Should You Trust a Chat AI for Medical Diagnosis?

AI is the hot new technology these days.

It’s been around long enough that the new is kind of worn off.     There’s still a lot of hype about what people have done with AI. There’s a lot of claims about what AI is capable of doing.  It’s rather staggering how many AI services and products are currently available. And how quickly the average person has taken to using them.

There’s already a lot of AI use in just about every kind of industry.   From programming, Investing. Literature, Visual arts, movies, t.v., In research papers; It’s even been used in researching and writing legal briefs.

I have seen on  different forums where multiple individuals said they put their Loved One’s symptoms into an AI and asked  for a diagnosis.   That concerns me.

An AI answer is not a medical diagnosis.  

 The results  are not something trustworthy enough to make life decisions. Until you have a doctor do an examination, use reliable testing, take a reliable history for a proper diagnosis,  you can’t actually know what is going on.   The AI that is available for general public use is not capable of making a diagnosis.

Headlines claim  AI can more accurately diagnose cancer than actual doctors, BUT. . .

  There is a big difference between what researchers used to claim those more accurate results, and what’s available for the average person to use.   It is like the difference between having a  dry cleaners clean the custom draperies or using a home washer and dryer.  Both will produce a result but, IF you can cram the draperies in the  washer and then the dryer, the results are not even close — permanently wrinkled, shrunken and don’t fit.   Dry cleaners are specially equipped for a specific type of cleaning, needs trained personnel, to do it faster and ensure perfect results.   

The data  medical AI are trained on is a lot more narrow.

For lung cancer, as an example,  researchers deal with lung x-rays and other imaging, and decades of specific medical records about lung cancer.  Rigorous attention is given to the information provided to the AI.  ChatgGPT is trained on everything “it”  can get their hands on.   Including fiction. The issue is more than just what training data is used, however.

It’s how AI produces results.

 AI doesn’t actually “produce” anything.   Their results are more like crazy quilts assembled from a large collection of what other people have written in the past. AI looks at the prompt provided for clues, then spits out text that occurs at  the highest percentages of that particular information occurring in documents about that  topic.  Including wrong and incorrect information. 

 AIs don’t – and can’t-  check the accuracy of the information is uses to answer questions.  AI cannot recognize if A is true and B is true, then C cannot possibly also be true.  Yet they pass that misinformation as true in the results.

AIs lack some very important resources to make a full diagnosis.

Starting with patient medical records and history.  AI also lacks results of medical labs to rule out other possibilities and other factors.  AI does not consider the patient in the “natural environment” – how they live their life.  A lot of neurological symptoms can simply be the result of dehydration.  Or sleep deprivation. Others may be the result of medications they’re currently taking.   

. A diagnosis also requires a process of ruling out other possibilities and being truly certain it’s not another issue. Without medical labs, CAT scans and X rays, and a whole host of other investigative procedures,  a general AI just is not equipped to make a diagnosis.

Filtering through all of these possibilities and information, requires actually thinking.

It requires logic and reason and understanding. Asking questions.  

Always consult a doctor. An actual trained, certified and capable of thinking doctor.