Is AI a Future Frenemy?

Artificial intelligence is becoming increasingly precise, offering ophthalmologists powerful tools for diagnosis and imaging. However, to maintain the best quality of care, some boundaries need to be established.

According to Luis Abegão Pinto MD, PhD, some learning machines already outperform the physician’s eye—for example, pattern recognition in glaucoma, influencing diagnosis.

“Something that took you 10 or 15 years to train yourself to look at in the optic disc can compress to a couple of hours in an AI with the proper algorithm. This is the game changer,” he admitted. “We spent a lot of time and effort going to medical school, studying ophthalmology and so on, and then suddenly we must deal with this. It is challenging.”

One of the primary reasons for defining the relationship between the doctor and AI is the patient. As Dr Pinto underlined, research shows patients are primarily concerned with proper diagnosis.

Missing a diagnosis is a shared concern among doctors and patients, Dr Pinto observed, as AI is already good at prediction and extrapolation—and getting better. But AI is not always right, since it still guesses, he said. It is a trash in, trash out scenario, meaning if taught the wrong algorithm, the program will continue to make the same mistake. For example, if the algorithm is trained on the wrong data sets or searches for something it was not meant to recognise, it will derail and fail—a type of daily occurrence today, he said.

This balance between benefit and concern often leaves ophthalmologists and diagnosticians feeling uneasy because AI is presented as a tool capable of doing such tasks without their supervision.

However, AI continues to improve at an amazing speed. In just 35 years, mobile phones have transformed from massive telephones into portable devices with more computing power than the machines that put humans on the moon in 1969, he observed. It is nearly impossible to predict life in 25 years, yet some informed guesses can be made, he added.

Dr Pinto suggested an affordable, easily accessible AI could fill the gap of undiagnosed patients with support from policymakers to build the infrastructure for easier screenings—not only for glaucoma, but for a variety of conditions that can be diagnosed in a single test, such as glaucoma and diabetic retinopathy. Fundus pictures, for example, are already used to diagnose Alzheimer’s disease and cardiovascular conditions, he explained.

If properly trained, these future algorithms could likely use the baseline exam to make an excellent risk profile assessment, allowing the physician to focus on the cases likely to progress while reassuring those expected to remain stable.

“We will have identified the most effective treatment—faster, with [fewer] exams and more tailored [than] ever—rather than through trial and error,” he concluded. “But don’t worry. In the words of American cardiologist Eric Topol: ‘AI won’t replace doctors, but doctors who don’t use AI will be replaced by those who do.’”

Dr Pinto presented at the 2025 ESCRS Annual Congress in Copenhagen.

Luis Abegão Pinto MD, PhD, FEBOS-G is Head of the Glaucoma Clinic of the Department of Ophthalmology at the Centro Hospitalar Lisboa Norte, Portugal. abegao.pinto@ulssm.min-saude.pt