OpenAI Podcast Episode 14 – Building AI for Better Healthcare

Date: March 16, 2026
Host: Andrew Mayne
Guests: Dr. Nate Gross (Head of Health, OpenAI), Karan Singhal (Lead, Health AI Research, OpenAI)

Episode Overview

This episode explores how OpenAI is developing AI systems—particularly ChatGPT Health—to improve healthcare for patients, clinicians, and the overall healthcare ecosystem. The discussion goes deep into the technical and ethical strategies for securing sensitive data, collaborating with medical professionals, and rigorously evaluating AI performance in high-stakes health scenarios. The hosts and guests also reflect on the transformative potential of health AI, notable feedback from real-world deployments, and their vision for accessible and effective healthcare technology.

Key Discussion Points and Insights

1. Backgrounds and Motivations in Health AI

• Nate Gross shares how his journey began with an interest in health policy and making healthcare accessible, later moving into technology after witnessing the profound contrast between outdated tools in medicine and proliferating digital consumer tech ([00:40]).
  • Quote: "Comparing the technology that we had as doctors...to what my friends had or what the patients had in the waiting room was pretty profound." – Nate Gross [01:15]
• Karan Singhal discusses his early fascination with the philosophy of mind, intelligence, and AI's impact on humanity, ultimately leading him to apply his expertise in safety and privacy to healthcare ([01:47]).
  • Quote: "The healthcare and clinical AI world was kind of not fully aware of that gap. And so I just thought it was a really amazing opportunity and responsibility to bring us there." – Karan Singhal [02:53]

2. OpenAI’s Vision for Healthcare

• OpenAI's mission—to ensure AGI benefits all of humanity—is especially resonant in healthcare, where fragmentation, missed care, and limited access are rampant ([03:10]).
• The aim is to move healthcare from a “reactive” to a more “proactive” model, leveraging AI to address gaps in care and empower both patients and clinicians ([03:10]).

3. Strategy and ChatGPT Health

• With around 900 million weekly ChatGPT users (40 million daily health-related queries), OpenAI’s strategy is both proactive (seizing responsibility for safe health AI) and reactive (meeting current demand) ([05:14]).
• Security and privacy are paramount: encrypted, “one-way valve” infrastructure ensures health conversations are never used to train models ([05:14]).
• Contextualization is prioritized—unlike “one-size-fits-all” search engines, ChatGPT Health enables users to share and ground queries in their personal health context ([05:14]).

4. Model Training and Evaluation for Healthcare

• The approach is “safety and grounding first,” not an afterthought ([07:06]).
• Healthbench: An exhaustive evaluation tool created alongside 250 physicians, simulating multi-turn conversations and evaluating 49,000 dimensions of performance ([07:06], [09:33]).
  • Quote: "Healthbench in particular actually measured around 49,000 different dimensions of performance. And that's just an example of one possible dimension of performance." – Karan Singhal [09:33]
• Models are trained and improved via thousands of physician-generated conversations, with detailed rubrics (~48,500 criteria) for assessing responses—from literacy adaptation to uncertainty handling ([11:18]).
  • Quote: "Healthcare is not multiple choice. Patients are coming in with a tremendous amount of complexities in their own stories and nuance and context..." – Nate Gross [12:25]
• Addressing uncertainty is key: models are now better at admitting when they don’t know, reducing harmful overconfidence ([13:20]).

5. Challenges and Blockers Ahead

• Access and Cost: Lower costs of intelligence drive wider adoption, but integrating diverse data sources and modalities (e.g., wearables) is challenging ([14:39]).
• Trustworthiness: Ensuring AI’s recommendations are grounded in the latest and region-specific medical knowledge ([18:10]).
• Interoperability: Healthcare’s patchwork of analog, digital, and decentralized systems makes unified AI integration difficult ([18:10]).
• Continuous Monitoring and Post-Deployment Evaluation: Deployment models include real-time monitoring (“AI Clinical Copilot” study in Nairobi led to measurable reductions in diagnostic/treatment errors) ([14:39], [16:30]).

6. Collaborative Development with the Healthcare Ecosystem

• Cooperation with clinicians and institutions is foundational—adopting national/international EHR standards, patient consent, and integrating with popular biosensors ([21:24]).
• Partnerships extend ChatGPT’s capabilities, enabling patients to more easily follow care plans, integrate health data into everyday decisions, and play an active role in their own health management ([23:34]).

7. Reducing Physician Burden and Expanding Capacity

• OpenAI aims to:
  • Raise the floor (expand access)
  • Sweep the floor (reduce administrative burden)
  • Raise the ceiling (empower advanced care and innovation) ([25:33])
  • Quote: "If there's one thing that healthcare professionals are short on, it's time." – Nate Gross [25:33]

8. Notable Moments and Real-World Impact

• Rapid, organic growth in health queries—health AI has become an indispensable tool for both professionals and patients ([26:52], [27:29]).
• AI has helped rediscover overlooked medications and treatments, demonstrating direct value in patient lives ([27:37]).
• Case study in Nairobi: Clinicians found it “dangerous” to not allow all professionals to use the AI tool due to the reduction in errors and improved outcomes ([28:52]).
  • Quote: "They actually felt that it was dangerous to have a group of clinicians not using the AI. And so that's the point at which I was like, wow, we have done something major here." – Karan Singhal [29:22]
• Patient and caregiver stories highlight AI’s role in solving “miracle cases,” helping with complex diagnoses, and assisting overloaded healthcare workers ([29:44]).

Memorable Quotes

• "Healthcare is not multiple choice. Patients are coming in with a tremendous amount of complexities in their own stories and nuance and context..." – Nate Gross [12:25]
• "I want everybody to have this protective effect. With health AI, there are these studies showing that if you have a doctor in your family, that adds a protective effect to your health as well." – Karan Singhal [15:50]
• "If there's one thing that healthcare professionals are short on, it's time." – Nate Gross [25:33]
• "They actually felt that it was dangerous to have a group of clinicians not using the AI. And so that's the point at which I was like, wow, we have done something major here." – Karan Singhal [29:22]

Timeline of Key Segments

• Backgrounds and Vision: [00:40]–[03:10]
• ChatGPT Health Strategy: [05:14]–[06:47]
• Model Evaluation & Healthbench: [07:06]–[11:18]
• Challenges and Solutions: [14:22]–[18:10]
• Ecosystem Collaboration & Interoperability: [21:05]–[23:34]
• Reducing Physician Burden: [25:18]–[26:37]
• Notable Moments & Real-World Feedback: [26:52]–[29:44]

Tone and Style

Throughout the episode, the hosts and guests maintain an informed, candid, and pragmatic tone. They blend technical detail with stories from the front lines of healthcare, consistently emphasizing collaboration, patient empowerment, and the ethical imperatives of health AI.

For Listeners:

This episode provides a comprehensive, honest look at the state of AI in healthcare—its promise, the complexities of real-world deployment, and the evolving partnership between technologists and clinicians. Whether you’re a patient, professional, or tech enthusiast, the insights here illuminate both how far we’ve come and the careful work still required to make safe, effective AI-driven healthcare a reality.