Outsourcing the Self: How Using AI Words Changes What—and How—We Self-Disclose

Wed 21.05 10:30 - 11:30

ABSTRACT

Using language to reveal ourselves is a process called self-disclosure. Self-disclosure and specifically written disclosure are important because they have been found to improve psychological and physiological well-being.  Large Language Model (LLM)-based conversational agents are integrated into writing applications and influence what individuals write. Thus, LLMs might affect self-disclosure processes. The current research tested the effects of using AI while writing on the self-disclosure process, including its effectiveness and implications on well-being. We hypothesized that using AI's words while self-disclosing would result in less personal texts than when consulting with AI but not using its words or having no AI access (H1). Following previous research that found that texts written with AI are more positive than texts written without AI involvement, we hypothesized that using AI's words while self-disclosing would result in more positive texts than when only consulting with AI and not using its words or having no AI access (H2). Finally, we hypothesized that texts' personal and positivity levels would mediate the effects of types of AI usage on well-being (H3). Specifically, we hypothesized that texts including AI's words would be more positive and less personal, leading to contradicting effects on well-being. Across two studies, participants self-disclosed in writing about three (Study 1) or one (Study 2) past immoral behaviors. Results showed that, as expected, texts written with AI's words while self-disclosing were less personal and more positive than texts written while consulting with AI or having no AI access. Also, as expected, texts' personal and positivity levels mediated the effects of types of AI usage on well-being. This research highlights the consequences of the type of AI usage while self-disclosing. Using AI's words but not consulting with AI resulted in less personal and more positive texts in comparison to having no AI access.

Speaker

Atil Mansour

Technion

  • Advisors Liat Levontin

  • Academic Degree M.Sc.