Post

I listened to this yesterday and they talk about this at 25 minute mark: > the rate of thyroid-cancer diagnoses in the Republic of Korea was 15 times that observed in 1993, yet thyroid-cancer mortality remains stable[:] the problem is overdiagnosis [due to] screening. doi.org/10.1056/NEJM... 🧵👇🏼

In the context of so-called AI cancer screenings — yes, some stooges suggest apps as actual solutions — these and many other related findings, indicate that people singing the praises of such apps forgot basic stats: > the balance of benefits and harms including overdiagnosis should be reconsidered

Many who push AI solutionism here (and elsewhere) propose that cancer screenings using AI can be "fixed" by collecting more "diverse” (whatever that means) data. This implies datafication is good or even always good maybe and that these models are actually useful so long as we remove the racism...

The problem is not just that a model is racist it's also that there is no reason to assume that we should extract that data, from vulnerable groups, think of HeLa, Tuskegee, these things do not end well, from e.g. Black people (patients!) and put their data into what? An app owned by a company?

Plus, there is something serious also on the other end! These is no reason to assume that taking these people’s data and putting it inside the models will actually solve anything — and in fact there is a great deal of research showing that this causes the opposite issues: doi.org/10.1016/B978...

There is no reason to assume the model is actually as accurate as it seems for people whose data is included as this is just a snapshot in time of cancers — and importantly a snapshot during which there are still skilled technicians & doctors. If we deskill those professions such tech might fail!

And to add insult to injury: such AI solutionism itself also contributes to the deskilling too, so it’s quite the predicament, as through job losses the skills die, etc. See example below from 1998: olivia.science/before/ web.mit.edu/slava/space/...

When AI proponents who frame statistics in this solutionist way are using "cancer: we also must detect that it is a rhetorical sleight of hand to obscure that often seemingly technical apps are created NOT for medical altruistic reasons but FOR profit. (Also see blue box at the link.)

As you know, this is something the pharmaceutical industry is known to do and is worthy instead of extreme regulation (think of Theranos, Palantir, etc). So much so in fact that journals are unable, even though they claim they want to, keep their pages free from tobacco industry propaganda!

Also see: Baba, A., Cook, D. M., McGarity, T. O., & Bero, L. A. (2005). Legislating “sound science”: the role of the tobacco industry. American Journal of Public Health, 95(S1), S20-S27. doi.org/10.2105/AJPH... For examples of extreme entryism: olivia.science/entryism/

If you ignore false positives they could cripple the medical system, think of the NHS especially, or force people into needless testing, e.g. doi.org/10.1016/j.pv...