AI: From productivity dividend to better patient outcome

Over a year ago, our predecessor in the role of Editor-in-Chief at Clinical and Experimental Ophthalmology (CEO), Prof Justine Smith, wrote an editorial 'Welcoming artificial intelligence into ophthalmology'.1 That is not that long ago but in view of the volume of publication on the subject of artificial intelligence (AI) in the intervening period, I feel another look at the subject is warranted. As she acknowledged, we at CEO are playing our part in this tsunami. Our ever-inventive researcher-authors provide us with one more innovative use of the technology after another, one of which is featured in the letters section of this issue of the journal.2 In my role as an editor, I have the privilege of seeing a wide range of applications described in our constant stream of submissions. I see the same every time I open a table of contents in any other ophthalmology journal. With that overview, a few things are noticeable. The productivity dividend arising from AI for us in ophthalmology is clearly evident in the vast majority of AI innovations. We see it enhancing (if not quite perfecting) data collection, image analysis and image pattern recognition, and patient and health care professional communication. All of these promise to relieve those of us working in clinical settings of burdensome activity.3-7 For researchers, the production of manuscripts is eased and trial design can be enhanced, data can be more easily accessed and analysed and literature searches are performed in seconds. Of course, the veracity of the product of research must always be the province of the human behind the research. Newer tools are emerging to address the problem of so-called AI 'hallucinations' or, in real English, errors in large language models.8, 9 The implied effect of this productivity gift is the freedom for us to move away from time consuming but vital activity. For example, if the machine will screen for a disease as effectively as a human, screening personnel are free to move to other tasks.10 If a researcher takes 20 min to produce a reasonable first draft of a manuscript instead of 3 days, the same applies. The promise is that the new freedom will benefit not only the work-life balance of the professional but also speed developments and streamline services.11 So what is wrong with that? Nothing. But something may still be missing. When a new technique is brought to us in health care, primary in our minds is the question 'is this of benefit to the patient?'. Undoubtedly, if clinicians and researchers spend their newly liberated time speeding developments and streamlining services, there is a benefit to the patient body as a whole. What I have yet to see in any volume is the application of AI and its comparison with conventional treatments with a specific focus on specific patient outcomes.12 Yes, there are some randomised trials of AI techniques versus traditional methods and I expect, with the lead time required for substantial randomised trials, there will be more.13 We need to keep in mind that, fascinating though AI is, it may not always improve on traditional methods.14, 15 Further, the ability of AI to ascertain large data sets and provide so called 'real-world' answers to clinical questions is being explored.16 Do we have the tools to assess the utility of this approach by comparison with formal epidemiology and randomised trials? Clinical and Experimental Ophthalmology looks forward to a future where AI is a routine part of our clinical, teaching and research world and judging by the speed of change, that will be in the near future. Proof of advantage to disease modification and patient management has always been the main focus of clinical research and AI will, undoubtedly, assist us in that goal. CEO will welcome research where that kind of proof can be demonstrated, as it always has. That search for constant improvement in patient outcome will not change. None. The author declares no conflicts of interest.