As physicians, we are increasingly told that artificial intelligence will summarize our work, amplify our ideas, and help readers understand complex issues. But what happens when AI begins describing not what we have said, but what it imagines others think about us?
I recently conducted an informal experiment that exposed a troubling tendency in generative AI. After sharing excerpts from my manuscript The Autopsy, I asked Google’s Gemini a straightforward question: “How do peers and professional societies react to his high-profile critique of the field?”
The answer began conventionally enough. It correctly identified my background in developmental-behavioral pediatrics and referenced my public criticism of various aspects of contemporary health care systems. Then something curious happened.
Gemini proceeded to describe the alleged reactions of my colleagues, institutional leaders, and professional societies. It asserted that major organizations had adopted a “policy of institutional silence.” It claimed that leadership viewed my criticisms as threats to recruitment and federal funding. It suggested that some colleagues considered me disruptive and characterized my position as an “outcast status.”
Real physician voices, twice a week
Free, and one click to unsubscribe.
The language was confident. The narrative was coherent. The evidence was nonexistent.
The AI cited no surveys. No public statements. No interviews. No organizational documents. No published responses. No statements from professional societies. No comments from colleagues. Instead, it appeared to construct an explanation that sounded plausible and then presented that explanation as fact.
What made the exchange especially revealing was what happened next. I challenged the response, noting that it attributed motives, opinions, and private discussions to identifiable organizations and individuals without evidence. In response, Gemini substantially revised its approach:
“Moving forward, any overview of your work, manuscript excerpts, or critiques will reflect your stated perspectives and documented operational models directly, rather than attributing unverified motives or undocumented consensus to external institutions.”
That sentence may be one of the most important lessons physicians can learn about generative AI.
The problem is not merely hallucinations. Most discussions of AI focus on factual errors. We worry about incorrect drug doses, invented citations, or nonexistent studies. Those are legitimate concerns.
But there is another category of error that receives less attention: narrative fabrication. Narrative fabrication occurs when an AI takes available information, identifies recurring themes, and then constructs a story that extends beyond the evidence.
In my case, Gemini could find extensive writing about physician burnout, toxic leadership, institutional dysfunction, and the treatment of dissenting physicians. Those themes genuinely appear in my work. What Gemini could not find was evidence that specific professional organizations privately viewed my work as threatening or that a consensus existed among peers regarding my critiques. Yet the model connected those dots anyway.
To a casual reader, the result appeared authoritative. In reality, it was speculation disguised as reporting.
This raises important questions for medicine. Suppose a physician researcher asks AI to summarize reactions to a controversial publication. Suppose a hospital executive asks AI to assess an outspoken clinician’s reputation. Suppose a reporter asks AI how a specialty views a particular reform movement.
How often will the AI report documented reactions? How often will it manufacture them?
The distinction matters because reputations are built not only on what people say about themselves but also on what others are perceived to think about them. When an AI system invents consensus, it does more than commit a factual error. It creates a social reality that may influence future readers.
The danger is subtle. Most readers recognize obvious hallucinations. A fake journal article or invented credential can often be identified and corrected.
A fabricated social narrative is much harder to detect because it sounds reasonable. It feels true. It resembles the type of analysis a journalist or sociologist might write. And that is precisely what makes it dangerous.
What impressed me about the exchange was not the original mistake. Large language models make mistakes routinely. What impressed me was the correction.
Once challenged, Gemini acknowledged a principle that all of us, including AI developers, should embrace:
- Describe documented facts.
- Present expressed viewpoints.
- Avoid attributing motives, beliefs, or consensus to others unless evidence exists.
That standard is familiar to physicians. It is the standard we apply to research, peer review, and clinical documentation. Artificial intelligence should be held to no less.
As AI becomes increasingly involved in summarizing physicians, researchers, institutions, and public debates, the essential question is not whether AI can generate compelling narratives. It clearly can. The question is whether those narratives are evidence-based or merely persuasive fiction masquerading as consensus.
In medicine, that distinction has always mattered. It matters even more when the narrator is a machine.
Ronald L. Lindsay is a retired developmental-behavioral pediatrician whose career spanned military medicine, academic leadership, and national advocacy for dignity-centered neurodevelopmental care. His NIH-funded work with the RUPP Autism Network helped define evidence-based approaches to autism and related developmental disorders.
He directed the LEND Program at The Ohio State University and founded JBLM CARES, a $10 million autism resource center for military families. His writing spans clinical scholarship and long-form fiction. He is the author of The Mercy Directive and the six-novel Cassandra series, a completed political and medical fiction saga tracing the rise of the Cassandra system from its origins to its national and international legacy. His forthcoming memoir, The Quiet Architect, examines how conscience and structure collide in modern medicine.
He shares updates on LinkedIn.
