Conversations around artificial intelligence (AI) are gathering pace, particularly in the healthcare space.
The topic of AI was key to the education programme at 100% Optical, from AI models in practice, to how AI is already transforming care, to developments in research.
OT spoke to speakers, experts, and suppliers in the field, about the scope of the technology and what practitioners need to consider when looking at AI in practice.
1 Is AI progressing faster than governance?
Jason Halsey, Rodenstock UK product and training manager told OT: “I think AI currently is very ‘friend’ and ‘foe.’”
Halsey led a discussion workshop during the event titled From daily life to diagnosis – making AI meaningful in optical practice, and explained to OT that currently the focus is on understanding the limitations, where AI is best utilised, and “understanding that not all AI is created equal.”
He noted: “I think the challenge right now is AI is progressing faster than clinical governance can keep up, so it’s trying to find the middle ground.”
Professional bodies are trying to direct professionals on which technologies are usable and which don’t have the evidence behind them, he said.
2 How was the model built?
Kishan Devraj, founder of Ask Fellow Optoms and a locum optometrist, told OT: “When we incorporate this technology within healthcare there is huge potential, but the risks are higher.”
Devraj led a session on Artificial intelligence models in practice, providing an introduction to large language models, their use in healthcare, and the future potential.
Discussing what practitioners need to know about AI when considering its use in practice, Devraj shared with OT: “It’s always important to understand how the AI was developed, to understand how you can relate it to your patient.”
Understanding the foundations of how AI models are built can shape its use, he said, emphasising that if the model has been trained on information that is not representative of the population: “then you could increase inaccessibility.”
3 How could AI provide a trusted second opinion?
Dr Peter Thomas, chief executive officer of Cascader, highlighted the potential of AI for making eye care more efficient.
He said: “We have this huge number of patients who need their eyes treated, both in primary and secondary care. We know that's going to grow. So how can we implement AI in a way that means we make best use of our clinicians’ time to get a really efficient workload?”
The partnership will work to understand how tools developed by Cascader could help practitioners in the community.
Speaking to OT about the potential for AI in healthcare, Thomas said: “I think the key thing is, how can we provide a trusted second opinion for a clinician?”
“For me as a hospital ophthalmologist, if I’m seeing a child and there’s something very strange on the very peripheral retina, I can go next door and ask a vitreoretinal surgeon. How do we deliver that same sort of capability into an optometry practice so that trusted second opinion is available anywhere,” he added.
4 Has the technology been clinically validated?
Cascader’s Horler encouraged optometrists to look at the clinical validation of technology, including the evidence behind it and whether the tool has received regulatory approval.
He noted that one key question around AI use in future will be how often it is used: would practitioners use it with all of their patients, all of the time?
With the potential for oculomics in mind, Horler emphasised that this may involve technology that practitioners would want to run for all patients to produce a risk profile.
“If we can then refer them to their GP, or whoever it is, to prevent that becoming manifest, that’s got to be good,” he said, but emphasised: “I think all of that needs to be worked out. I don’t think anybody really knows how that is going to happen in the fullness of time.”
5 Who is responsible for the data?
Dr Carlos Ciller, co-founder and CEO of RetinAI, explained that as the AI funnel expands, he could see its use for screening growing.
Advising practitioners, he suggested ensuring the framework of any tool or solution is contained and safe, asking who is responsible for taking data, and who owns the data.
Ciller said: “My suggestion will be, just give it a try.”
The best outcome, he suggested, would be finding a new tool for routine clinical care that can mean more time dedicated to the patient.
Watch the interviews with these experts and more in OT’s video above.
Comments (1)
You must be logged in to join the discussion. Log in
Don Williams07 May 2026
This is a very important discussion, but I think we need to be careful with the idea that AI is merely 'gathering pace'.
In many ways, AI has already outpaced large parts of the professional conversation. The recent visibility of large language models has created the impression that AI is suddenly arriving but AI as a field is not new. Its conceptual roots go back many decades, from Turing’s early work to machine learning, deep learning and now generative AI. What is perhaps new is not AI itself but the speed at which healthcare professions are now being forced to confront it.
The governance question is where it becomes particularly complex. We often speak about 'AI governance' as though it exists on a level playing field but it does not. The same AI model may carry very different risks depending on the country, healthcare system, patient population, legal environment, clinical workflow, data infrastructure and professional accountability framework in which it is deployed. What is considered safe, acceptable or proportionate in the UK may not be interpreted in the same way in another jurisdiction, even if the underlying model is technically identical.
That is why I think the conversation around AI risk needs to move beyond the simplistic question of whether a model is 'validated' or 'approved'. Validation is important, of course, but it is not the end of the story. A model may perform well in one setting and fail silently in another because of dataset shift, different imaging devices, different disease prevalence, different referral thresholds, different ethnic representation in the training data or different clinical pathways. Risk is not located only inside the algorithm. Risk emerges from the interaction between the model, the data, the clinician, the patient, the workflow and the system into which it is deployed.
This is also why governance can easily become a paper exercise if we are not careful. AI will almost always move faster than formal regulation, guidance and professional consensus. Governance therefore cannot simply be about having a policy document, a risk register or a regulatory label. It has to involve continuous monitoring, post-deployment evaluation, calibration, audit, accountability and a clear understanding of who is responsible when an AI-supported decision causes harm.
I also think we need to distinguish more carefully between autonomous AI and Augmented Intelligence. In healthcare, and certainly in eye care, the immediate future is unlikely to be AI replacing clinicians wholesale. The more realistic and safer model is clinician plus AI. AI supporting triage, risk stratification, image interpretation, decision support and pattern recognition, while the clinician remains responsible for context, judgement, communication and accountability.
So, yes, AI has enormous potential. But the difficult question is not simply whether AI can improve efficiency. The deeper question is whether we can deploy it in a way that is clinically meaningful, contextually safe, legally accountable and continuously governed after it has entered real-world practice.
For me, the real debate is no longer whether AI is coming. It is already here. The real debate is whether our governance, clinical culture and professional understanding are sophisticated enough to keep pace with it!
ReportLike2