Text to avoid paywall
The Food and Drug Administration is planning to use artificial intelligence to “radically increase efficiency” in deciding whether to approve new drugs and devices, one of several top priorities laid out in an article published Tuesday in JAMA.
Another initiative involves a review of chemicals and other “concerning ingredients” that appear in U.S. food but not in the food of other developed nations. And officials want to speed up the final stages of making a drug or medical device approval decision to mere weeks, citing the success of Operation Warp Speed during the Covid pandemic when workers raced to curb a spiraling death count.
“The F.D.A. will be focused on delivering faster cures and meaningful treatments for patients, especially those with neglected and rare diseases, healthier food for children and common-sense approaches to rebuild the public trust,” Dr. Marty Makary, the agency commissioner, and Dr. Vinay Prasad, who leads the division that oversees vaccines and gene therapy, wrote in the JAMA article.
The agency plays a central role in pursuing the agenda of the U.S. health secretary, Robert F. Kennedy Jr., and it has already begun to press food makers to eliminate artificial food dyes. The new road map also underscores the Trump administration’s efforts to smooth the way for major industries with an array of efforts aimed at getting products to pharmacies and store shelves quickly.
Some aspects of the proposals outlined in JAMA were met with skepticism, particularly the idea that artificial intelligence is up to the task of shearing months or years from the painstaking work of examining applications that companies submit when seeking approval for a drug or high-risk medical device.
“I don’t want to be dismissive of speeding reviews at the F.D.A.,” said Stephen Holland, a lawyer who formerly advised the House Committee on Energy and Commerce on health care. “I think that there is great potential here, but I’m not seeing the beef yet.”
Or maybe that is part of the allure of automation: the eschewing of human responsibility, such that any bias in decision making appears benign (the computer deemed it so, no one’s at fault) and any errors - if at all recognized as such - become simply a matter of bug-fixing or model fine-tuning. The more inscrutable the model the better in that sense. The computer becomes an oracle and no one’s to blame for its divinations.
I saw a paper a while back that argued that AI is being used as “moral crumple zones”. For example, an AI used for health insurance acts allows for the company to reject medically necessary procedures without employees incurring as much moral injury as part of that (even low level customer service reps are likely to find comfort in being able to defer to the system.). It’s an interesting concept that I’ve thought about a lot since I found it.
I can absolutely see that. And I don’t think it’s AI-specific, it’s got to do with relegating responsibility to a machine. Of course AI in the guise of LLMs can make things worse with its low interpretability, where it might be even harder to trace anything back to an executive or clerical decision.