Stanford CS547 HCI Seminar | Spring 2026 | HCI and Human-Centered AI for Digital Health

Why It Matters

Effective, personalized AI can deliver timely health interventions, but only if HCI strategies ensure user compliance and meaningful evaluation, shaping the future of scalable digital therapeutics.

Key Takeaways

• Personalized AI models predict health events from wearable data.
• Self‑supervised learning reduces label dependence for individual models.
• Intervention timing hinges on accurate, just‑in‑time predictions for patients.
• User adherence challenges undermine AI performance in real‑world settings.
• HCI design must balance model accuracy with user burden.

Summary

The seminar introduced a human‑centered AI approach for digital health, emphasizing personalized machine‑learning models built on multimodal wearable streams. Rather than a single, population‑wide diagnostic model, each user receives an AI that learns from their own biosignals to predict repeat health events—such as substance‑use cravings, blood‑pressure spikes, or stress episodes—and trigger just‑in‑time digital interventions. Key technical insights include leveraging self‑supervised, foundation‑model‑style training on unlabeled wearable data, then fine‑tuning with minimal labeled events. This personalized, self‑supervised pipeline dramatically outperforms generic models on public stress datasets and requires only a handful of calibration points. The speaker highlighted concrete studies on vaping‑related cravings and blood‑pressure spikes, noting that predicting a spike (e.g., >155/102) is clinically actionable even without precise numeric forecasts. Challenges surfaced when real‑world adherence faltered: a COVID‑era nurse study suffered from inconsistent self‑reporting and low wearable compliance, rendering personalization ineffective. The talk underscored the need for HCI solutions—active learning to query users only when model confidence is low, layered context sensing, and careful metric selection beyond standard precision/recall—to reduce user burden while preserving model efficacy. Overall, the work illustrates that AI breakthroughs alone won’t deliver digital therapeutics; integrating thoughtful HCI design to manage labeling, adherence, and evaluation is essential for scalable, patient‑centric interventions.

Original Description

For more information about Stanford’s graduate programs, visit: https://online.stanford.edu/graduate-education

May 8, 2026

This lecture covers:

• HCI issues related to the design, evaluation, and deployment of AI-enabled systems in digital health

• How "easy to implement" AI engineering decisions that are usually default parameters in Python machine learning code can have dramatic effects on the user experience of patients and clinicians

• How the choice of evaluation metrics to optimize for can affect downstream user experience

To follow along with the seminar schedule, visit: https://hci.stanford.edu/

Dr. Peter Washington, PI of the UCSF TECH Lab (techlab.ucsf.edu), is an Assistant Professor in the Division of Clinical Informatics and Digital Transformation (DoC-IT) in the Department of Medicine at the University of California, San Francisco (UCSF).