Artificial Intelligence Discontinuation Effects (AI-DICE): An Emerging Phenomenon in Mental Health Applications

Abstract

Artificial intelligence (AI) has emerged as a powerful tool for fostering positive behavior change and enhancing mental health support. However, the abrupt discontinuation or functional degradation of AI-driven interventions, particularly those featuring conversational agents, may trigger unintended psychological consequences. Therefore, we introduce and examine the concept of Artificial Intelligence Discontinuation Effects (AI-DICE), drawing parallels from abandonment-like experiences observed from problematic termination experiences with therapists. We propose a conceptual framework for AI-DICE mitigation that draws on evidence-based behavior change principles and explores clinical modalities that may inform mitigation toolkits (eg, Acceptance and Commitment Therapy, Cognitive Behavioral Therapy, Dialectical Behavior Therapy, and Motivational Interviewing). We also ground our approach in user experience research and community-engaged research. AI-DICE raises critical ethical challenges, including transparency, the ability to withdraw or adapt participation as the users’ knowledge of the intervention grows, and access to support postintervention. Prioritizing long-term continuation, or at least some form of ongoing access, over the best-planned complete discontinuation strategy may help ensure that AI-driven mental health solutions deliver lasting benefits rather than unintended harm. Finally, although not yet empirically established, incorporating discontinuation planning and AI-DICE mitigation from the outset may also improve intervention effectiveness by strengthening user autonomy, supporting skills transfer, and reducing dependence-related vulnerabilities.