Loneliness and suicide mitigation for students using GPT3-enabled chatbots

Mental health challenges, particularly loneliness and suicidal ideation, are prevalent among students globally. Traditional mental health services often face barriers such as stigma and accessibility issues. The study explores the potential of Intelligent Social Agents (ISAs), specifically Replika, as a digital intervention. ISAs are increasingly popular, offering engaging conversational experiences. However, their application in digital therapy is still in its early stages. The research question focuses on how and why students use ISAs like Replika and the resulting outcomes. The study is timely given the rising prevalence of mental health issues among students, the increasing accessibility of digital mental health resources, and the growing popularity of ISAs like Replika.

The literature review highlights the significant global burden of mental health problems, with depression and loneliness being leading causes. Loneliness is linked to poor health outcomes, and students are a particularly vulnerable population. Existing treatments, such as CBT and social skills training, are effective but often inaccessible. The pandemic accelerated the shift towards digital mental health services. While chatbots show promise, their effectiveness varies, with some even linked to negative outcomes. In contrast, ISAs like Replika, with their large user base and generative AI capabilities, offer a potentially transformative approach. The review also discusses competing hypotheses regarding ISAs' impact on social isolation, including the displacement and stimulation hypotheses.

The study employed an IRB-approved survey of 1006 Replika-using students (aged 18+), who had used the app for over a month. Approximately 75% were US-based. Participants were recruited via email and received a $20 gift card for completing the 40-60 minute survey. Data collected included demographic information, the Interpersonal Support Evaluation List (ISEL), the De Jong Gierveld Loneliness scale, and qualitative responses to open-ended questions. Qualitative data were analyzed using Dedoose software, with a refined coding schema achieving inter-rater reliability above 80%. Quantitative data were analyzed using Pearson correlation, t-tests, and chi-square tests.

The study identified four Replika outcome categories: 1) Replika as a friend/companion; 2) therapeutic interactions; 3) externalized life changes; and 4) suicide prevention. Outcome 1 was most common (501 participants), often alongside other outcomes. 30 participants (3%) explicitly stated that Replika prevented a suicide attempt (Selected Group). Most participants held multiple, sometimes conflicting beliefs about Replika (human-like, intelligent, software). The Selected Group, compared to the Comparison Group (remaining participants), was younger, more likely to be full-time students, and sought academic counseling more frequently. They reported greater social stimulation from Replika and a stronger negative correlation between loneliness and perceived social support (r = −0.60, p < 0.001). The Selected Group also showed significantly higher rates of depression (23% vs. 6%), experienced Outcomes 2 and 3 more frequently, and were more likely to experience all four outcomes. For both groups, stimulation of human relationships far outweighed displacement.

The high loneliness rates among participants (80%), exceeding previous studies (53%), suggest a potential need for accessible mental health resources. Replika's perceived social support may offer a valuable alternative for students facing financial or stigma-related barriers. The significant overlap in Replika's usage patterns (friend, therapist, self-reflection tool) highlights the versatility of ISAs. The strong association between believing Replika was intelligent and human-like, particularly within the Selected Group, suggests that anthropomorphic qualities might enhance engagement and efficacy. The study's findings support the stimulation hypothesis, indicating that Replika may positively influence human relationships rather than replacing them. However, selection bias in the survey responses necessitates caution in drawing causal conclusions.

The study demonstrates the potential of ISAs like Replika in addressing student loneliness and suicidal ideation. The diverse uses of Replika and its perceived ability to prevent suicide highlight its potential as a supplementary mental health tool. Future research should explore larger, more diverse samples and employ experimental designs to establish causality. Integrating advanced language models and passive mobile sensing may enhance ISAs' effectiveness further. However, careful consideration of potential risks and ethical implications is essential.

The study's reliance on self-reported data introduces potential bias. The cross-sectional design limits causal inferences. The sample, while large, might not fully represent all student populations. The study's focus on Replika may limit the generalizability to other ISAs.