Eyeworld

DECEMBER 2023 | EYEWORLD | 27 of the physician in real time and suggest further testing to them. As for ethics and best practices, Mr. Wang said there are several considerations. Overall, he said it's a been a best practice in the chatbot and AI assistant world to explicitly tell people the response is driven by AI. This serves as a disclaimer and helps set expectations to the human being on the other side. Mr. Wang said the FDA is struggling with the blurry line for some clinical aids that don't specifically diagnose. "It's a bit of a gray zone, and the FDA has had questions of when you step over the line" in terms of a diagnosis, he said. Another challenge for AI in practice will be getting all areas of the practice, from the front- of-the-office staff to technicians and nurses to the physician, on board with adopting certain technologies. Dr. Hovanesian said everyone has to do their piece of the puzzle for newly adopt- ed technologies to succeed. Still, Dr. Hovanesian said nothing will substitute the experience and judgment of the human physician. "AI is going to get better and better and may replace some physician function, but we need a level of judgment passed on it and some supervision of it for its initial rollout. There clearly needs to be some healthy skepticism about how we approach this and what we hand over. At the same time, we have to not be too skeptical. We need to recognize that our brains and bodies can only work so fast, and if we have tools that can help us deal with this grow- ing population more effectively, we should look at it as an opportunity to do what we do best, which is use the skills that only experienced doctors have," he said. Mr. Wang said he expects AI in medicine to be more constrained than people may think. The challenge, especially from a diagnostic standpoint, is "how do you ensure that in rare cases and exceptions that [AI] always works?" he said. AI in medicine is probably "less inevitable than the tech enthusiasts say, but it's also not as scary and not as threatening as certain clini- cians think it is," Mr. Wang said. "This is a narrow area in healthcare where AI can help us, and it's an important one be- cause it's one where it would be offering quality assurance processes that we currently don't do," he said. Dr. Hovanesian also gave an example of how AI could help in patient education. If a patient says their dry eye is more prevalent in the morning, for example, AI could serve up educational material about the unwanted side effects of fans blowing on the patient at night. "That's innocuous advice, nothing that would be high risk, but helpful to patients who haven't thought about it. That kind of thing I could see us including and supplementing our care," he said. Another broad area is record searching. AI, Dr. Hovanesian said, could identify glaucoma patients who should come in for updated visual field imaging, for example. It also could alert physicians to things they need to address, for example, if a biopsy is conducted but there isn't an indication that the result came in. "When we do a test, we're responsible for following up on the results medicolegally," he said. "AI could assist us by combing through our records and looking for patients who have fallen through the cracks." Dr. Hovanesian and Mr. Wang spoke about how chatbots, similar to ChatGPT, can be used in ophthalmology, such as for appointment scheduling, providing educational content, and triaging to the correct department. A study pub- lished earlier this year evaluated AI chatbots' ability to respond to patient questions posted on a social media forum. The authors concluded that the chatbot "generated quality and empa- thetic responses." 2 "A lot of tools like that would be fairly easy applications of AI to enhance the practice of medicine. They're not sexy but they would be great for patient care," Dr. Hovanesian said. Shawn Lin, MD, envisioned a future where AI was used not only for scribing and patient education and communication but also for com- bating insurance denials or preauthorization let- ters and real-time decision support. He said that an AI system could process the natural language References 1. Lu W, et al. Applications of artificial intelligence in oph- thalmology: general overview. J Ophthalmol. 2018:5278196. 2. Ayers JW, et al. Compar- ing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Intern Med. 2023;183:589–596. Relevant disclosures Hovanesian: Alcon, various eyecare technology companies Lin: None Wang: None Contact Hovanesian: DrHovanesian@harvardeye.com Lin: slin@jsei.ucla.edu Wang: james@creativeventures.vc

• Cover