Download PDF
I read somewhere that a good way to connect to a middle school–aged son is to play video games together. So, when my son was in seventh grade, I tried playing Skyrim with him. This lasted about 8 minutes because the ophthalmic surgeon couldn’t navigate the handheld controller. Things went a little better when, on a road trip, I asked him to teach me about rap music. Ten hours later, I could identify songs by Chance the Rapper and Jay-Z.
So, I was prepared to appreciate the ChatGPT-generated rap lyrics posted by Baochan Nguyen, a comprehensive ophthalmologist in Newport News, Virginia, in a private Facebook group for ophthalmologists who are moms.
The lens, it flexes, like a zoom on display,
Adjusting focus, for both night and day,
Cataracts might come, but we’ll take ’em away,
With a surgical touch, we’ll make the clouds sway.
AI can write clever, rhyming stanzas, but its promise goes well beyond this. AI may be the innovative disruption that ophthalmologists will need in order to care for the predicted crush of patients with eye disease expected in the decades ahead. If we open our doors to AI, though, our practice models will probably need to change. For example, several AI-based diagnostic screening systems for diabetic retinopathy have been approved by the FDA. Perhaps patients with diabetes will get their annual screening for retinopathy at their primary care provider’s office—when they get their HbA1c checked. In some respects, getting an AI-reviewed retinal photograph from a non-mydriatic camera in the primary care office is like screening kindergarteners for decreased vision. It’s not as comprehensive as an exam by an ophthalmologist, but it has the potential to screen more patients, identify those who need specialty care, and free us up to spend more time treating patients.
Glaucoma specialists imagine the day when AI will integrate data from visual fields, OCTs, and photos to predict fast versus slow progressors. Target pressures and surgical interventions could be tailored so that slower progressors have less aggressive care and fast progressors get earlier, more aggressive care. With AI, we could shift resources to patients who are most likely to be impacted by vision loss. Our reimbursement models and patient flow models will need to change, too. It isn’t clear how AI will be reimbursed within the current fee-for-service system.
AI’s impact on health care is now accelerating—and not just for improving care. For example, machine-learning systems that support diagnostic workflows and create predictive models are vulnerable to adversarial attacks—small, carefully designed alterations of inputs, sometimes just a few pixels, that can affect the output in a calculated way. These so-called clinical-support tools could mislead physicians to a particular diagnosis or treatment. AI can also be misused to generate information quickly and secretly, allowing a participant to speak confidently about a topic with little real knowledge during Grand Rounds or a remote oral test.
Some of the ethical and practical issues of emerging technology are impossible to predict so it’s imperative that technical, legal, ethical, and medical experts, along with industry and payers, work together to optimize AI-assisted care. One such endeavor is The Collaborative Community on Ophthalmic Imaging Foundation, which includes participants from ophthalmic organizations, academic institutions, government agencies, and the private sector in its forums.
In its outro, the AI-generated ophthalmology rap lyrics sum up the future of machine learning–assisted care:
So there you have it, the eye’s grand symphony,
Ophthalmology, a world of possibility,
From nearsighted to farsighted, we find clarity,
In the realm of vision, we rewrite destiny.
But I still can’t play video games.