Abstract
This paper outlines a proposal for an international workshop designed to bridge the significant gap between Artificial Intelligence (AI) and health psychology. While AI has demonstrated remarkable success in domains like computer vision and gaming, its potential to facilitate long-term health behavior change remains largely untapped. The authors argue that this stagnation is due to a lack of integration between three critical research tracks: adaptive intervention design, cognitive modeling, and social robotics. By synthesizing these fields, the workshop aims to develop more effective, personalized, and engaging digital health interventions.
Introduction: The "Human-Hard" Challenge
Digital health interventions have been a subject of research for over twenty years, yet they consistently face two major hurdles: a lack of sustained effectiveness in real-world settings and limited commercial success. A fundamental irony exists in current AI development: AI excels at tasks that humans find intuitive, such as object recognition, but struggles with health behavior change—a task that humans themselves find notoriously difficult.
Currently, research in this area is fragmented into three isolated silos:
- Just-In-Time Adaptive Interventions (JITAI): Systems focused on identifying the optimal moment to deliver an intervention.
- User Modeling: The creation of computational simulations to understand and predict habit formation.
- Social AI: The use of chatbots or social robots to provide the "human touch" often missing from mobile applications.
The authors contend that significant breakthroughs will only occur when these fields are integrated—for example, by using cognitive models to train more sophisticated intervention algorithms.
Methodological Frameworks for Synthesis
The proposed workshop focuses on several core research paradigms that, when combined, could revolutionize health behavior change:
- Reinforcement Learning (RL) for JITAI: Moving beyond static, rule-based systems toward algorithms that treat user compliance as a "reward" signal. This allows the system to learn dynamically when a user is most receptive to a "nudge."
- Psychological Computing: A framework for building cognitive models (e.g., habit formation models) that enable digital systems to track a user’s internal psychological progress rather than just external metrics like step counts.
- Social Agency and Control: Leveraging social robots and chatbots to foster emotional bonds with users. This addresses the lack of social influence in traditional digital tools, which is often a primary reason for user disengagement.
- Micro-Randomized Trials (MRTs): A statistical methodology essential for developing and validating the intelligent algorithms that power adaptive interventions.
Key Insights and Strategic Ideas
The workshop proposal highlights several critical insights for the future of the field:
Cross-Pollination as a Catalyst
A major "aha moment" in the proposal is the suggestion that cognitive models of human behavior can serve as "simulators" to train RL agents. This "warm-start" approach could significantly improve the effectiveness of AI interventions from the moment of deployment, bypassing the initial data-gathering phase that often leads to user drop-off.
The Role of Social Presence
The authors suggest that the "social gap" in digital tools is a primary driver of their failure. Integrating Social AI is not merely a cosmetic addition; it is a strategic method to provide the social control and emotional support necessary for sustained behavior change.
Identifying Boundary Conditions
The workshop aims to identify which specific health behaviors or demographic groups are most susceptible to AI-driven change. Acknowledging that AI is not a "silver bullet," the goal is to define the specific contexts where these technologies can be most effective.
Expected Outcomes and Future Directions
As a preface and proposal, this work does not present quantitative experimental results but rather defines a roadmap for the field. The expected outcomes include:
- The identification of "boundary conditions" for AI application in health.
- The formation of new multidisciplinary collaborations between computer scientists and psychologists.
- The development of a comprehensive perspective paper to define future research priorities.
Limitations and Challenges
Despite the potential, several challenges remain. Most current interventions lose effectiveness once the "novelty" wears off. Furthermore, there is a significant scarcity of high-quality, long-term behavioral data required to train robust AI models. Finally, the psychological and ethical implications of using AI to "persuade" or "control" human behavior require rigorous investigation. The primary goal for future work is to bridge the current isolation of research teams to achieve fruitful outcomes at the intersection of social AI, machine learning, and cognitive science.
Summary updated