The impact of "relational" Artificial Intelligence on human well-being: A self-determination theory analysis.

Abstract

Advances in generative artificial intelligence (AI) have given rise to relational AI-AI agents that mimic human relational capabilities while possessing unique nonhuman features, including constant availability, extensive malleability, and a lack of intrinsic psychological needs. This article uses self-determination theory to examine how relational AI may support or undermine the three basic psychological needs self-determination theory posits as essential for well-being: relatedness, competence, and autonomy. Across roles including that of tutor, copilot, social mediator, companion, and therapist, relational AI may address critical challenges to these needs, facilitating goal attainment, alleviating loneliness, and promoting mental health. However, relational AI also carries potential risks for well-being, including reduced self-direction, diminished efficacy, and altered expectations within human relationships. We propose that relational AI's ultimate impact on users' well-being will depend on moderating factors that may shape both the strength and direction of its effects on well-being, including users' motivational orientation toward the goal for which they use relational AI, their motivational orientation for using relational AI, their baseline psychological need satisfaction, dynamics particular to each relational AI role, and various situational factors. Crucially, it is likely impossible to fully understand relational AI's impacts on well-being without examining all three psychological needs, as focusing on only a single need risks overlooking dynamics in which impacts on multiple needs interact to amplify or neutralize well-being. Future research on these processes can deepen our understanding of human relationship dynamics, inform responsible AI development, and reveal novel theoretical mechanisms underlying relational AI's impacts on well-being. (PsycInfo Database Record (c) 2026 APA, all rights reserved).