

Can we trust AI in health?
7 snips Sep 23, 2025
In this engaging discussion, Professor Marzia Ghassimi from MIT shares her expertise on AI in healthcare, highlighting challenges like demographic bias and model vulnerabilities. She emphasizes the need for stronger regulations to ensure the safety of AI applications. Marzia reveals how small changes in input can drastically alter AI recommendations, raising safety concerns. She also uncovers AI's ability to infer sensitive demographic information from medical images, prompting a cautionary message about potential biases in health outcomes.
AI Snips
Chapters
Transcript
Episode notes
Deploy Simple AI Screening Widely
- Use AI stethoscopes to screen common heart conditions when echocardiograms are unavailable.
- Deploy them with a smartphone and basic connectivity to reach low-resource settings quickly.
Verify AI Health Answers With Sources
- Check sources behind AI health answers and follow the original trusted references.
- If uncertain, consult regulated resources like NHS-approved guidance rather than trusting raw AI outputs.
Treat Wearables Appropriately
- Distinguish wellness wearables from regulated clinical devices before relying on results.
- For clinical decisions, prefer devices with regulatory approval and seek medical advice on flagged issues.