Munjal Shah’s Hippocratic AI: Using Language Models to Resolve the Health Care Staffing Crisis

The health care industry is facing a monumental staffing crisis. As highlighted by Munjal Shah, CEO and founder of Hippocratic AI, nearly 17% of US hospitals are anticipating critical shortages of health care professionals. The nursing shortage is especially dire, with over 200,000 open positions expected annually through 2031. Shockingly, 30% of nurses have contemplated leaving the field after the stresses of the COVID-19 pandemic.

Munjal Shah believes large language models (LLMs) like ChatGPT and Google’s Bard can help resolve this crisis through “superstaffing.” While productivity tools may provide marginal capacity increases, LLMs offer exponential growth in health care resources. As Shah explains, “Superstaffing can give us 10 times or 100 times more capacity.” By leveraging the conversational abilities of LLMs, they can fill gaps left by departing nurses and expand services not viable with human staffing levels.

The Goal of Hippocratic AI: Deploying LLMs Ethically in Health Care

Munjal Shah founded Hippocratic AI to explore safe applications of generative AI in health care. The name ties to the medical maxim “first, do no harm.” Shah recognizes that despite their promise, LLMs still “hallucinate” incorrect information potentially harmful if relied upon. So Hippocratic AI focuses these models on low-risk, non-diagnostic tasks like patient coaching and coordination. The aim, per Munjal Shah, is “to help patients using powerful LLMs while avoiding high-risk applications that require diagnostic trust.”

For instance, LLMs could provide chronic care support for diagnosed patients, ensuring prescription adherence and appointment attendance. With human-level conversation, they can also address social determinants like nutrition access. As Shah explains, “Chronic care nurses don’t diagnose you. They just ask questions like: ‘Did you take your meds? Do you need a ride to your next appointment?” While still early, Hippocratic AI’s LLM has surpassed state-of-the-art models like GPT-4 on over 100 medical exams. Its health care specialization makes it uniquely apt for these applications.

How Munjal Shah Believes AI “Superstaffing” Can Resolve Health Care Shortfalls

Currently, resource constraints prevent nurses from providing consistent chronic care outreach. With only around 3 million RNs nationally, Shah points out “They couldn’t possibly handle 68 million people on a chronic care basis.” But LLMs can scale infinitely at a fraction of human costs. He estimates LLMs at $1 per hour versus upwards of $90 per hour for nurses. This vast savings enables new interventions financially unviable before. As Shah notes, “Would we call every patient two days after they start every medication just to check in and see if they’re having any weird side effects? Of course we would.”

By collaboratively developing safe use cases with medical experts, Hippocratic AI aims to deploy LLMs ethically. Munjal Shah explains “Nurses who do the chronic care calls today are the best ones to judge whether our algorithm is good at doing chronic care calls.” With their guidance, the LLM learns to produce approved responses. Ultimately, Shah believes this superstaffing can provide substantial relief to the nurses bearing the brunt of shortages partly driven by pandemic-related departures. In his view, “it may be nurses who help tune LLMs” to fill persistent gaps.

Concerns Around Generative AI in Health Care Applications

With game-changing potential comes apprehension. And generative AI models like ChatGPT prompt reasonable worries when applied insensitivity domains like health care. While mostly harmless, these systems do “hallucinate” false information seemingly true. When asked an impossible question about relocating the Golden Gate Bridge, ChatGPT fabricates a detailed, bogus response. And they can also display frightening accuracy, recently passing medical licensing exams. This dichotomy leaves patients rightly questioning whether LLMs merit trust for medical uses.

As Munjal Shah cautions, some applications are too “high-risk” currently, like autonomous diagnosis. But he believes lower-risk, nondiagnostic tasks still offer massive upside for supervision-minded deployment. Particularly for overburdened nurses facing untenable workloads, Munjal Shah sees LLMs as a supporting resource to expand care. But responsible development demands deliberate constraints.

Building Responsible LLM Applications Through Collaboration

Munjal Shah insists responsible generative AI in areas like health care requires collaboration between both medical experts and AI developers. At Hippocratic AI, experienced nurses and doctors actively evaluate output quality and shape appropriate use cases. Per Shah, “help us test the model…Let’s have them judge the product on a daily basis.” With practitioner guidance, the risks shrink substantially.

And the benefits only grow. Chronic care may improve exponentially with LLMs handling routine check-ins, freeing up nurses for more meaningful work. Patients see more holistic support. Costs drop dramatically. And burnout cases could taper amid alleviated workloads. But reaching this potential hinges on building these systems cooperatively, tuning model responses until they align with expert standards. With dedication from leaders like Munjal Shah to enable generative AI ethically, health care could witness an innovation renaissance unseen since the dawn of modern medicine.

Realizing the Promise: Deploying Generative AI Responsibly to Transform Care

Munjal Shah resonates optimism balanced with diligence regarding AI’s prospective health care impact. By picking lower-risk applications and developing them collaboratively, he believes generative models can bring languishing visions to life. Ample challenges demand solutions before AI can deliver on its immense promise.

Concerns linger about patient safety and responsible oversight. Workforce anxiety persists surrounding AI integration. And health equity worries concentrate on marginalized community access. But the sheer scale of suffering from systemic care gaps begs the question: what if AI can help? And according to Munjal Shah, applied conscientiously, generative models could soon support millions lacking adequate care.

The road ahead remains long, as new technologies often progress in fits and starts. But leaders like Munjal Shah are carefully laying the tracks for AI to uplift patients safely. If successful, Shah’s vision of supersized staffing could spur care model transformation. More support flowing to more people. Patients feeling consistently heard. Providers feeling consistently supported. And LLMs helping pave the path, learning dynamically to elevate care standards for all.

About Post Author

Roy Reyna

Roy Reyna is a congressional reporter for Pluralist. He covers the U.S. Congress and its members, focusing on the House of Representatives. Prior to joining Pluralist, he worked as a reporter for National Journal and The Hill, covering Congress and politics. Reyna has a background in political science and has written extensively on the inner workings of Congress, as well as the political and policy issues that drive it. He is based in Washington, D.C.

See author's posts