- The Cautionary
- Posts
- Learning AI #30
Bethany Crystal woke up from a nap and found red spots all over her legs. She ran the symptoms by ChatGPT. It told her she needed immediate evaluation for possible bleeding risk.
She went to the ER and doctors diagnosed her with a rare disorder that drops your platelet count and can cause you to bleed internally without warning. Crystal has said she isn't sure if she would have gone to the emergency room in time without the AI pushing her.
A woman named Lauren Bannon spent four months being told she had rheumatoid arthritis, despite testing negative for it. Stomach pain and a 14-pound weight loss got waved off as acid reflux. She asked ChatGPT what else could cause her symptoms, and it suggested Hashimoto's disease, a condition that attacks the thyroid.
Her doctor was skeptical. She pushed for the test anyway. It came back positive, and the follow-up ultrasound found two small lumps in her neck. Thyroid cancer.
Neither of these is a fluke. Hundreds of millions of people now ask ChatGPT for health advice every week, and a growing number of doctors use it too. There's a reason for that.
There's an old line in medicine: if you hear hoofbeats, think horses, not zebras.
Most symptoms have a boring, common explanation. A doctor with 15 minutes and a packed schedule is trained to find the horse and move on.
AI doesn't get tired, doesn't have a waiting room backing up, and doesn't stop asking questions after the obvious answer fails to fit. It can evaluate your symptoms for as long as you're willing to type.
Doctors are good at horses. They don’t have time to be curious.
Then there are the problems. A 60-year-old man asked ChatGPT how to cut sodium from his diet and it steered him toward sodium bromide, an industrial chemical, not sodium chloride, the food seasoning we know as salt. He ended up hospitalized with hallucinations and paranoia.
AI is a good second opinion or source of information for a conversation with your medical provider. It is not a decision maker.
Same tool. One case finds a cancer doctors missed. Another almost kills someone over table salt.
You don't need to trust AI for medical advice. You don't need to distrust it either. You need to use it the way the people in the good stories used it: as a second opinion generator, not a doctor.
If AI flags something, ask your doctor to test for it. It costs you one conversation and a blood draw. If it's wrong, you've lost twenty minutes. If it's right, you've caught something that could save your life.
What you should not do is take an AI’s answer and act on it directly, especially anything involving medication, dosage, or a chemical going into your body. The people those failures happened to weren't stupid. They just skipped the step where a human with a license checks the work.
This is the pattern for AI everywhere, not just medicine. It has infinite patience and zero accountability. Bring its output to someone who is qualified and accountable.
Things I think about
The US adds about one person to its population every 18 seconds.
*****