Generating AI’s Potential

In healthcare—and more precisely, in the field of ophthalmology—the use of generative AI may bring some important benefits to everyday clinical practice. However, there is a flip side to consider. A panel of key opinion leaders in ophthalmology discussed its many aspects from the perspective of efficiency and effectiveness, as well as ethical and legal concerns.

“What we have to keep in mind is what is the data we train the system with,” Nino Hirnschall MD, PhD said. “It must be the absolute truth, or we will face some problems in the future.”

According to Dr Hirnschall, as human beings suffer the cognitive bias of the Dunning–Kruger effect, so too does AI. As a black box, he observed, it is very difficult to trace the way it processes information, and the technology always considers its output as the absolute truth, despite its sensitivity to limitations such as culture, language, and computational mistakes.

“If you talk to clinicians, they will tell you that patients would never find AI replacing their doctor as a positive thing. However, large language models are shown to increase patient satisfaction—with a better bedside manner than the average doctor,” Sorcha Ní Dhubhghaill MBBCh, PhD observed.

“The real problem of AI in healthcare is sycophancy, because these programs are built to please the human user,” she added, considering the scenario where a patient might be more likely to seek a truth that pleases them rather than one that might consider the harsh reality of a diagnosis and treatment.

However, as underlined by Ernest Lim MBBS, PhD, there are some applications of generative AI in healthcare that might increase efficiency without the risk of sycophancy or mistakes. After four years of research, Dr Lim created a conversational AI assistant for the sole purpose of automating preoperative checks and collecting patient surgical outcomes, and he plans to expand it on generative AI, with a pinch of caution.

“There is a lot of amazing potential, but there’s a lot that can go wrong,” Dr Lim admitted.

While considering ethics and legality, Mor Dickman MD, PhD observed generative AI represents a constant breach of intellectual propriety and copyright, something that demands serious attention.

“AI principles are directly in [opposition to] the principles of data protection expressed in the GDPR, in complete violation of many basic human rights,” he stressed.

Such an oversight, according to Dr Dickman, will soon be tackled by the European Union AI Act, which will provide a single, harmonious system for European use that will directly impact the world of ophthalmology.

However, such a standpoint from the EU already dramatically differs from some of the ways of thinking expressed in other countries. As Bruce Allan MD observed, the cautionary angle of the EU stands in stark contrast to the more disruptive position of the US and its stakeholders.

So, what is AI for? Dr Ní Dhubhghaill asked. Agentic AI programs make preliminary medical decisions now considered an effortless way to cut personnel costs, demonstrating the technology’s social complexities. Still, Dr Hirnschall said the power of a tool providing helpful indications to the physician is undeniable. Nic Reus MD, PhD, noted that some programs are already capable of making very precise diagnoses, such as spotting signs and symptoms of diabetic retinopathy previously undetectable by the human eye, and this may represent a real opportunity to improve health in patients.

AI technology, however, comes at a great cost. According to Dr Ní Dhubhghaill, the consumption from data centres in Ireland jumped from 5% of the overall national energy consumption in 2015 up to 21% in 2023. As she observed, the energy consumed to create an image in DALL-E mini, the least advanced generative AI, would charge a cell phone for a day. A video generated with AI consumes the energy equal to more than half a million views of a single episode of Friends on Netflix.

“The data consumption is enormous, and we must think very carefully about the environmental cost of this technology. Who is really paying for this?” she asked.

Dr Allan said only the future will tell. “Hopefully, we are not spiralling downwards. I am cautiously hoping for more efficient and less power-hungry models, for a golden period within a data revolution that will make our lives as doctors better,” he concluded.

The presenters spoke during the Generative AI in Ophthalmology TOGA Session at the 2025 ESCRS Winter Meeting in Athens.

Bruce Allan MD, FRCS is consultant ophthalmic surgeon at Moorfields Eye Hospital and Professor of Anterior Segment and Refractive Surgery at University College London, UK. bruce.allan@ucl.ac.uk

Nino Hirnschall MD, PhD (med Physics), MBA, FEBO is senior surgeon at the Kepler University Clinic, Linz, Austria. nino@hirnschall.at

Sorcha Ní Dhubhghaill MBBCh, PhD, MRCSI(Ophth), FEBO, FEBOS-CR, BaO, Dip(stats) is chair and head of the Department of Ophthalmology at University Hospital Brussels, Belgium. nidhubhs@gmail.com

Mor Dickman MD, PhD is Professor of Ophthalmology, University Eye Clinic, Maastricht University, Netherlands. mor.dickman@mumc.nl

Ernest Lim MBBS, BSc(Hons), PhD is an ophthalmic surgeon based in London and Science Director of Ufonia, UK.

Nic J Reus MD, PhD is an independent ophthalmologist at Amphia Hospital, Breda, Netherlands. nreus@amphia.nl