An AI problem

[AI highlights potentially precancerous tissue — adenomas or polyps — in the colon of patients undergoing endoscopy © Stacey Wescott/Chicago Tribune/TNS/Alamy]

Having had a few colonoscopies myself, the piece at reference 1 in the Financial Times caught my eye this morning. A piece which is mainly built on the paper at reference 2, published yesterday.

The actual paper is a bit too expensive for me at around $40, but Google turned up an open access pre-print from an outfit called SSRN, to be found at reference 3. A sort of half-way house to open access.

I have not read this pre-print very carefully, but the drift seems to be that over the past five years or so, AI has been very helpful in colonoscopies - outcomes have been improved - but at the cost of deskilling the colonoscopists. Which all seems very plausible: if you have got a machine to do the work, you don't need to bother yourself and your skills will fade. So if this were true, should we care? Are we the customers happy to trust the AI to be there and to be working properly? It also sounds as if the AI is a money saver, one way or another.

I was also interested to read that an accepted indicator of colonoscopy quality is something called the adenoma detection rate (ADR), where an adenoma is the proper term for an unwanted growth in the colon. A rate which will vary according to patient mix and circumstances. No doubt the paper addresses this point.

I believe that AI is also used to help with the interpretation of scans looking for cancers. Presumably, we have the same issue there. Then what about if AI is used to help with medical procedures in a more direct way?

PS 1: I don't suppose that the image is taken from the present study, but it does look very much like a colonoscopy. I was struck today by how awkward it looks for the colonoscopist, with hands down and eyes up. A much more natural posture is for the eyes to follow the hands. Presumably they get used to it.

PS 2: given that a lot of the authors come from Poland and other places where English is not the first language, I was also struck by the quality of the writing. Presumably academics from such places who want to get on have to take the time needed to be at least competent in English. I don't suppose very many take the time needed for Chinese.

References

Reference 1: Routine AI assistance hits skills of health experts performing colonoscopies: Study comes amid rapid adoption of the fast-developing technology - Clive Cookson, Financial Times - 2025.

Reference 2: Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study - Krzysztof Budzyń and others, Lancet - 2025.

Reference 3: https://www.ssrn.com/index.cfm/en/.