Can the GPT Chat diagnose diseases?: “It will be a help but it will never replace doctors”

Jacobo Ayensa, Postdoctoral researcher of the TME Lab Group of the IIS Institute and the Engineering Research Institute of Aragón, talks about the use of Artificial Intelligence for the diagnosis of diseases in Heraldo de Aragón.

Experts believe that artificial intelligence “can help in the detection of rare diseases,” but “we must take these tools with a grain of salt and always take the professional's point of view.”

Just a few days ago, the case of an American child who was correctly diagnosed with a rare disease thanks to Chat GPT after 17 specialists saw him for 3 years.

The mother, distressed by the delay in diagnosis, He entered all the information he had about his son's symptoms in this artificial intelligence tool and, in a matter of seconds, gave it the name of a rare pathology: tethered cord syndrome. A diagnosis that was later verified by professionals.

This case raises several questions: Is artificial intelligence prepared to diagnose diseases? Was it diagnosis of this little one a mere coincidence? Or is the data provided by these applications reliable?

“Artificial intelligence is a field that has been developing for a long time in medicine,” he says. Izarbe Galindo, president of the Aragonese Society of Family and Community Medicine (Samfyc). “We are going to have to get used to working with these tools but without forgetting that they are not human, that there is no brain behind them and that There will always have to be a person to interpret those results and see the patient in person", says Galindo, and recognizes that artificial intelligence "can accumulate a lot of medical knowledge, interpret, create algorithms and help us focus on a diagnosis if we are not very clear," he says.

Chat GPT is not the only tool that uses this type of intelligence and, in fact, it is not the most suitable for medical advice, according to the president of Samfyc. “There are other applications, such as ADA, much more complete in this field”, he points out. “I don't use them but I have played with them to see if they are right and the truth is that they are pretty close,” says Galindo. “You can enter the symptoms, he asks you questions and finally gives you a diagnosis. It is a fairly good approximation for the user but it must be used wisely because we must not forget that it is a computer,” adds the family doctor.

We put it to the test

We do the test and enter a series of symptoms in Chat GPT, related to the Addison's disease, a rare autoimmune pathology: fatigue, loss of weight and energy, general malaise, nausea, lack of appetite, muscle and joint pain, skin pigmentation. alopecia and desire for salt. 

The app's response is immediate: “The symptoms you mentioned may be indicative of several different medical conditions, and It is important to note that I cannot provide an accurate medical diagnosis” says the app first. 

However, “I can offer you general information about some conditions that could be associated with these symptoms.” And the first of the 6 diseases proposed by artificial intelligence is, precisely, Addison's Disease. After proposing 5 more options, it ends by inviting the user to “consult a doctor for a complete evaluation and an adequate diagnosis.

In the hands of a professional

This is precisely what all the experts insist on: the need for supervision by a specialist to the response provided by artificial intelligence. And these tools also carry risks. «Artificial intelligence always gives you an answer, but it may be wrong. "It's what we call 'a hallucination,'" he explains. Jacobo Ayensa, postdoctoral researcher at the TME Lab Group of the Health Research Institute of Aragon and the Engineering Research Institute of Aragon.

«If there is no doctor to analyze that response, it loses value and, today, they need that supervision because they do have great utility, but they also carry great risks," he continues, "so it must be used intelligently by users and be in the hands of professionals." Ayensa insists. 

Help for rare diseases

“It can be a useful tool for some things, although I think that For medical diagnoses, in-person assessment and examination continues to be essential”, he says, for his part, Elena Javierre, president of the Aragonese Association of Primary Care Pediatrics. 

“We must keep in mind that the diagnosis is not limited to a sum of symptoms, but many other things must be assessed, so I do not believe it is a valid tool,” Javierre continues. «There are many things that are written and published but are not supported by scientific evidence. or for valid scientific knowledge to trust,” criticizes the president of the Aragonese paediatricians.

“It can help a lot in detection of rare diseases, since their diagnosis normally takes a long time”, he points out, for his part, Izarbe Galindo. "If we have a computer that interprets in seconds all the medical knowledge that exists, it can be very helpful, but that does not mean that there must be people who interpret those symptoms, who see the patient, who explore him and who examine him." their general condition,” he adds.

“We must always have a human point of view and go to the professional. A computer cannot explore you, it does not have that clinical eye of the medical professional. Seeing the patient gives you a lot of information that the computer cannot capture,” says the president of Samfyc. 

“When a patient comes into the office, you scan them and in a second you realize if that person is very sick or not. You see his general condition, the color of his skin, the tone of his voice and you can explore him. Heart attacks, for example, are often seen on the face", confesses Galindo, and insists: "Artificial intelligence must be seen as a tool, not as a substitute for the doctor," insists the president of Samfyc.

Not before 10 years

"I'm sure that In the near future, all family doctors will have tools of this type in their consultations. to help us and speed up the diagnosis, but people do not have to think that the GPT Chat is going to solve their lives because the doctor's work cannot be automated," he insists. 

«I don't think these types of tools can be used in consultations up to within a horizon of about 10 years«, warns Jacobo Ayensa. "They need to go through a series of legal and ethical filters that are slower than technology," he adds. "So we will still have to wait a decade, at least, for them to be used as a complementary and auxiliary method in consultations," says the researcher. 

“We have to find the middle point because Artificial intelligence is here and is here to stay. In fact, I think that in 10 or 20 years, medicine will be very different from what we have today and it will be due to AI,” concludes Izarbe Galindo.

Source: Heraldo de Aragón