Serial entrepreneur Munjal Shah sees generative AI as a way to provide supplemental healthcare services amid a growing staffing crisis. He argues that while unfit for diagnosis, large language models (LLMs) can take on critical nondiagnostic tasks if focused narrowly and appropriately trained. Shah’s new startup, Hippocratic AI, aims to create virtual assistants that act as chronic care nurses and patient navigators. His vision is “super staffing,” with AI agents providing support at a fraction of the cost to ease the burden on human health workers.
Munjal Shah has decades of experience in AI but believes generative models represent an unprecedented breakthrough. Unlike past classifier systems focused on categorization, LLMs can generate new content while capturing the nuance and empathy needed for sensitive conversations. They have near limitless capacity for patient engagement if designed thoughtfully. However, as seen with public tools like ChatGPT, mistakes inevitably occur. Applying them clinically requires tight constraints, both technically and legally. Diagnosis, in particular, remains firmly in human territory.
Fortunately, Shah sees plenty of space for diagnostic applications, which form the core of Hippocratic AI’s offerings. This includes medication and appointment reminders, coordinating transportation, connecting patients with food assistance, and overall care plan guidance. While not hands-on medical care, these supplementary services are crucial for outcomes yet must be addressed due to overworked staff. Dedicated support channels can also free clinicians to focus on more complex cases. The economics likewise work favorably – unlike human nurses who require shifts and time off, virtual assistants can scale cost-effectively with no risk of burnout.
To minimize mistakes, Hippocratic AI trains its models exclusively on medical language. Shah stresses the importance of a human-like conversational style, with empathy and bedside manner as crucial as factual knowledge. This helps ensure patients feel comfortable engaging openly. The company refines its system based directly on feedback from actual doctors and nurses. Over time, Shah believes purpose-built healthcare LLMs can become handy tools for medical teams. They will never fully replace human judgment but may one day provide a complete infrastructure for personal guidance.
In the race to find applications for generative AI, healthcare stands out with an enormous deficit of staff hours relative to patient needs. Munjal Shah and Hippocratic AI are betting that LLMs can help provide that supplemental support, even if a diagnosis remains off the table. With the proper constraints and training, virtual health assistants may one day offer the super staffing needed to manage chronic conditions worldwide properly. For now, small steps like appointment reminders and care coordination show the promise of thoughtfully applied LLMs in health.