PhilSci Archive

Algorithms on Regulatory Lockdown in Medicine

Babic, Boris and Gerke, Sara and Evgeniou, Theodoros and Cohen, Glenn (2019) Algorithms on Regulatory Lockdown in Medicine. Science.

[img]
Preview
Text
science_manuscript_revised.docx.pdf

Download (380kB) | Preview

Abstract

Regulators of medical Artificial Intelligence and Machine Learning (AI/ML) are faced with a difficult problem: Should they limit marketing to a version of the system that was submitted for initial premarket review (a “locked” regime), or permit marketing of an algorithm that can adapt to changing conditions (an “adaptive” regime)? In April 2019 FDA issued a draft framework to address this problem that may become a model worldwide. In this essay, we argue that the locked/adaptive distinction and FDA’s proposed approach to it miss the central risks of medical AI/ML. Such risks emerge from structural features of predictive analytics, like concept drift, covariate shift, and AI/ML model instability. Paying attention to these risks suggests a continuous and cooperative monitoring framework of how the systems work in different and likely evolving environments, which we outline by way of conclusion.


Export/Citation: EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL
Social Networking:
Share |

Item Type: Published Article or Volume
Creators:
CreatorsEmailORCID
Babic, Boris0000000328001307
Gerke, Sara
Evgeniou, Theodoros
Cohen, Glenn
Subjects: Specific Sciences > Artificial Intelligence > AI and Ethics
Specific Sciences > Artificial Intelligence > Machine Learning
Depositing User: Boris Babic
Date Deposited: 13 Mar 2021 00:03
Last Modified: 13 Mar 2021 00:03
Item ID: 18796
Journal or Publication Title: Science
Subjects: Specific Sciences > Artificial Intelligence > AI and Ethics
Specific Sciences > Artificial Intelligence > Machine Learning
Date: 2019
URI: https://philsci-archive-dev.library.pitt.edu/id/eprint/18796

Monthly Views for the past 3 years

Monthly Downloads for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item