Blogs

Insights from with Yindalon Aphinyanaphongs, MD, PhD, Head of Applied AI at NYU Langone Health

Tell us about NYU Langone Health and your role there.

I wear two hats at NYU Langone Health. On the research side, I'm a research professor in Population Health and Medicine, where my lab focuses on novel applications of artificial intelligence (AI) and large language models in healthcare, with a particular interest in reliability, stability, benchmarking, and model performance.

On the operational side, I serve as Head of Applied AI. My team is responsible for deploying, scaling, and evaluating AI technologies across the health system. Our goal is simple: deliver meaningful value to the organization through responsible and effective use of AI.

NYU has been investing in this work for nearly a decade. That commitment from leadership, combined with strong technical expertise, high performance computing, and deep integration with clinical workflows, has allowed us to move AI projects from concept to deployment in a way that creates value.

What do health systems need to get right as they build and deploy AI?

One of the biggest misconceptions is that AI success is primarily about the model itself. In reality, the technology is only one piece of the equation. Organizations need a clear understanding of the problem they are trying to solve and how AI fits into that strategy. Even the best-performing model won't create value if it isn't integrated into clinical workflows or if end users don't know how to act on its outputs. In some instances, AI may not even be the best solution and bringing the right technical experts to a team from across the organization can make sure even the choice to use AI is considered thoughtfully.

I've seen organizations invest heavily in AI solutions only to struggle with adoption because clinicians have to leave their normal workflow to access the tool, or because implementation planning wasn't fully thought through. The human systems around AI are as important as the technology itself. Health systems should be thinking not only about acquiring AI, but also about how people will use it, trust it, and incorporate it into day-to-day decision-making. Baked into this statement is relying on existing processes to routinely hear from your stakeholders on the impact of the AI based tool.

How does NYU Langone approach AI governance?

At its core, governance is about ensuring that AI is safe, effective, and delivering meaningful value to patients and providers.

Our process starts with understanding the level of risk associated with a particular project. From there, we bring together the right stakeholders across technical, clinical, operational, legal, and compliance teams to evaluate the opportunity.

One lesson we've learned is that governance cannot be owned by a single group. Successful governance requires representation from everyone involved in the lifecycle of an AI solution, from model developers and technical teams to clinicians and frontline users.

Equally important is maintaining strong documentation and creating clear pathways for ongoing monitoring. Governance doesn't end when a model is deployed. Organizations need processes in place to evaluate performance over time and ensure solutions continue to deliver value as workflows and environments evolve.

A nice side benefit is that the documentation creates an inventory with project meta-data that allows easy identification of projects that may be relevant to future regulatory questions.

What AI initiatives are you most excited about right now?

One area I'm particularly excited about is patient-facing AI. At NYU Langone, we've been deploying patient-friendly radiology reports and are expanding similar approaches into other areas, including discharge summaries and cardiac imaging reports. The goal is simple: help patients better understand their health information.

As clinicians, we often assume patients leave an encounter with a clear understanding of what happened and what they need to do next. In reality, that isn't always the case. AI gives us an opportunity to translate complex medical information into language that is more accessible and actionable.

I believe helping patients better understand their care may ultimately be one of the most meaningful applications of AI in healthcare.

How did NYU Langone become involved with CHAI?

As AI adoption accelerated across healthcare, there was a clear need for organizations to come together around shared standards, best practices, and responsible implementation.

What attracted us to CHAI was its ability to bring together stakeholders from across the healthcare ecosystem, including health systems, academics, startups, technology companies, and policymakers. That's a difficult undertaking, but it's also what makes the organization valuable.

Healthcare organizations are often solving similar challenges independently. CHAI creates an opportunity to learn from one another, share ideas, and develop common approaches to responsible AI deployment. For organizations navigating a rapidly evolving landscape, that type of collaboration is incredibly important.

What value have you found in participating in CHAI's community and working groups?

One of the biggest benefits has been the opportunity to connect with peers facing many of the same challenges. When we first started building AI capabilities years ago, there were very few opportunities to learn from others. Today, through communities like CHAI, organizations can exchange ideas, share frameworks, and accelerate progress without everyone having to start from scratch.

I think that's one of the most exciting developments in healthcare AI. We're beginning to reach a point where solutions, lessons learned, and best practices can be shared across institutions. That creates tremendous opportunities for collaboration and ultimately helps the entire industry move forward more quickly and responsibly.

Looking ahead, where do you hope to see the greatest progress in AI and healthcare?

I'm particularly excited about two areas: access and safety.

AI has the potential to expand access to healthcare in ways we've never seen before. Intelligent assistants and care-support tools could provide patients with personalized guidance and support far beyond what is possible today. That could be especially meaningful for patients in underserved communities or areas with limited access to specialists.

I'm also optimistic about AI's ability to improve patient safety. As models become more capable, they will increasingly help identify potential errors, recognize clinical risks, and support providers in making decisions.

While there's still significant research and validation work ahead, I believe we're moving toward a future where AI becomes a trusted layer of support for both patients and clinicians. If we get it right, healthcare can become more accessible, more efficient, more impactful, and ultimately safer for everyone.