to once a day was associated with .07 standard deviations fewer symptoms of depression, P = .007). There were no significant effects of respondents' initial health status (P = .234) or role as a caregiver (P = .911) on the association between health-related Internet use and depression. Neither type of use was associated with changes in general health (P = .705 for social uses and P = .494 for health uses). CONCLUSIONS: Using the Internet for health purposes was associated with increased depression. The increase may be due to increased rumination, unnecessary alarm, or over-attention to health problems. Additionally, those with unmeasured problems or those more prone to health anxiety may self-select online health resources. In contrast, using the Internet to communicate with friends and family was associated with declines in depression. This finding is comparable to other studies showing that social support is beneficial for well-being and lends support to the idea that the Internet is a way to strengthen and maintain social ties.

Psychological research online: report of Board of Scientific Affairs' Advisory Group on the Conduct of Research on the Internet.

As the Internet has changed communication, commerce, and the distribution of information, so too it is changing psychological research. Psychologists can observe new or rare phenomena online and can do research on traditional psychological topics more efficiently, enabling them to expand the scale and scope of their research. Yet these opportunities entail risk both to research quality and to human subjects. Internet research is inherently no more risky than traditional observational, survey, or experimental methods. Yet the risks and safeguards against them will differ from those characterizing traditional research and will themselves change over time. This article describes some benefits and challenges of conducting psychological research via the Internet and offers recommendations to both researchers and institutional review boards for dealing with them. ((c) 2004 APA, all rights reserved)

Visual copresence and conversational coordination.

Pickering & Garrod's (P&G's) theory of dialogue production cannot completely explain recent data showing that when interactants in referential communication tasks have different views of a physical space, they accommodate their language to their partner's view rather than mimicking their partner's expressions. Instead, these data are consistent with the hypothesis that interactants are taking the perspective of their conversational partners.