Unleashing the Power of Generative AI and LLMs in Healthcare: Insights from Industry Experts
At the 2023 Re View Summit, AI Panelists explored how CEOs can get started, viable opportunities within the market and what investors and health systems are seeking when it comes to practical solutions. The panel, moderated by Venture Chair James Quarles, featured industry experts including Fawad Butt, Redesign Health Advisor and Executive in Residence at Canvas Ventures; Mike Desjadon, CEO of Anomaly (a company built at Redesign Health); Jon Moore, Head of Engineering Project & Program Management at Cohere; and Dr. Justin Norden, Partner at GSR Ventures and Adjunct Professor at Stanford Medicine. Here are some of the key takeaways from this session: Companies can get started by leveraging in-house talent When it comes to integrating generative AI into existing solutions, startups don’t need to rush into hiring tech talent. “Before you hire, you should make sure what you’re doing [with AI] is core to the problem you’re trying to solve,” Desjadon said. Moore noted, “AI's biggest value-add is not replacing expertise, but allowing experts to spend less time doing busy work and more time on interesting problems that are best suited to humans.” “AI's biggest value-add is not replacing expertise, but allowing experts to spend less time doing busy work and more time on interesting problems that are best suited to humans.” - Jon Moore Once you identify the right use case, ask your team to explore free tools like Chat GPT, Coral Chat and Bard to craft your vision and generate your thesis. “AI today isn’t where it was eight years ago, where you needed to understand 800 different constructs to use it,” Butt said. “You can take a software engineer or data scientist on your team, give them a week to learn it and come back, and they’ll be as efficient as most of the people you’ll hire out there." “You can take a software engineer or data scientist on your team, give them a week to learn it and come back, and they’ll be as efficient as most of the people you’ll hire out there.” - Fawad Butt AI isn’t foolproof, which means it isn’t ideal for use cases that require 99.99% accuracy — especially in healthcare. “Start with something you don’t have to be perfect in, like summarization,” Dr. Norden said. “Then always be sure to keep a human in the loop.” New AI technologies, like retrieval-augmented generation, can help improve accuracy, Moore explained. These tools ground results in specific data sources that teams can determine in advance, rather than attempting to create answers purely from within the model. Nimble startups will have an advantage over larger incumbents The panelists agreed that startups have a significant head start when it comes to incorporating this technology due to their agility. “Faster implementation means that you can try more AI integrations with lower risk,” said Moore. “Startups also benefit the most from using AI internally. I’d encourage leaders to explore AI-powered use cases that help scale and amplify your limited HR, marketing and customer service staff.” Meanwhile, incumbents—especially health systems and payers—will be hindered by their own internal processes. “You can do in six months what would probably take them three years and $100 million [to do],” Butt said. But Dr. Norden cautioned entrepreneurs to avoid competing with large technology companies, including powerhouses like Epic and Amazon, that are building these models and their components. These organizations will likely be able to develop commoditized components more quickly. Startups can then purchase and combine these components in novel ways. “This is a capability that you can either build or buy,” Dr. Norden said. “Today we’re building it. Tomorrow, we’ll buy it.” Companies must show tangible AI value to investors As companies rush to incorporate generative AI into their business models, investors and buyers—especially health systems—want to see how developers plan to use these tools to solve real problems. “This is the most transformative technology we’ve ever seen,” Dr. Norden said. "If you’re not mentioning generative AI, that’s an issue.” “This is the most transformative technology we’ve ever seen. If you’re not mentioning generative AI, that’s an issue.” - Justin Norden But the field has evolved rapidly. Six months ago, it was enough to have a slide on AI in a pitch deck. Today’s investors are savvier. They’re looking for tangible evidence of how companies leverage LLMs to deliver measurable value in terms of reduced costs, better outcomes, accelerated processes or improved customer relationships. Panelists recommended that CEOs focus on the problem their companies set out to solve and how technology like generative AI can help address it. “Given limited burn, budget and time, you need to think very specifically about how generative AI will make your product better and which problems can’t be solved without this technology,” Desjadon said. “It’s not enough to mention generative AI – you still need to solve a problem that people care about.” “It’s not enough to mention generative AI – you still need to solve a problem that people care about.” - Mike Desjadon Startups should lead with the business goal they’re trying to help achieve and not with the technology. “Rather than concentrating your company's AI strategy into one slide, I recommend spreading it out across different parts of your roadmap, to show that you understand the potential and range of the technology's application,” Moore suggested. Gen AI will transform healthcare data management In their closing remarks, the panelists unequivocally affirmed that generative AI and similar LLMs represent the future of data management in healthcare. These innovations are poised to act as indispensable co-pilots, empowering skilled clinicians and healthcare administrators to process information with unprecedented speed and precision. As the healthcare industry continues to evolve, the integration of generative AI and LLMs promises to be a driving force in achieving more efficient and effective healthcare solutions.
.webp)
