Harvard, MIT researchers warn of adversarial attacks in AI

Hacker accessibility and the potential for security breaches weigh heavily on the development of successful AI systems, but a more pressing threat might lie with healthcare regulators like insurance providers and billing companies, the New York Times reported of a Science study March 21.

While neural networks train themselves to complete certain tasks by analyzing mass amounts of data, it takes just a small digital manipulation to change the core function of a technology, Samuel Finlayson, a researcher at Harvard Medical School and MIT and an author of the paper, told the Times. Just as researchers have trained AI systems to detect breast cancer, a hacker could manipulate code to not flag malignancies.

Researchers have proven time and again that AI can have unintended negative consequences—self-driving cars have interpreted stop signs as yield signs, for instance, and AI-powered eyeglasses have been tricked into thinking they perceive celebrities. But on the less clinical side of things, Finlayson and his colleagues said they’re concerned about what powerful healthcare stakeholders could achieve with data manipulation.

“The inherent ambiguity in medical information, coupled with often-competing financial incentives, allows for high-stakes decisions to swing on very subtle bits of information,” Finlayson told the Times.

He said that since there is so much money exchange within the healthcare industry, insurance and billing companies are already “bilking the system” by changing billing codes and other data within their computer systems. As AI becomes more commonplace in those settings, there’s more of a possibility stakeholders could manipulate scans for better payouts or alter images for expedited regulatory approval.

“Some of the behavior is unintentional, but not all of it,” Hamsa Bastani, an assistant professor at the University of Pennsylvania who’s studied the manipulation of healthcare systems, told the Times. “There are always unintended consequences, particularly in healthcare.”

Read the full story below:

""

After graduating from Indiana University-Bloomington with a bachelor’s in journalism, Anicka joined TriMed’s Chicago team in 2017 covering cardiology. Close to her heart is long-form journalism, Pilot G-2 pens, dark chocolate and her dog Harper Lee.

Around the web

The tirzepatide shortage that first began in 2022 has been resolved. Drug companies distributing compounded versions of the popular drug now have two to three more months to distribute their remaining supply.

The 24 members of the House Task Force on AI—12 reps from each party—have posted a 253-page report detailing their bipartisan vision for encouraging innovation while minimizing risks. 

Merck sent Hansoh Pharma, a Chinese biopharmaceutical company, an upfront payment of $112 million to license a new investigational GLP-1 receptor agonist. There could be many more payments to come if certain milestones are met.