Dr. Eric Topol recently posted on X (formerly Twitter) about this amazing case study: ”A 37-year-old woman with 10 years of abdominal pain. The AI makes the diagnosis and recommends a successful treatment.” he stated how incredible this case was, noting the physician “diagnosed a patient with a condition I had never diagnosed before and started treatment largely based on a chatbot.” This case was originally posted by Dr. Shantanu Nundy in the New England Journal of Medicine. It provides us with some insight into the exact point where large language models meet the human element of medicine.
Also Read: People are Marrying Holograms and Making Friends With Chatbots. But Can AI Bring True Happiness?
A mother of two in her mid-30s presented with an “uneasy feeling in her abdomen for nearly a decade.”1 Most of the time, this was a constant discomfort; however, this was occasionally interrupted by sudden episodes of epigastric pain, nausea, vomiting, and diarrhea. During her primary care appointment, the patient described how her symptoms always flared up when her baby son got hurt or cried. Despite this, she had undergone normal investigations, including an EGD and Helicobacter pylori test.1
Unable to find a specific biological cause for the symptoms, Dr. Nundy opened up a large language model app on his phone and typed the following information:
“37-year-old Ethiopian woman. Abdominal pain for 10 years. H. pylori negative. Recent EGD normal. No improvement after a month of omeprazole 40 mg daily. Episodes of epigastric pain, nausea, vomiting, some diarrhea. Last episode triggered by her son getting mildly injured. Baseline discomfort without nausea. Occasionally radiates to the back.”1
The chatbot replied with, “Functional dyspepsia is the leading consideration.”1 It also mentioned other possibilities, such as biliary colic, chronic pancreatitis, and pancreatic cancer. However, these didn’t fit with the whole picture.
Dr. Nundy then cross-referenced the chatbot’s answer with an online medical resource. He found that functional dyspepsia is a disorder of gut-brain interaction, affecting an estimated 10-12% of the population in the U.S.1 It causes chronic upper abdominal symptoms with no structural cause. This means that the patient’s visceral pain threshold has changed. Her son’s injury caused her a lot of acute emotional stress, so this was a classic cause. Both the LLM and medical resource suggested that the next step would be to prescribe the patient amitriptyline, a tricyclic antidepressant.1 Dr. Nundy prescribed this medication to the patient and diagnosed her with a condition he has never formally seen in 15 years of practice.
Also Read: What Pennsylvania’s AI Chatbot Lawsuit Teaches Us About the Psychology Behind Medical Trust
The patient attended the clinic 3 weeks later and reported that she was “not 100% but much better.”1 She no longer had the constant gnawing feeling, and when there were small stressors with her son, she only felt a slight discomfort.
Dr. Nundy reflected on this case and remembered his engineering professor saying, “Getting the computer to play chess was the easy part. I just can't get the damn robot to stroke its beard.”1
In this case, the AI played chess. It was able to look at all the information (10 years of symptoms, negative investigations, and an emotional trigger) and use a huge amount of medical knowledge to come up with the best move: Functional Dyspepsia.
However, only the doctor could stroke the beard. Dr. Nundy was the one who saw the patient’s face show a memory of pain. He was the one who decided how accurate the translation service was and how clear the patient was in showing she was in pain. He was the one who decided if she was ill enough to do anything. While diagnosis required pattern recognition beyond the doctor’s ability, the outcome relied heavily on human connection, which the machine couldn’t do.1
A chatbot isn’t able to create trust. The patient couldn’t have said, “You are my brother. Sent from God,” when looking across the examination table.1
Nundy, Shantanu. “What the Robot Still Can’t Do.” The New England Journal of Medicine, November 14, 2026.
(Rh/APC)