Latest / Becker’s Healthcare Digital Health + Health IT / AI chatbots spread medical misinformation: Mount Sinai, UPMC + UChicago Medicine back AI startup in $55M funding round + more
Transcript
- 0:00This is Naomi Diaz with the Becker's Digital
- 0:02Health podcast. Here's your daily industry news briefing
- 0:05for today, August 6.
- 0:07First,
- 0:08researchers from the Icahn School of Medicine at
- 0:11Mount Sinai found that widely used AI chatbots
- 0:14designed for medical support
- 0:16are vulnerable to repeating and elaborating on false
- 0:19health information.
- 0:21Published August 2 in communications medicine,
- 0:24the study tested how large language models respond
- 0:27to fictional patient scenarios containing made up medical
- 0:30terms.
- 0:31The chat bots often accepted the false information
- 0:34without question,
- 0:35generating detailed responses about nonexistent
- 0:38conditions, symptoms, or treatments.
- 0:42Researchers then introduce a simple safeguard,
- 0:45a one sentence warning at the start of
- 0:47the prompt reminding the AI that the information
- 0:49provided may be inaccurate.
- 0:52With this added prompt, the rate of errors
- 0:54drops significantly,
- 0:55suggesting that even minimal guardrails
- 0:58can help reduce misinformation
- 1:00generated by AI systems.
- 1:02The findings highlight the risk of relying on
- 1:04AI chat box for medical decision making and
- 1:07underscore the importance of prompt engineering,
- 1:10system design, and human oversight.
- 1:13According to the research team, the results support
- 1:15the need to test AI tools using fabricated
- 1:18inputs before they are deployed in clinical settings.
- 1:21Researchers plan to apply the method to de
- 1:23identified patient data and explore additional safety props
- 1:27and retrieval tools as AI technology
- 1:30continues to evolve.
- 1:32Second,
- 1:33the venture capital arms of UPMC and UChicago
- 1:36Medicine have invested in a health care AI
- 1:39startup as a part of a $55,000,000
- 1:41series c funding round.
- 1:44UPMC
- 1:45enterprises and UCM ventures back all Ultramix,
- 1:49which says it offers its first FDA cleared
- 1:52Medicare reimbursed
- 1:53AI technology
- 1:54to aid in detecting cardiac amyloidosis
- 1:57and heart failure with preserved ejection fraction.
- 2:00The company developed the tool in collaboration with
- 2:03Mayo Clinic.
- 2:05UltraMix other health system partners include university hospitals
- 2:09and Northwestern
- 2:10Medicine.
- 2:11The startup says this platform helps hospitals cut
- 2:13unnecessary
- 2:14tests, enhance workflows, and start treatments earlier.
- 2:18Third,
- 2:19health systems are increasing their adoption of Microsoft's
- 2:22Ambient AI for clinical documentation.
- 2:26Here are four things to know from a
- 2:27July 30 Microsoft earnings call.
- 2:30First, Microsoft had a breakout physical 2025
- 2:33for the use of its Dragon co pilot,
- 2:35AKA DAX, which ambiently listens to patient visits
- 2:38and drafts clinical notes for the EHR, said
- 2:41Microsoft chair and CEO, Sato Nadala.
- 2:44Second,
- 2:45health care customers employ the tool to document
- 2:47over 13,000,000
- 2:48patient encounters in the most recent quarter, up
- 2:50nearly seven times year over year.
- 2:53Third,
- 2:54at Mercy, more than 1,000 physicians are now
- 2:57using Dragon Co pilot.
- 2:59Four, mister Nadella said of Mercy, quote, they
- 3:02have saved more than a 100,000
- 3:05to date and plan to expand to all
- 3:075,000 providers.
- 3:08As one physician put it, the best thing
- 3:11to happen to my practice in ten years,
- 3:13end quote.
- 3:14If you would like the latest health IT
- 3:16and digital health industry news delivered to your
- 3:18inbox every morning,
- 3:20subscribe to the Becker CIO and health IT
- 3:22review e newsletter through our website at ww.beckershospitalreview.com.