Saltar al contenido

#20 Google Wants Your Fitbit Data for AI! And it’s not (all) bad news!

1 de julio de 2026 · 58 min

Notas del episodio

Rob and Stephan dissect Google’s groundbreaking "Sensor FM" paper, exploring how a foundation model, trained on trillions of wearable data minutes, could revolutionize preventive healthcare and disease prediction.


📝Summary

In this episode, biological data scientists Rob and Stephan break down Google's latest research introducing Sensor FM, a massive foundation model trained on over one trillion minutes of multimodal wearable data across 5 million users. They explore the technical mechanics under the hood, including its encoder-decoder architecture, its alignment with deep learning scaling laws, and a unique "AI classroom" setup where collaborating virtual agents optimize 35 distinct clinical health prediction tasks. The hosts discuss the profound implications of using generative AI like Gemini to transform these raw sensor representations into clinician-approved lifestyle recommendations, while critically evaluating the delicate trade-offs between continuous behavioral nudging, hidden human blind spots, and systemic data privacy risks. Finally, Stephan shares his upcoming personal testing protocol for the new Fitbit Air as a potential alternative to his long-term Oura ring setup.


⏳Chapters

00:00:00 Wearable Data Predictions: Google’s motivation for tracking disease and lifestyle indicators

00:01:48 Google Health Platform: The strategic endgame of consolidating consumer health tracking

00:04:00 The Interface Layer: Transforming raw hardware signals into actionable health insights

00:07:13 Introducing Sensor FM: A deep learning foundation model for wearable health data

00:10:15 Overcoming Data Labels: High phenotypic diversity and few-shot learning

00:12:37 Data Privacy vs. Preventive Care: Evaluating the societal trade-offs of deep tracking

00:17:18 Encoder-Decoder Architecture: The science of signal compression and reconstruction

00:18:43 Trillion-Minute Dataset: Mapping 5 million global wearable users

00:22:27 Empirical Scaling Laws: Maximizing compute and parameters for improved performance

00:27:19 The AI Classroom Experiment: Peer collaboration mechanisms among virtual agents

00:39:07 Gemini Judged by Clinicians: Blind evaluation of AI-generated health recommendations

00:44:57 Behavioral Nudging and Alignment: Addressing the blind spots of metric-driven optimization

00:50:42 Insurance Risk Paradigms: The dangers of continuous data history in for-profit healthcare

00:53:26 Fitbit Air vs. Oura Ring: Stephan’s upcoming personal logging experiment


📚Resources

SensorFM from Google

SenseFM: Towards a General Intelligence and Interface for Wearable Health Data

Google Fitbit Air, das Fitness - Tracker-Armband

Google Health

Foundation model

Embedding (machine learning)

Generative AI

Supervised learning

Few-shot learning

Wearables with hypertension features: Huawei D2, Apple Watch, Whoop, Oura

Nudging

Dimensionality reduction

Prince Charles’ “middle finger” dimensionality reduction example

Sam Altman tweet: “there is no wall”

Correction: The Netherlands is divided into 12 provinces and 3 special overseas municipalities

"Life can only be understood backwards; but it must be lived forwards." - Søren Kierkegaard

…There is more: complete show notes here


🎙️About

Fit For Science is a deep-dive podcast hosted by two biological data scientists, Rob and Stephan, exploring the intersection of research, health tech, and data-driven lifestyle design. The hosts provide evidence-based systems, layered with practical "N=2" personal experimentation, to cut through the noise and enable everyone to become their best N-of-1.

Learn more and subscribe on your favorite platforms:

YouTube

Spotify

Apple Podcasts

Amazon Music

Collection of all show notes


⚠️Disclaimer: This podcast represents our own opinions and is for informational purposes only. It does not constitute medical or financial advice or a professional relationship.

Todos los episodios de Fit For Science

Ajustes

Visualización

Idioma

Determina qué emisoras ves primero en la página de inicio.

Historial

Las emisoras y episodios escuchados recientemente y tus últimas búsquedas solo se guardan en este navegador.

Avisos

Un breve aviso cuando un pódcast que sigues tiene un episodio nuevo. Puedes desactivar pódcast concretos con la campana de su página.

Cuenta

Guarda tus favoritos: en la app y en la web.

Atajos de teclado

/
Abrir búsqueda
Espacio
Reproducir o pausar
← →
En el reproductor: 15 s atrás, 30 s adelante (episodios) · en una fila: siguiente tarjeta
↑ ↓
En una fila: destino dentro de la tarjeta
Inicio / Fin
En una fila: primera o última tarjeta
Esc
Cerrar ventana o menú
?
Este resumen
Reproduciendo ahora

Ir a la emisora

A continuación

    Añade episodios con «A continuación»: sonarán después del episodio actual.