OpenAI and Karan Singhal: ChatGPT Aims to Revolutionize Healthcare

OpenAI and Karan Singhal: An Ambition for Health

Karan Singhal, a principal researcher at OpenAI, is leading an ambitious initiative to transform ChatGPT into a major player in health, comparable to Google in its scope. Currently, ChatGPT is already consulted by over 230 million users each week for health and wellness advice.

In an interview with Business Insider, Singhal shared his efforts to integrate medical context into ChatGPT. OpenAI is increasingly investing in health research, responding to a growing demand from users who turn to ChatGPT for urgent medical questions.

Singhal's goal is to make ChatGPT sufficiently effective to improve users' lives, prevent health disasters, and convince skeptics of its utility. He is already observing a shift in patient behavior, as they begin to see OpenAI's model as an ally in their care journey.

A Focus on Health at OpenAI

Before joining OpenAI, Karan Singhal earned a reputation as a researcher at Google, where he contributed to the development of Med-PaLM, a series of AI models specialized in medical questions. However, Google has since reduced its investments in these models, preferring to focus on general-purpose AI models.

In 2024, when Singhal joined OpenAI, the GPT-4o model was the company's flagship product. This model faced criticism and legal challenges for providing harmful advice, particularly in mental health. OpenAI has denied any responsibility in these cases, which are still ongoing.

Despite these challenges, OpenAI has continued to explore AI applications in the health sector. Singhal felt a strong "responsibility" to improve the quality of medical responses provided by OpenAI's models. He quickly formed a new team of health researchers and established partnerships with over 200 doctors to leverage their collective expertise.

A year after his arrival, Singhal contributed to the launch of HealthBench, a series of assessments developed with physicians to measure the performance of AI systems in health. "Once you know how to evaluate performance, it becomes much easier to improve it," he explained.

Advances in AI in the Medical Field

The recently launched GPT-5.5 Instant model by OpenAI has outperformed responses provided by doctors and the GPT-4o model in comparative tests, according to an announcement from the company. By analyzing billions of anonymized health messages, OpenAI has observed a 71% reduction in inaccurate responses over the past two months.

There is constant pressure to maintain these improvements, both for patients and healthcare professionals using these tools. Singhal noted that doctors are quickly adopting tools like ChatGPT for Clinicians, and he does not perceive any reluctance from hospitals and clinics regarding AI. "The adoption of technology in healthcare is actually incredibly fast," he stated.

Towards a Better Understanding of Patients

For a long time, Google Search has dominated the online medical information sector, redirecting users to sites like WebMD. Singhal sees chatbots as an improvement, as they offer more personalized advice through conversational interactions.

One of the main challenges in providing accurate medical information via a chatbot lies in the lack of knowledge about the patient. Unlike a doctor who can access your medical records or know you personally, a chatbot must simulate these interactions.

In January, OpenAI announced a health product integrated into ChatGPT, allowing users to connect health applications and upload their medical records. Singhal illustrated this by sharing how he imported his sleep data from his Apple Watch, enabling the application to analyze this data and discover that he was lacking deep sleep due to a room that was too warm.

ChatGPT Health, OpenAI's health product, still has a waiting list more than five months after its launch. Singhal's team prioritizes ensuring that the AI model can seek additional information, much like a doctor asking questions to provide accurate answers.

The team also aims to demonstrate that AI can be beneficial for everyone in health, not just for "advanced users." "People's adoption will only progress at the speed of their readiness in practice, and you need to guide them towards that, especially as technology improves," Singhal said.