Why Is Medical AI So Hard to Trust?
Artificial intelligence is transforming healthcare, but building AI that clinicians can truly trust is far more challenging than achieving high accuracy.
In the latest episode of Twin Things Podcast, Katarzyna Baliga-Nicholson talks with Jan Fiszer from Sano’s Extreme-scale Data and Computing team about the challenges of applying AI to medical imaging, MRI and radiology.
The conversation explores what happens between raw medical data and a clinically useful AI system. Medical data is complex and often noisy, while differences in scanners, protocols, image quality and patient populations can significantly affect how AI models perform.
That is why accuracy alone is not enough. Medical AI also needs to be validated across different settings, its limitations need to be understood, and its results need to make sense within real clinical workflows.
The episode also looks at explainable AI and synthetic data, and at the broader question of how we can turn increasingly complex healthcare data into tools that support — rather than complicate — clinical decision-making.
Ultimately, trustworthy medical AI is not just about building better models. It is about understanding the entire journey from data to decision.
Listen to the full episode
🎙️ Twin Things Podcast #9
Katarzyna Baliga-Nicholson × Jan Fiszer
