From Prediction to Interaction: A Behavioral Informatics Framework for AI in Preventive Health.
- 1Bioinformatics Section, Department of Computer Science, Vrije Universiteit Amsterdam, De Boelelaan 1081A, Amsterdam, 1081 HV, The Netherlands, 0031 6 34 64 99 52.
Abstract
Unlabelled
In this viewpoint, we present our perspective-informed by published evidence and by our own work on AI-driven digital health systems-on why AI has had limited real-world impact in preventive health and how it should be redesigned. Most AI systems in health care are built as episodic prediction tools, whereas preventive care depends on sustained behavioral change; the consequences are visible in consistently high attrition and low long-term adherence across digital health interventions. The aim of this viewpoint is to argue that this failure is structural rather than incremental and to propose an integrated, interaction-centric alternative. We advance 3 key messages. First, the limited impact of AI in preventive health reflects a structural misalignment between episodic, prediction-centric system design and the longitudinal dynamics of health behavior. Second, addressing this misalignment requires treating continuous interaction as the primary unit of design, which we articulate through a behavioral informatics framework integrating 6 interdependent layers: data, knowledge, reasoning, interaction, behavior, and governance. The contribution of this framework is explicitly integrative: it synthesizes previously separate strands, including just-in-time adaptive interventions, behavioral intervention technologies, and human-AI interaction research, into a single system-level account. Third, evaluation must shift from predictive accuracy and short-term engagement toward measurable behavioral trajectories and longitudinal outcomes, with explicit attention to feasibility and to conditions under which simpler, non-AI interventions remain preferable. This viewpoint is addressed to digital health researchers and AI developers, behavioral scientists, clinicians involved in preventive care, and policymakers responsible for governing continuously interacting AI systems.
Related
- Hybrid Explainable Deep Ensemble Framework with self-attention mechanism for early prediction of depression severity using PHQ-9 clinical and behavioral data
- PRInter: an inductive multimodal heterogeneous network embedding framework with contrastive learning for lncRNA-protein interaction prediction
- MDM-DDI: a drug–drug interaction prediction framework integrating multi-dimensional molecular features and memory-augmented networks
- Foundation model-driven framework for human–object interaction prediction with segmentation mask integration
- HAFI-DTA: an interpretable multimodal framework for drug-target affinity prediction based on hierarchical attention guidance and feature iterative interaction
BibTeX
@article{gavai2026from,
title = {{From Prediction to Interaction: A Behavioral Informatics Framework for AI in Preventive Health.}},
author = {Anand Gavai and Jaap Heringa},
journal = {Journal of medical Internet research},
year = {2026},
month = oct,
doi = {10.2196/98961},
url = {https://europepmc.org/article/MED/42950341},
note = {CC BY},
}
RIS
TY - JOUR
TI - From Prediction to Interaction: A Behavioral Informatics Framework for AI in Preventive Health.
AU - Anand Gavai
AU - Jaap Heringa
PY - 2026
DA - 2026/10/01
JO - Journal of medical Internet research
DO - 10.2196/98961
UR - https://europepmc.org/article/MED/42950341
AB - Unlabelled In this viewpoint, we present our perspective-informed by published evidence and by our own work on AI-driven digital health systems-on why AI has had limited real-world impact in preventive health and how it should be redesigned. Most AI systems in health care are built as episodic prediction tools, whereas preventive care depends on sustained behavioral change; the consequences are visible in consistently high attrition and low long-term adherence across digital health interventions. The aim of this viewpoint is to argue that this failure is structural rather than incremental and to propose an integrated, interaction-centric alternative. We advance 3 key messages. First, the limited impact of AI in preventive health reflects a structural misalignment between episodic, prediction-centric system design and the longitudinal dynamics of health behavior. Second, addressing this misalignment requires treating continuous interaction as the primary unit of design, which we articulate through a behavioral informatics framework integrating 6 interdependent layers: data, knowledge, reasoning, interaction, behavior, and governance. The contribution of this framework is explicitly integrative: it synthesizes previously separate strands, including just-in-time adaptive interventions, behavioral intervention technologies, and human-AI interaction research, into a single system-level account. Third, evaluation must shift from predictive accuracy and short-term engagement toward measurable behavioral trajectories and longitudinal outcomes, with explicit attention to feasibility and to conditions under which simpler, non-AI interventions remain preferable. This viewpoint is addressed to digital health researchers and AI developers, behavioral scientists, clinicians involved in preventive care, and policymakers responsible for governing continuously interacting AI systems.
SN - 1439-4456
SN - 1438-8871
ER -