Well, hopefully the patient won't have acne *and* skin cancer, or rosacea, or an allergic dermititis, or fungal folliculitis, or a lupus rash...

@dalfen I pulled up one of the research publications they list of their website. It looks like they had at least some variation in their training data, but my expectation is still they did the bare minimum to make a splash and satisfy some auditor (who may have an agenda of promoting AI tools).

https://www.nature.com/articles/s41591-023-02728-3

@dalfen more broadly, I have a few questions/suspicions:
1. if this medical treatment is so simple and low risk, why can't we self-medicate?
2. Do doctors diagnose these conditions based on photos/videos, or would they expect to see someone in person to provide a proper diagnosis? If not, then the 'human oversight' doesn't seem to be worth much if it's just looking at the photos.
3. Is the subscription model going to produce bad incentives for the company?
4. Can they show that this triage system increases access to dermatologists (like they claim it will)?

I could imagine incorporating this system into a care procedure, as a triage step. If you need to wait two months to see a dermatologist anyway, maybe starting this treatment immediately will help speed up the whole process, regardless of the illness (sometimes treatment is part of diagnosis). Maybe patients will require fewer visits to the doctors to get their acne under control. Maybe it could flag cases that look very bad so that doctor's could prioritize those patients on a cancellation list. But here, it sounds like the company wants to replace doctors.

@DecaturNature Oh it definitely wants to replace doctors. The literature explains a phased system where the AI tech is increasingly used as doctors' services are utilized less. I also saw an interview where the company's founder basically said as much.

You have great points. I chuckled a bit at point one (and said "Great point!" or something to myself lol).

1/4

@DecaturNature Concerning point two, the company plans to eventually have a physician retroactively review a sampling of prescriptions only once a week per the article – and it seems only with pictures. So human oversight doesn't seem to much matter.

(hmm... I wonder who the patient will sue when the AI misdiagnoses or misses something important )

2/4

@DecaturNature An aside – I have issues with tele-medicine in general (w/ human assessors) because having worked at patients' bedsides for many years and felt thready pulses before heart attacks and smelled the breath of those who are ill, etc., I believe assessment takes multiple senses, and much is already lost with only pictures and video.

AI basically only "sees" & maybe "hears," and its assessments will be confined to its limitations, causing it to miss patient problems, imho.

3/5

@DecaturNature And then concerning your 4th and 5th points, can you explain more what you mean by "bad incentives for the company"?

What I see is this AI company (and others) would seek to make a lot of $ off of people through subscription models & use the idea of a subscription model as a selling point to dermatologists also seeking to make $. (That said, I am not sure it increases the likelihood that patients will actually *see* dermatologists in person.)

4/5

@DecaturNature (backtracking)... though the company plans to eventuallly have a human doctor retroactively review a only a sample of the prescriptions once a week, the article says the program aims "to complement" physicians by passing "low acuity" cases to AI... (hmm)

Like I said before, there's great value in human assessment & much can be missed relying on AI (even patient problems unrelated to whatever AI is directly evaluating, like catching labored breathing when assessing the skin).

5/5