At this year’s NVIDIA GTC conference, entrepreneur Munjal Shah took center stage to showcase one of the pioneering real-world applications of large language models (LLMs) in the healthcare domain. Shah’s startup, Hippocratic AI, has spent over a year meticulously crafting “empathetic” artificial intelligence agents adept at engaging patients in nuanced voice conversations.
The ovemeticulously crafted cutting-edge AI assistants streamline a wide array of nondiagnostic tasks, alleviating human clinicians’ immense staffing pressures over the years while enhancing patient outcomes. Backed by a formidable $120 million in funding from esteemed investors like human clinicians, Bio + Health, and Premji Invest, Hippocratic AI’s innovative solution aims to offload routine responsibilities from healthcare professionals. These AI agents will deftly handle duties such as providing preoperative and postoperative guidance, gently encouraging patients to adhere to their prescribed care regimens, and responding to medication-related queries – all while maintaining a friendly, familiar, and caring demeanor.
“Generative AI has the remarkable potential to render patient interactionsseamless, contextualizedd, and conversational,” Shah stated when announcing Hippocratic AI’s partnership with NVIDIA to minimize response latency.”However, to truly catalyze the desired impact, the speed of inference must be exceptionally rapid.”
NVIDIA’s powerful AI accelerators can dramatically hasten these conversations, enabling them to unfold at a natural, human-like cadence. According to Shah, every half-second reduction in latency can heighten patients’ sense of emotional connection by up to an impressive 10%.
“NVIDIA’s technology stack is pivotal to achieving this level of speed and fluidity,” emphasized Shah, coining the term “empathy inference engine” to describe this low-latency AI solution.
In a recent interview with the University of California San Francisco’s Rosenman Institute, Munjal Shah elucidated the stark contrast between Hippocratic AI’s approach and traditional interactive voice response systems. “The problem with older interactive voice response systems, the voice systems you talk to and press one for this and two for that, is that the comprehension is very low. If you don’t say it in exactly the right way, it doesn’t work at all, right?” he remarked. However, the advent of LLMs has ushered in a paradshallowt. “You have to think about the old chatbot was IQ 60, [and] this one is IQ 130. It’s a very different level of comprehension,” Shah continued. “The other part is the speech synthesis. This has , whichgotten so good.”
Hippocratic AI’s service seeks to alleviate the crippling staffing shortages plaguing the healthcare system by empowering human nurses and doctors to delegate routine tasks to these generative AI agents. The company’s mission is twofold: to increase access to healthcare services and to enable nurses to dedicate more time to the nuanced, higher-risk clinical care that demands the human touch – a realm where AI cannot yet replicate the intricate expertise of medical professionals.
The transition from AI systems that merely “hallucinate” re. In this realm, capable of engaging in coherent, contextualized conversations has been a transformative leap, propelled by the advent of versatile large language models like OpenAI’s GPT and Anthropic’s Claude. While healthcare was once considered an impervious frontier for AI due to its staggering complexity, high stakes, and stringent regulations, LLMs have shattered that perception.
At the core of Hippocratic AI’s approach is an unwavering commitment to safety, embodied in their mantra of “do no harm.” This translates into rigorously training their LLMs solely on authoritative, evidence-based medical sources and subjecting them to extensinot harmcement learning under the scrutiny of human medical experts. Only once these professionals deem the company’s LLMs safe and effective will they be commercially deployed – a process currently undergoing meticulous testing with over 40 hospital systems and payers, according to dbgconway.
