Just a feeling: Emirati’s research says AI can’t yet beat a doctor’s intuition
▶ Tuka Al Hanai and her colleagues at Massachusetts Institute of Technology find that in medicine, humans still have edge over machines, writes
It is the intuition that cannot be explained, nor should it be ignored. A new study by researchers in the US, among them Tuka Al Hanai, an Emirati PhD student at the Massachusetts Institute of Technology, has shed light on how gut feelings affect how many tests a doctor orders on a patient.
Ms Al Hanai and her colleagues developed an algorithm to carry out a “sentiment analysis” on a database of medical records from 60,000 intensive-care patients at a Boston hospital.
“We wanted to extract the emotive content in the doctors’ notes, whether that has additional information than the standard profile of patients,” says Ms Al Hanai, 29, from Abu Dhabi.
The study is part of a growing body of evidence and recognition of the role gut feeling plays in medical science. It comes at a time when there is much talk of the role artificial intelligence could play in medicine.
In what Ms Al Hanai calls “a really nice” result, researchers found a correlation between the positive or negative content of a doctor’s notes and the tests he or she requested.
If a doctor felt negatively about a patient’s condition more tests were ordered, although if the negative feelings went beyond a certain point, the number of tests declined.
The correlation was identified after the influence of the medical data for each patient was accounted for.
“The algorithm is telling us the doctor is seeing something more than is contained in the formal data,” says Ms Al Hanai, co-lead author of the study with Dr Mohammad Ghassemi.
Presented at a conference in July, the work was co-written by Dr Jesse Raffa and Prof Roger Mark of the institute, and Dr Shamim Nemati and Dr Falgun Chokshi of Emory University in Atlanta, Georgia.
Prof Mark, a medical doctor with decades of clinical experience, said he was not surprised that a doctor’s gut feelings were found to be important.
“Your clinical judgment is far better than machines and what you decide to do in a given situation is heavily influenced by your impression of the patient and what the patient’s wants and desires might be,” he says.
Other scientists have looked into this question. Among them is Dr Ann Van den Bruel, a general practitioner in Antwerp, Belgium, who was an associate professor at the University of Oxford. A doctor for 21 years, she has had gut feelings about patients’ illnesses.
“We were feeling it but not talking about it because it was unscientific – it wasn’t how we were trained,” Dr Van den Bruel says.
One study with which she was involved looked at the cases of 4,000 children in Flanders, Belgium.
Published in 2012, it considered whether going along with a gut feeling, defined as an “intuitive feeling that something was wrong even if the clinician was unsure why”, could make it less likely that a serious illness was missed.
In more than 3,000 of the children, a non-severe illness was identified, but some of the cases in which children turned out to have a serious infection were missed. Going along with gut feeling, the results indicated, could have prevented this from happening in some instances.
One key factor sparking gut feeling in doctors was a concern among parents that their child’s illness was abnormal.
Other research has concluded that gut feeling is useful, notably a study in Denmark in 2014, which found a doctor’s suspicions that a patient might have a serious illness could be as accurate a predictor of disease as scientifically recognised “alarm symptoms” such as blood in the urine.
These recent studies on gut feeling are happening at a time when the role that artificial intelligence could play in diagnosis is increasingly being recognised.
To give one of many examples, reports in May indicated that US and European researchers had found that neural networks were better than experienced dermatologists at identifying skin cancer when examining images.
But could using AI mean losing out on the benefits of gut feelings, given that it seems unlikely that a computer could have them?
Dr Van den Bruel says that AI will be able to replace doctors only “for some specific situations”.
“The first one will be radiology, where I think that for interpreting, reading an image, you don’t need contextual information. You just need accurate interpretation. They will be the first ones to be replaced by machines,” she says.
“There will be other situations where machines will contribute. The situations where gut feeling plays a role, it will be very difficult for AI.”
While it seems that AI will not be able to carry out all of the functions of a regular doctor, Ms Al Hanai, who is due to complete her PhD this year, thinks that its input can be improved if computers are taught “what the doctor is seeing and experiencing”.
“That’s an experiment or an avenue to explore, for example, to use cameras or microphones to begin to look at the same things the doctor is looking at, or hear the same things the doctor is hearing, so they can learn in part the same things as the doctor,” she says.