AI’s Next Movement

It feels slightly strange looking back at the piece now, because the three years since I wrote it have covered the duration of my PhD. In that time, AI has moved quickly. In ophthalmology, the discussion has shifted from relatively narrow image-based tools, such as diabetic retinopathy detection or OCT segmentation, to broader systems that can draw on images, clinical notes, reports, and patient-facing language.

I still think the main tension in the essay holds. We are very good at producing impressive retrospective studies, but much less good at showing these tools work safely and usefully in real clinics. The difficult questions have not gone away, like how we validate AI, how we avoid bias, how transparent these systems need to be, who is responsible when they are wrong, and how patients are meant to trust them.

But I am optimistic about AI in ophthalmology; perhaps more cautious than excited. The specialty is clearly well suited to it because we rely so heavily on imaging and longitudinal data. But I don’t think the best version of this future is one where machines replace ophthalmologists. The real value will probably be in less dramatic tools that reduce administrative tasks, help detect disease earlier, support triage, and give clinicians better information while keeping the doctor–patient relationship intact.

To read the original 2023 John Henahan Writing Prize-winning essay, click here.