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

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 (arxiv.org/abs/2605.22759v1)
Google Fitbit Air, das Fitness - Tracker-Armband (store.google.com/de/product/google_fitbit_air)
Google Health (healthapp.google)
Foundation model (en.wikipedia.org/wiki/Foundation_model)
Embedding (machine learning) (en.wikipedia.org/wiki/Embedding_(machine_learn…))
Generative AI (en.wikipedia.org/wiki/Generative_AI)
Supervised learning (en.wikipedia.org/wiki/Supervised_learning)
Few-shot learning (en.wikipedia.org/wiki/Few-shot_learning)
Wearables with hypertension features: Huawei D2, Apple Watch, Whoop, Oura
Nudging (en.wikipedia.org/wiki/Nudge_theory)
Dimensionality reduction (en.wikipedia.org/wiki/Dimensionality_reduction)
Prince Charles’ “middle finger” dimensionality reduction example (reddit.com/r/interestingasfuck/comments/xu1270…)
Sam Altman tweet: “there is no wall” (x.com/sama/status/1856941766915641580)
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 (docs.google.com/document/d/1LCIm780Aue5573FWVp…)

🎙️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 (creators.spotify.com/pod/profile/fitforscience) and subscribe on your favorite platforms:
YouTube (youtube.com/@FitForScience)
Spotify (open.spotify.com/show/56TjUxuMsPETb0kGEJ7nwf)
Apple Podcasts (podcasts.apple.com/us/podcast/fit-for-science/…)
Amazon Music (music.amazon.de/podcasts/c3e54ee7-4a2c-442e-a5…)
Collection of all show notes (docs.google.com/document/d/1LCIm780Aue5573FWVp…)

⚠️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.