If AI is biased, blame the humans

Presented by Better Medicare Alliance: The ideas and innovators shaping health care

Dec 18, 2023 View in browser   By Daniel Payne, Evan Peng, Erin Schumaker and Ruth Reader Presented by PROGRAMMING NOTE: We’ll be off next week for the holidays but back to our normal schedule on Tuesday, Jan. 2. TECH MAZE If AI inputs aren't diverse, don't trust the recommendations. | Christophe Archambault/AFP via Getty Images Bias may be among the most difficult problems to solve within artificial intelligence systems. But some health leaders look to a future in which AI helps correct human bias. “The human is the problem … that’s where the bias comes from,” said Forrest Pascal, principal of AI/ML model governance at Kaiser Permanente. During a panel at the Healthcare Information and Management Systems Society conference Friday in San Diego, he joined several industry leaders to discuss implementing the technology in an ethical way. “How do I put guardrails on that human that’s in the loop?” he said. AI could be a powerful tool to check clinicians’ biases — not just as individuals but also across systems, Sophia Henry, clinical consultant at Beckman Coulter Diagnostics, said. One key, Pascal and Henry agreed, is having a diverse group of people involved in making AI tools and principles that guide the use of those tools. Another is carefully considering all the pieces of data fed into AI systems. Even so: Creating an AI system without bias is a near-impossible task, health and tech experts say — because the data on which the models are trained often comes from biased clinicians. “Our AI products are really only as good as the data that we bring to the table,” said Rachini Moosavi, chief analytics officer at UNC Health. The end goal, then, must be developing AI that can detect bias and mitigate it. A message from Better Medicare Alliance: We support policy solutions to strengthen Medicare Advantage and improve the beneficiary experience. Our policy priorities include modernizing prior authorization, establishing marketing guidance clarity, and improving provider directories. Learn more about Better Medicare Alliance's recommendations for Medicare Advantage priorities.   WELCOME TO FUTURE PULSE San Diego, Calif. | Daniel Payne/POLITICO This is where we explore the ideas and innovators shaping health care. Researchers are using drones and sensors to scan for mosquito breeding sites — like stagnant water — in Peru's Amazon rainforest, The Washington Post reports. Their aim is to map where diseases are going next as rising temperatures make the country more hospitable to the infection-spreading bugs. Share any thoughts, news, tips and feedback with Carmen Paun at cpaun@politico.com, Daniel Payne at dpayne@politico.com, Evan Peng at epeng@politico.com, Ruth Reader at rreader@politico.com or Erin Schumaker at eschumaker@politico.com. Send tips securely through SecureDrop, Signal, Telegram or WhatsApp. Today on our Pulse Check podcast, host Kelly Hooper talks with E&E News' public health reporter Ariel Wittenberg, who breaks down her investigation into the heat-related illnesses — and one fatality — afflicting letter carriers and how the Postal Service allegedly falsified its training records at the risk of endangering workers' health. Listen to today’s Pulse Check podcast   A message from Better Medicare Alliance:   FORWARD THINKING AI adopters are looking for the low-risk, high-reward options. | Damien Meyer/AFP via Getty Images AI will change pretty much every part of health care, according to Harjinder Sandhu, CTO at Microsoft Health and Life Sciences. That poses an immediate challenge for health systems: Where to start? At the HIMSS conference on AI in health care, Sandhu offered a few ways for health organizations to think about how advanced AI could best fit into their work. A funnel: One way to weed out AI applications that don’t make sense is to run them through a “funnel” of questions, such as: — Is the problem too risky for AI to tackle? — Will the system perform well? — Is the cost of an AI system appropriate? — What systems will ensure a model works? — How will the deployment of these systems be sustained as models change? And a chart: Health systems could also use four quadrants on a chart with two axes — risk and value — to help decide where to introduce AI. High-value, low-risk options, such as transcribing doctor’s notes from a conversation with patients, could rise to the top of the list of priorities, Sandhu said. Why it matters: Central to many considerations, Sandhu said, is that large language models aren’t considered trustworthy in many contexts. But humans themselves aren’t always trustworthy, he countered — so systems exist to check us. Those systems are all the more important now that AI is part of the mix, he said. A message from Better Medicare Alliance: More and more seniors are choosing Medicare Advantage than ever before, including in rural communities. In 2023, 40% of all eligible Medicare beneficiaries in rural counties were enrolled in an MA plan—nearly four times the share in 2010. As rural Medicare Advantage enrollment grows, the program is making care more affordable for rural communities because Americans in rural areas are more likely to face financial challenges than those in urban areas. Medicare Advantage covers all of the same services as Fee-For-Service (FFS) Medicare, but offers additional cost protections, including an annual cap on out-of-pocket expenses and additional benefits that support beneficiaries' overall health, like dental and vision. Ultimately, beneficiaries save $2,400/year on average compared to FFS Medicare beneficiaries. Read about our proposed solutions for improving and maintaining care for seniors.   INNOVATORS Microrobot makers hope they can deliver medicine more efficiently and safely. | Robyn Beck/AFP via Getty Images Microrobots that deliver drugs to targeted locations or groups of cells in a patient’s body hold great promise for curing disease, argue Bradley J. Nelson and Salvador Pané, microrobotics professors at ETH Zurich, in a new piece for Science. Say what? Nelson and Pané explain that microrobots as small as a few nanometers long, one-millionth of a millimeter, could eventually carry drugs precisely to where they’re needed in a patient’s body. A catheter could carry the microrobot close to the target before releasing it into the bloodstream. From there, external magnetic fields could guide the robot. Why it matters: Most drug candidates fail during clinical trials. Among the causes: Drugs turn out to be toxic to humans. In other words, a pill can kill a disease, or whatever medical problem it was designed to solve, but simultaneously damage healthy parts of a patient’s body. Microrobots could overcome toxicity since only a localized area would be exposed instead of whole organs. Microrobots could also speed delivery of drugs to target areas, overcoming slow drug-diffusion rates that can undermine the medicine’s effectiveness. What’s next? Nelson and Pané say researchers will need to: — Find a suitable, nontoxic material for the microrobots that responds to magnets — Develop a system for tracking the microrobots since conventional medical imaging techniques aren’t there yet — Determine how many microrobots are needed to deliver a sufficient and effective dose of medicine — Obtain regulatory approval of the technology   Follow us on Twitter Carmen Paun @carmenpaun Daniel Payne @_daniel_payne Ruth Reader @RuthReader Erin Schumaker @erinlschumaker   Follow us

To change your alert settings, please log in at https://www.politico.com/_login?base=https%3A%2F%2Fwww.politico.com/settings This email was sent to rouf@idiot.cloudns.cc by: POLITICO, LLC 1000 Wilson Blvd. Arlington, VA, 22209, USA Unsubscribe | Privacy Policy | Terms of Service