Dive Brief:
- OpenAI on Wednesday rolled out dedicated health information chatbot ChatGPT Health in one of the artificial intelligence giant’s biggest moves into healthcare.
- Users can connect their medical records and wellness apps, like Apple Health and MyFitnessPal, so ChatGPT Health can walk them through recent test results, help them prepare for doctor’s appointments, put together workout routines or compare health plan options.
- The health chatbot includes additional privacy protections, including purpose-built encryption, and information is stored separately from other conversations with ChatGPT, according to OpenAI. Data from ChatGPT Health also won’t be used to train foundation models, the company said.
Dive Insight:
Users are already turning to ChatGPT for health information. More than 40 million people ask the chatbot healthcare questions each day, including queries on navigating the healthcare system and understanding their insurance, according to a report released this week by OpenAI.
And a majority of health conversations take place outside normal clinic hours, suggesting people are taking questions to ChatGPT when they can’t easily access their doctors, the AI firm said.
Releasing a health-specific offering that can utilize users’ personal information should help customers manage their health — often a challenge for patients navigating the complex and fragmented U.S. healthcare system, Fidji Simo, CEO of applications at OpenAI, wrote in a Wednesday blog post.
Many Americans report skipping needed care due to cost concerns, and people living in rural areas often face long travel times to reach providers.
Plus, clinicians often have limited time with patients. Providers have long complained about a growing burden of administrative work and tasks in electronic health records that they say distract from patient care and drag work into off-hours.
“Under these conditions, it’s almost impossible for doctors to look beyond the immediate problem in front of them and see the full picture of someone’s health,” Simo wrote. “That lack of context often means treating individual symptoms in isolation, rather than addressing the underlying drivers.”
The chatbot will first be available to a small group of early users, and OpenAI will expand access to ChatGPT Health in the coming weeks, the company said.
But there are risks to using AI for healthcare. The tools can hallucinate, offering inaccurate or misleading information that many patients could struggle to recognize.
A spokesperson for OpenAI said it worked with physicians to develop and run health-specific evaluations of ChatGPT Health as well as review the model’s responses. More than 260 doctors worked with OpenAI to create HealthBench, a framework that evaluates the chatbot’s answers for safety, clarity and how well it encourages follow-up care with clinicians when needed.
New GPT models are also more accurate and reliable compared to older models, as they’re more likely to ask follow-up questions, browse the internet for the latest research and direct users to professional evaluation, the OpenAI spokesperson said.
Still, ChatGPT Health should not replace medical care, and it isn’t designed for diagnosis or treatment, OpenAI noted. The goal is to ask everyday questions and help users understand patterns in their health over time, the company said.
