Munjal Shah, founder of medical AI startup Hippocratic AI, wants to leverage large language models (LLMs) to provide crucial but nondiagnostic services that address staffing shortages and improve patient relationships. His vision is for AI to act as an infinitely patient assistant that absorbs medical knowledge and communicates it conversationally. Shah believes this AI “bedside manner” could increase patient engagement, free up clinician time, and combat their burnout.
Hippocratic AI’s name connects to the physician’s oath to “first, do no harm.” The company focuses specifically on safe applications that avoid diagnostics while potentially expanding healthcare access. This article explores Munjal Shah’s inspiration, his views on AI’s promise for patient communication, Hippocratic AI’s responsible training methodology, and early testing results.
The Empathetic Potential of AI Patient Interactions
Munjal Shah sees intimate patient conversations as foundational to good healthcare. However, the average emergency room doctor cuts patients off after only 20 seconds. The system lacks the time and capacity for meaningful engagement. Shah believes LLMs uniquely address this deficit. He explained, “Language models have time, they have infinite time actually, and [they] speak every language.” Unlike overwhelmed physicians, AI can dedicate extensive time to understanding patients and explaining treatment plans conversationally.
Shah called this patient focus “bedside manner with a capital B.” It moves beyond basic instructions to forge understanding through dialogue. He envisioned AI chatbots performing formerly unscalable relational tasks like 35-minute chronic care discussions or personalized discharge follow-ups. This expanded capacity and attention could “truly” transform patient interactions.
Intriguingly, research already indicates LLMs can surpass physicians at empathetic patient communication. One study found blind reviewers preferred AI-generated responses nearly 80% of the time regarding quality and empathy. The AI appeared more consistently empathetic than the doctors.
Shah believes this result exposes a key AI advantage: lack of emotional fatigue. While AI currently falls short of human emotional intelligence overall, its tirelessness gives it an edge for skills involving great patience and attention, like counselor-patient dialogue. Unlike burnt out physicians, it can dedicate full focus to each interaction.
Training AI Responsibly for Medical Applications
While optimistic about AI’s relational potential, Munjal Shah stresses responsible implementation. He focuses on supplemental support services, avoiding high-risk diagnostics. Shah also emphasizes specialized medical training for his company’s LLM.
He explained that despite innovations like ChatGPT, most LLMs still lack sufficient health data for medical applications. Their pre-training utilizes comparatively few tokens from evidence-based sources. Hippocratic AI addresses this by sourcing content ranging from textbooks to research rather than scraping the general internet. Shah believes teaching the intricacies of health standards and processes is essential for safe functionality.
His team also refines the LLM’s knowledge through reinforcement learning with medical professional feedback. So far it has tested performance on 114 health certification exams, including role-specific tests and published benchmarks. Hippocratic AI claims significantly stronger results than both GPT-4 and other medical LLMs across most evaluations.
This rigorous methodology aligns with the startup’s mission statement – uphold the credo to “first, do no harm” by carefully targeting applications to supplement rather than replace credentialed providers. Shah sees AI as a means to expand access and fill gaps, not make high-risk autonomous decisions. His aim is improving connections and freeing up clinician time, not replacing human judgment.
The Inspiration Behind Munjal Shah’s Healthcare Vision
Unlike founders drawn to AI by technological novelty alone, Munjal Shah has direct experience both in medicine and advanced computing. He studied computer science at MIT before heading product teams at Microsoft Health, Applied Semantics (Google AdSense), and Facebook. He also mentored healthtech startups at the incubator Rock Health.
However, his most formative healthcare exposure came earlier in a personal capacity. Shah spent over five years supporting his sibling through severe chronic illness. He experienced firsthand the vital role of patient empowerment and clinician availability. But he also observed the system’s scarcity of time and staff for meaningful sustained relationships.
Shah came to recognize relational care’s importance as foundational. He told audiences it often has more influence over outcomes than diagnostic or procedural accuracy alone. Yet overload and burnout commonly force it to become a luxury not equitably accessible.
When progress in large language models revealed their aptitude for information retention and empathetic dialogue, Shah drew on his healthcare background. He recognized an opportunity to alleviate strained access to patient-centered support. Hippocratic AI could address suffering exacerbated more by resource constraints than failures of technical skill.
Current State and Future Outlook
Munjal Shah founded Hippocratic AI in 2021 to begin realizing this vision for AI supplemental health services. Just over a year later, the company has assembled an advisory board including clinical professors,DATA-DRIVEN HEALTHCARE LEADERS, and former regulators. It also recently completed a funding round adding financial backing to its staff expertise.
Hippocratic AI currently provides two main services on its website. First is a clinical evidence support tool granting subscribers searchable access to the latest peer-reviewed data. Second is a health literacy assistant chatbot for patients and caregivers. This AI assistant clarifies diagnosis details, treatment guidelines, medication interactions, and more while tracking engagement metrics for providers.
Shah ultimately plans for Hippocratic AI to offer a wide range of supportive modules handling triage, care coordination, patient education, and chronic care at large scale. He’s also exploring multilingual applications to expand access across demographics.
With responsible design, Shah believes services like appointment coordination and discharge follow-ups could help patients worldwide. They require personalized communication mastery over specialized medical prowess – an ideal application for AI strengths. At the same time, focusing less definitionally human work to AI could give caregivers more capacity for uniquely human judgment, creativity, and compassion.
In the startup Hippocratic AI, entrepreneur Munjal Shah sees an opportunity to place patient relationships at the center of care once more. He believes responsible implementation of large language models can help clinicians, not replace them. AI chatbots handling cumbersome supplementary tasks could renew human attention to the interpersonal ASPECTS OF HEALING. Meanwhile for underserved groups, AI assistants covering basics could provide on-ramps to greater inclusion.
Of course, these hopes rest on AI consistently proving itself safe and reliable for such sensitive data. But the theory aligns with evidence so far regarding the technology’s communication capabilities. Perhaps AI really could lift burdens to make space for human hands again. For the sake of both patients AND OVERTAXED CAREGIVERS, the health world will be watching Munjal Shah’s venture closely.