John Ayers has watched healthcare AI generate massive excitement and investment, but he believes most of the hype has not translated into meaningful patient impact. This frustration drove Ayers and a team of researchers to create ChatCPR, an AI agent launched this week that coaches users through CPR in real time.
The tool represents a shift from backend healthcare AI applications toward direct patient care. With more than 350,000 Americans suffering cardiac arrests outside hospitals annually and a 90% fatality rate, the researchers focused on an area where seconds count and existing systems often fail.
From hype to reality in healthcare AI
Ayers, head of AI at the University of California San Diego’s Altman Clinical and Translational Research Institute, gained prominence after leading a widely discussed 2023 JAMA study. The research found AI chatbots’ responses to patient messages are often more accurate and empathic than those written by human doctors.
Despite his reputation as a healthcare AI expert, Ayers remains skeptical about the field’s actual impact. During a recent television news appearance, a reporter asked him about AI saving lives. “What are you talking about? It’s not really helping anybody yet – it’s all hype, no reality. The way it’s being delivered is all on the back end,” he responded.
Targeting cardiac arrest response gaps
The timing challenge in cardiac arrest response is stark. Only 2% of Americans are CPR-certified, meaning most people simply call 911 and wait when someone collapses. This delay proves costly since CPR efficacy drops with each passing minute.
Ayers and his UC San Diego research team partnered with researchers from Johns Hopkins and UPMC to address this gap. They tested major AI models including OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini on CPR instruction tasks.
While most models handled basic CPR guidance reasonably well, they often missed more nuanced but clinically important instructions. This gap led the team to build ChatCPR with more advanced, guideline-specific capabilities.
Outperforming 911 dispatchers in testing
A study published Monday in JAMA introduced ChatCPR and demonstrated its effectiveness. When tested against recordings from real 911 calls, the AI tool outperformed 911 dispatchers in guiding bystanders through CPR procedures.
The researchers chose to release ChatCPR as an open-source public resource rather than pursuing commercial development. They are making the training materials, guidelines, prompts, and architecture publicly available for companies and emergency-response organizations to build upon.
Implementation over sophistication
Ayers believes the key challenge in healthcare AI is implementation rather than having the most advanced model. The team intentionally built ChatCPR on a relatively small, lower-performing language model and still achieved strong results through careful design and domain-specific training.
This approach means the tool could eventually run directly on smartphones without internet connectivity. The offline capability could prove crucial in emergencies where network access is limited or unreliable.
Addressing healthcare access disparities
The researchers see ChatCPR as a potential tool for reducing disparities in emergency care access. “CPR should not be a luxury good. But even in the wealthiest country in the world, depending on where you’re at and the resources around you for emerging medical services, it is often a luxury,” Ayers noted.
By open-sourcing the tool, Ayers hopes to accelerate its adoption and improvement across the healthcare system. The approach prioritizes widespread access over proprietary control, aligning with the team’s goal of creating measurable patient impact rather than commercial success.