Playing with Fire: Artificial Intelligence at the Bedside

My patient held out their phone. “Do I need this?” they asked.

On the screen was a conversation with an AI chatbot about a recent lipid panel. Their LDL was only slightly above the laboratory reference range, but the chatbot had raised the possibility of apheresis, a specialized procedure that removes cholesterol from the blood and is generally reserved for rare, severe conditions.

This was not one of those conditions.

The recommendation made little clinical sense. Yet it was written fluently, confidently and in the language of expertise. My patient had not misunderstood it. They had trusted something that sounded trustworthy.

The moment was unsettling, but it was also more complicated than a simple story about technology getting something wrong. The chatbot had prompted my patient to look at their results, ask a question and bring that concern to me. In that sense, it had made them more engaged in their own care. The problem was not that they had used artificial intelligence. The problem was that the tool had offered a dramatic recommendation without the clinical context needed to interpret a mildly abnormal number.

That exchange captured the tension of AI in medicine for me. Using these tools can feel like handling fire. Fire can illuminate, warm and transform. It can also spread quickly and cause harm. The difference often lies in how carefully it is used.

As an internal medicine resident and an AI researcher, I see the promise firsthand. Physicians work through an enormous volume of information: laboratory results, imaging, medications, prior notes and subtle changes in a patient’s condition. AI may help organize that information, surface patterns, reduce repetitive work and give clinicians more time for the parts of medicine that require human presence: the conversation, the examination and the judgment call.

It may also help patients understand medical language, prepare questions and participate more actively in decisions about their health. That possibility should not be dismissed. Nonetheless, fluency is not judgment.

Still, looking away is not an option, and I would not want it to be. These tools are already in our patients’ pockets, already shaping how they understand their bodies, and already improving parts of medicine. Our responsibility is to understand them well enough to recognize both their strengths and their limits, and to speak openly with patients about how to use them safely.

I want to continue working at the forefront of clinical AI. The forefront should not mean running toward every bright new tool without caution. It should mean helping determine what the fire is good for, how it should be contained and where it might burn.

We should not be afraid of its light. We simply should not mistake its glow for truth.

---

References

• AI is Not a Magic Wand
• It’s a Little HAIzy: Understanding the Current Use of Artificial Intelligence in Scientific Research
• From Research to Clinic: Regulatory Frameworks for AI in Medicine
• Unveiling the Dangers of AI Through Dune’s Insights