Dr. Mertcan Sevgi is a clinical academic specialising in Artificial Intelligence (AI) research and innovation. He is a Clinical Research Fellow at the UCL Institute of Ophthalmology and an Honorary Clinical Research Fellow in AI at Moorfields Eye Hospital.
We are discussing his NIHR-supported study on the use of AI models that communicate directly in patients’ preferred languages. This study is the first of a multilanguage clinical trial to evaluate the experiences of patients follow-up appointments after surgery. Preparatory work involved community leaders in the design of the trial. Community outreach is key to building a future where equitable artificial intelligence can enhance eye care for all.
Do you think AI in healthcare can help reduce health inequalities? If so, how?
Yes, but not automatically. AI learns from data and if the data does not represent everyone, the system will not work equally well for everyone.
We already know this is a problem in ophthalmic imaging. The datasets used to train these models cluster in a few parts of the world and patient demographics are not frequently reported. The solution is not complicated in principle, it is more representative data and checking performance across different groups.
Conversational voice AI faces the same question. Can these models work well across many languages or only the most commonly spoken ones? Speech technology has not developed evenly, largely because some languages have far more data for models training than others. Multilingual models are improving quickly, but that has to be tested rather than assumed, particularly in a clinical conversation. That is exactly why this trial is so exciting. It takes one specific, well defined clinical task and tests how Dora performs across ten different languages.
How important was partnering with Derman and community organisations in designing this work?
Partnering with Derman was central to this and we are very grateful to them for making it possible.
What we wanted was not simply a focus group with Turkish speakers, but with a group who are underserved and who rely on Derman’s help to access health and social care. Many of the participants did not speak English and the paper sets out the difficulties that creates for them. They are among the groups who potentially have the most to gain from a system like this and we wanted to hear their perspective directly.
They also arranged for one of their interpreters to join the focus group, which gave the participants a familiar face in the room. It is a very different thing to be asked for your views by researchers you have never met than to do so alongside someone you already know and trust.
The group were genuinely enthusiastic about taking part. They wanted to talk, in their own language about the challenges they face and the use of conversational voice AI in the clinical setting. They came in with reservations about AI, but once they heard the demo of Dora speaking Turkish they were positive about it as a way of communicating in their own language. That support was conditional though. They wanted verification that the call was genuine, assurances on confidentiality and anything complex or sensitive to stay with the clinical team. Practical details mattered too, such as accent, pace, choice of voice and how regional dialects are handled.
Can you tell us more about the upcoming trial and what success would look like?
It involves patients who have undergone cataract surgery at Moorfields, each still attending their usual in-person appointment alongside the AI call. The system should identify patients requiring review as reliably in one language as it does in another. Beyond that, I am keen to see patient feedback measures (such as NPS) so we can understand how patients from different backgrounds actually feel about this. I wanted to work on problems at a scale that goes beyond one clinic. In clinic, you help the person in front of you, which matters enormously, but the reach is always limited by how many people you can see. Research and technology are a way of changing what happens for everyone who comes through a pathway.
I work across both academia and industry. What draws me to the product side is the translation. You take research evidence and turn it into something people use, then evaluate and iterate with end users. That keeps you close to the people the work is meant to be for, which is where I think the most interesting questions tend to sit.