Our Voice on AI in healthcare

Key themes, in people's own words

Q. Is it okay for AI to help make health decisions? "Yes, but only with a human check." The most common shape of answer wasn't a flat yes or no, it was conditional. "Yes, if the doctor checks it." People seem open to AI as a second opinion, not a replacement for their doctor.

Q. How important is it to understand how the AI reached its answer? Fairness is the biggest worry and people can explain why. Several answers show real understanding of how bias creeps into AI, not just that it might:

"Depends on how the AI model learns. It would depend on how it could be taught to 'forget' about those biases." "It depends what data sets make up the knowledge hub that AI models are trained on. If one size fits all orchestration of AI agents is used then this will exacerbate inequalities but the potential improvement in outcomes is significant if inequalities are factored into the design of the system." "Underlying data and its biases need to be understood. It could also be used to highlight bias/inequalities."

Q. Would AI make the NHS more fair, or less fair? A strong warning about trust. One longer answer stood out as a caution against over-explaining AI's role, for fear of the opposite effect: "Using technology to determine appropriate treatment isn't new and AI is already in use in medicine and treatment research, just not in the format the wider public is becoming aware of now, i.e. gen AI and large language models. Drawing unnecessary attention to AI technology used in recommending appropriate treatment will just undermine public trust in the NHS and lead people to make assumptions that their treatment can't be trusted."

This is worth flagging on its own, it's a genuinely different take from "tell people more," arguing that how you frame the announcement could do more harm than the AI itself.

Accessibility was raised. One respondent gave a detailed, practical suggestion for people with visual impairments: "App on phone for visually impaired which reads the letter, braille on posters, magnetic under poster and it will read out the instructions." Another simply asked for "regularly updating information to keep it relevant."

The people answering were genuinely diverse. Ages ranged from 18 right up to 64+. Ethnic backgrounds included Arab, Indian, African, Caribbean, Pakistani, Bangladeshi, Sri Lankan, Somali, Irish, White British, and several mixed backgrounds. Housing answers ranged from "Own" and "Rent" to "Temporary Housing" and "Homeless." That spread matters it means the fairness concerns above are coming from people who are themselves more likely to be affected by unfair data.