The FDA ’s new A.I. action plan just got published. Here are the main points: 1. Tailored Regulatory Framework My take: As the draft guidance about this with details will come out this year, I hope adaptive algorithms will be covered. 2. Good Machine Learning Practice My take: The FDA wants to harmonize such practices across a range of international organizations. I'm not sure it’s possible, since their detailed plan is still not ready. 3. Patient-Centered Approach My take: I would’ve expected the digital health arm of the FDA to appoint a patient engagement advisory board, as promoting a patient centric approach without actively involving patients sounds like a risky business. 4. Regulatory Science Methods Related to Algorithm Bias & Robustness My take: This is why we expected that the FDA would take over our open access database of A.I.-based medical devices that we published in Nature Publishing Group's npj Digital Medicine paper. However, they confirmed to me that they won’t. 5. Real-World Performance My take: The FDA should promote the importance of prospective studies when it comes to analyzing algorithms' performance in medical tasks. …see more