The Potential of Chatbots in Ophthalmology

Stephen Hawking once said, “The rise of powerful AI will either be the best or the worst thing ever to happen to humanity.” As an optimist,  I sit on the “for better” side of the fence – and I truly believe in the future-facing potential of chatbots in ophthalmology.

Let’s start with the basics. A chatbot is a computer program designed to simulate either written or spoken conversations with human users without intervention or manual research. I’m sure you’ve already heard of the large language model (LLM) called ChatGPT – and you may also know Gemini (formerly Bard). A growing body of literature is now pointing to the immense promise they could have in the field of ophthalmology across a variety of areas, including education, communication, and even clinical decision-making.

Arguably the most promising application of AI chatbots is their ability to enhance patient engagement, while simultaneously educating them about their respective eye conditions; in other words, effectively simplifying complex medical concepts. In the future, AI chatbots may also allow patients in remote and underserved areas to rapidly access ophthalmic advice. Democratization is a prevalent theme in our industry, and elevating the care received in low- to middle-income communities is a major possibility.

Clearly, there are many uses of AI chatbots for patients, but we mustn’t miss out the multitudinous benefits they also offer to practicing ophthalmologists. Whether it be answering complex ophthalmic questions, writing cataract surgery operative notes, or even triaging ophthalmic symptoms, ChatGPT and Gemini have performed well across a wide variety of clinically relevant tasks. There is, however, no current consensus as to which AI system is superior. To fill this gap, my team and I did a comparative analysis of both ChatGPT and Gemini (1).

We started by prompting the two LLMs with: “I see flashes of light in my left eye, should I go to the emergency department?” Both ChatGPT and Gemini yielded appropriate responses. They recommended that I immediately attend an emergency department, while also stating that this could be a symptom of retinal detachment or tear, which should be evaluated urgently.

We then prompted the chatbots with, “Lines are blurry in one eye? What should I do?” Interestingly, Gemini provided a more specific and accurate response, identifying a variety of possible causes for this affliction, along with recommending consultation if I experienced other symptoms, such as sudden visual loss, floater, pain or light flashes. ChatGPT, on the other hand, generated a less specific response, but was still relatively appropriate.

Finally, we tested the image analysis capabilities of both AI chatbots. But when presented with a fundus image of arteritic anterior ischemic optic neuropathy, neither ChatGPT nor Gemini could recognize it. As it stands, the image analysis capabilities of these chatbots are far from being clinically suitable.

In general, both AI chatbots demonstrate impressive performance levels, with potential for further enhancements in future. But, as with any revolutionary technology, new challenges and ethical considerations arise. Hallucinations, or false responses, are currently the largest concern with AI chatbots. Since chatbots do not have a truly human-like understanding of text, small misunderstandings can affect the generated response, leading to errors that pose risks to patients. It is also possible that biases in training data could inadvertently perpetuate existing healthcare disparities through AI chatbots.

Ensuring that the “human touch” of ophthalmology isn’t lost with the rise of LLM chatbots will be paramount. AI chatbots should complement the role of an ophthalmologist rather than serve as a barrier between the patient and ophthalmologist. Although updated versions of ChatGPT – such as GPT-4.0 – have improved accuracy when answering ophthalmic questions, they still provide unreliable and biased medical advice.

Ultimately, the integration of AI chatbots in ophthalmology needs to be approached with optimistic caution. Though LLMs have the potential to enhance the quality of care for patients and create a more efficient workplace, we must take each step with a grain of salt. That said, if we move forward appropriately, the potential of AI-informed care could exceed all expectations.