In case you’d heard any scary tales about AI advancing to the purpose of taking clinicians’ jobs, here is your common reminder that we’re very, very removed from such a situation.Driving the information: In a latest thread, Twitter consumer Lucy Hao shared a number of screenshots from an MIT article detailing issues that befell a number of AI instruments designed to assist determine COVID-19. Why it issues: In even essentially the most superior of circumstances, well being care-related algorithms are usually being utilized in an effort to information medical determination making — not exchange a physician’s judgment.Even in relation to these kinds of instruments, builders have to be cautious to make sure their algorithms are skilled with the suitable knowledge and account for any confounding variables.One widespread problem that is surfaced a number of late entails AI instruments skilled on non-diverse populations after which deployed on various populations. Can you say error-prone?(Small) particulars: When unaccounted for, surprising variables rendered a number of COVID algorithms ineffective.In one instance from the article, builders skilled their instrument on a dataset containing chest scans of youngsters with out COVID. The ensuing AI instruments “realized to determine children, not COVID.”In one other, sufferers who have been scanned whereas mendacity down have been extra more likely to be critically ailing. So “the AI realized wrongly to foretell critical COVID danger from an individual’s place.” Oops.
https://www.axios.com/pro/health-tech-deals/2022/03/30/medical-ai-meet-reality
AI meets medical reality
