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

| Podcasts | May 20, 2026 | 481 views | 50:08

TL;DR

This seminar explores how personalized foundation models trained on wearable data can predict intervenable health events (like cravings or blood pressure spikes) to enable just-in-time digital interventions, but emphasizes that HCI is critical to solve adherence challenges, inconsistent labeling, and misaligned evaluation metrics.

🧠 Personalized AI Architecture 3 insights

Self-supervised learning on biosignals

Models predict missing portions of bio-signals from surrounding data, creating personalized foundation models per user without requiring extensive labeled datasets.

One model per person paradigm

Unlike traditional diagnostic AI using a single general model, this approach trains separate models on individual longitudinal data to predict intervenable events like substance cravings or blood pressure spikes.

Multimodal physiological relationships

Pre-training learns how signals like heart rate variability and accelerometry relate to each other within an individual's specific physiology to enable downstream predictions.

⚠️ Real-World Deployment Failures 3 insights

Nurse study adherence breakdown

During COVID-19, nurses showed inconsistent wearable adherence and highly variable stress labeling patterns both across individuals and within the same person over time, rendering personalized models ineffective.

AI's behavioral limitation

Personalization fails when users provide inconsistent ground-truth labels, proving that algorithmic innovation alone cannot overcome fundamental human behavioral complexities.

Actionable over precise predictions

Clinical utility comes from detecting intervenable events like blood pressure spikes rather than predicting exact numerical values that would not change the clinical action taken.

🎯 HCI Solutions for Burden and Evaluation 3 insights

Intelligent burden reduction strategies

Combining active learning with passive sensing detects optimal moments for user queries while automatically avoiding interruptions during activities like driving or school pickup.

Prevalence-sensitive metric awareness

Standard metrics like precision shift dramatically with population prevalence, while sensitivity and specificity provide stable evaluation across different clinical deployment contexts.

Causal adherence modeling

Intervention timing and content create causal chains affecting user burden, receptivity, and adherence, requiring HCI study designs that measure human behavior beyond algorithmic accuracy.

Bottom Line

Effective digital health AI requires pairing personalized foundation models with HCI methods that explicitly address adherence challenges, ensure consistent human labeling, and employ clinically relevant evaluation metrics to deliver interventions at the right moment without excessive user burden.

More from Stanford Online

View all
Stanford MS&E435 | Spring 2026 | Economics of Generative AI
34:13
Stanford Online Stanford Online

Stanford MS&E435 | Spring 2026 | Economics of Generative AI

Stanford instructor Apur frames generative AI as a supercycle with inverted economics where semiconductor and infrastructure costs dominate revenues while application-layer value remains elusive, questioning whether this structure represents a temporary capex cycle or a new permanent equilibrium.

2 days ago · 7 points
Stanford Robotics Seminar ENGR319 | Spring 2026 | Interactive Autonomy
1:11:12
Stanford Online Stanford Online

Stanford Robotics Seminar ENGR319 | Spring 2026 | Interactive Autonomy

UC Berkeley's Icon Lab presents game-theoretic frameworks enabling robots to safely interact with humans and other agents by modeling joint prediction as potential games, reducing computational costs by 20x while solving the challenge of multiple social equilibria in real-time navigation.

2 days ago · 8 points