Do you want a human overseeing the AI that is used in your care?
There is a good chance your answer is yes. We have seen it as a preference time and time again in social research. But does this approach expose you to more danger? According to different research, the answer is probably yes.
In our latest video, Ellie Asgari, kidney doctor at the UK's National Health Service and AI Scientist at Newton's Tree, explains why human oversight doesn't always make AI safe.
What the video covers
- A Harvard study where changes to a skin image, imperceptible to the human eye, led an AI model to a completely different clinical result. No human in the loop could have spotted the difference.
- A multi-site Newton's Tree study of two commercial chest X-ray algorithms, looking at automation bias - when a human agrees with an algorithm's output without properly reviewing it - and what clinician experience and workload have to do with it.
- Why innovating care does not just mean improving our tools, but improving our oversight of those tools, too.
If the human isn't always properly reviewing the AI's outputs, is it still an appropriate safety mechanism? Watch the full video to find out what the research says.
Watch the full video here
If you're working with AI in your health system, Newton's Tree works with hospitals to support safer AI use, helping strengthen the monitoring layer to ensure patients receive the best care.
