Lessons from an evolving regulatory framework: AI/ML Based Devices
Artificial Intelligence and Machine Learning based methods are adaptive by nature. Their evolving character makes them powerful, but poses a challenge for quality and regulation. The post below describes the type of modification that can be made to AI/ML based software as a medical device after formal regulatory approval.
The pre-submission steps that draw out a space for acceptable changes and methods to implement and test them. In addition, we discuss the responsibility of following good machine learning practices (GMLP) that fall on the developers to afford the flexibility of making changes to their AI systems.
Great article, Nishant - very well written! With the abundance of ML/AI applications out there, it will be quite the challenge to generate regulatory requirements that are even remotely similar across everything from tumor detection to self driving automobiles. I'm particularly concerned because the ML/AI algorithms are generally the "secret sauce" of start-up companies that will (legitimately) be afraid to share IP and may not work with those regulatory bodies - leaving the big tech companies as the only ones with inputs. So some encouragement to include small business in those regulatory discussions is critical.
Medium article on evolving regulatory framework for AI/ML devices: https://tinyurl.com/aimldevices