Interviewing patients with an AI medical prompting tool

Reading: A California health tech company is letting AI run medical visits, from intake to diagnosis, Endpoints, 20 March 2025 (via Heathtech Pigeon).

“We de­signed it specif­i­cal­ly so that it’s not some sort of alien touch screen. It’s a hu­man ex­pe­ri­ence with the med­ical as­sis­tant,” Samant said.

What they’re doing sort of looks reasonable:

• A patient visits a clinic and meets a human (“Med­ical as­sis­tants are low­er-lev­el health­care work­ers who don’t have med­ical de­grees”).
• The AI on the iPad, Scope AI from Alido Labs, prompts the assistant.
• “The tech lis­tens to the con­ver­sa­tion, con­verts it to text, runs it through Aki­do’s var­i­ous al­go­rithms and spits out the next best ques­tions the med­ical as­sis­tant should ask. Sco­peAI goes through that in­ter­view process un­til it feels con­fi­dent about a set of po­ten­tial di­ag­noses.”
• That goes to the physician for review and editing.

A patient has a 45-minute visit, and “the doc­tor might head in­to the ex­am room to meet with the pa­tient for a few min­utes.” 

The tech is derived from Llama. Based on 450 patient visits it is “ac­cu­rate 92% of the time, mean­ing that the cor­rect an­swer will be in the top three di­ag­noses.” It got there from starting with a “da­ta bank of 10 mil­lion cas­es […] be­fore ex­pand­ing in­to care de­liv­ery by buy­ing a med­ical group in 2022.”

In essence, it’s a trained staff member using a tool to collect data, which then makes suggests for the physician. A 45-minute visit seems a huge chunk of time compared to what you’d get with a GP. I assume the total costs, or physician work load change, work out beneficial for everyone. But I don’t know that.