Can artificial intelligence help medical professionals treat California’s unhoused population? Or will the technology open a can of worms that critics say may do more harm than good?
As CalMatters’ Marisa Kendall explains, Los Angeles-based Akido Labs plans to use an AI model it developed on unhoused patients next month in the Bay Area. Akido, a health care technology company, developed Scope AI, a tool that aims to increase health care access for homeless people.
Scope assists non-medically trained outreach workers in beginning the intake and diagnosis process with homeless patients. The tool generates questions that workers ask patients, and it listens to, records and transcribes the interview. Afterwards, Scope suggests diagnoses, medical tests and medication.
The information is then sent to a human doctor, who reviews the interaction and can sign off on Scope’s medication suggestions or make changes. For more complex cases, doctors can arrange to see the patient themselves.
As early as 2023, Aikido’s outreach workers have been using Scope in homeless encampments in L.A. County, where it has since seen more than 5,000 patients. Scope lands on the correct diagnoses within its top three suggestions 99% of the time, according to Akido, and street medicine doctors in L.A. and Kern counties have increased their case load from roughly 200 homeless patients at a time to 350 after implementing Scope.
But critics have concerns about the AI’s reliability, how it could put patient’s data at risk and how it could reinforce biases. AI is more likely, for example, to misdiagnose breast cancer in Black women than in white women, according to a 2024 study.
Because of their increased vulnerability and unique circumstances, homeless patients also may not be treated as precisely by AI compared to a human health care provider. A patient with scabies, for example, would typically be prescribed special shampoo or body wash, said Brett Feldman, founder of USC Street Medicine. But for an unhoused person who does not have regular access to a bathroom, an oral medication may be needed. Would AI know to flag that detail?
Feldman: “I would say, in general, that this would not work for this population.”