Utah to let AI examine patients and prescribe medication without human oversight
techspot.com
techspot.com
So here, AI isn't doing anything clever, it's just asking "are you sure you don't have a heart condition" and "are you sure you're not a crazy person". We don't really need LLMs for that, but if it's a catalyst for liberalization, fine.
And the medications that are controlled vs uncontrolled (at least here in Texas) seem completely random.
So are we just admitting that the "AI" here is the equivalent of a dialog box and this same change could have been enacted at any point in the last half century?
>We don't really need LLMs for that, but if it's a catalyst for liberalization, fine.
Is this liberalization or is it just further entrenching the bureaucracy that this change implies is already broken?
This whole thing just feels like a Chesterton's fence scenario in which we knock down the fence and then immediately rebuild it, only a few inches shorter and with weaker materials this time. I just don't see how that's any sort of victory for progress.
I am a big believer in AI for drug discovery. But AI applied to patient care will never be used to benefit the patient or increase quality of care.
Cost reductions will be absorbed for profits. The medical industry has already aggressively implemented AI automation (imaging, automated note taking, etc) and in that period, costs have only increased.
And doctors aren't infallible, AI (and rule based systems) can act as an extra safeguard for them as long as they don't blindly trust it.
Drugged with a topical ointment and complacent by having less acne? This is a one-year trial for patients with mild-to-moderate acne. Let's not blow this situation out of proportion. This trial seems rather narrowly scoped and overall a positive idea for well understood treatments that hardly need an MD to begin with.
If you want to get into conspiracy theories of a government wanting a drugged and compliant citizenry just look no further than the sudden push legalizing THC across the country.
I see this same fallacy repeated all the time when we speak about AI.
"I want AI to be 100% accurate and not 99.999%"
"If AI makes even one mistake, why would I even use it?"
"What's the point of using AI if I have to verify it anyway?"
Here's how it could work.
You could always have a CI sort of thing for doctors prescriptions where you run the prescription through AI and it can flag any potential error to the doctor.
The LLM can be accurate enough that it only flags things when necessary
You could always have a CI sort of thing for doctors prescriptions where you run the prescription through AI and it can flag any potential error to the doctor.
"Lets not write unit tests. We can hire developers to write good code that doesn't break"
We usually regulate a one-size-fits-all approach to everything in medicine. An M.D. (or whatever NPs have?) is required to prescribe medicines, even medicines which are just the same or slightly stronger formulations of OTC medications like a 400mg Ibuprofen or a 1% hydrocortisone cream. Compare this to Mexico, where you can buy many of our "prescription-only" meds OTC.
A lot of healthcare issues are just not high-stakes. Requiring a doctor in these cases doesn't help the average patient, because the average patient would rather ignore the problem, or now, ask ChatGPT how to treat it with OTC options, than deal with the mess of scheduling doctor appointments, dealing with insurance (or in Canada, long wait times to see the doctor for free), only to get just 5 minutes with the actual doctor who may not even listen to you.
I know it's a tough pill to swallow for doctors, and I recognize that it's not doctors who chose the terrible healthcare system we have or decided that they should have 5-10 minutes with a patient. But I do think we need to change the rules for things that have little risk of causing harm but a great chance of improving the aggregate outcomes by increasing access to healthcare.
I have for example had tonsilitis 5+ times, and a lot of common colds/flus/etc. I know how my body reacts and how it feels when I have one or the other.
Yet, I have to be screened by nurses, sent around, given weak antibiotics, and in general wasting a lot of time when I could just convince a model to give me the prescription.
I have a friend with acne who has the same issues with tretinoin prescriptions. Endless appointments and time waste.
Step in the right direction
I have for example hypersensitivity to radio waves. I know how my body reacts and how it feels when I'm exposed to such radiation, especially during vaxxine toxicity flareups.
Yet, I have to be screened by nurses, go through appointments with psychologists, and in general wasting a lot of time when I could just convince a model to prescribe the radium water I need.
Step in the right direction.
Sometimes doctors make choices that reflect a landscape of risk the patient is no more prepared to understand than the doctor is ready to run a network test. People who say they “know they need something” are often confidently wrong.
Where I am in BC, Canada, pharmacists can prescribe specific medications for certain pre-approved conditions, and there is also the concept of an "emergency" refill for if you lose your medication or are unable to get an appointment in time, a few days of a previously prescribed medication can be provided.
At the end of the day, a better solution is to have tiers of prescription meds. Not everything needs to be gate-kept by MDs. If an AI can prescribe it, there is no reason a nurse, pharmacist or other medical professional shouldn't be able to as well.
But there are. Costs and wait times go down. Can be handled like a drive-through. No mis-diagnoses due to fatigue, etc.
If a 20 minute telehealth visit with a PA is a sufficient level of screening for a medication, it's hard to imagine a compelling reason why an AI system would not be.
To make things more complicated, every message I sent was responded to by a different nurse practitioner. So it seemed everytime i sent a response or question, it would get routed to a random NP and she would respond without reviewing the full conversation.
Works if you know what you need, and I can't imagine AI being a worse process than whatever Amazon is doing.
> In the second phase, the next 400 patients will have their prescriptions submitted to pharmacies by the AI and reviewed retrospectively by a physician weekly.
> The final phase will involve a doctor reviewing at least 10% of the total prescriptions issued once a month.
This is a really good way to go. That's how you get to a proper system. It's the route that should've been taken from the very beginning with all human-facing AI-brokered services, and would've meant far fewer nightmare stories.
Acne is a real problem for some people, but comparatively low stakes, mostly visual evaluation, and with pretty well-known solutions. (Of course everything has corner cases, but in general this is a well understood area.)
I have no idea how well this will go, but starting here rather than, say, robotic cancer treatments, can help AI medicine develop into more.
If it does go badly, then after a year we can re-evaluate and either give up or refine the system.
Is it? At least as a customer I'd love to replace a doctor with AI. Humans suck, you have to wait so long for an appointment, they spend like 10 minutes on your case, and often have very outdated knowledge. The main issue is that vision still sucks and lack of embodiment is a problem. It could invoke "human subagents" for that.
unless frontier models (or whatever employs in medicine) are being kind enough to absorb the lawsuits here, I don't see how this ends well.
The accountability is with the company owners. Like with every healthcare startup, there's no reason to assume they'll love being sued or fined so much that they will bake violating HIPAA into their business plan.
They have lawsuits built into this, and either by hubris or bribes believe this is the most profitable route. "Its just acne, what could possibly go wrong?"
I don't think it's automatically true though. In the case of acne medication, I remember there's a well-known one that carries the very real caveat that it'll kill or severely deform the baby if you're pregnant. When you consider the old classic line "Think of the average person. Half of the people are dumber than that" I don't feel comfortable with that medication sitting on the shelf next to the Clearasil, where a busy person will just grab it because they heard it was super effective, or because the other types were out of stock.
The AI inference involved there no doubt costs pennies; so having it screen the patient to ensure that she understands the risks and has taken a pregnancy test recently if unsure, seems like a "why not" kind of thing.
Now, if some AI-healthcare company is capturing 90% of the money saved by not having doctors do this, then it would be barely worth it.
[0] - https://www.mobihealthnews.com/news/nolla-health-launches-ut...
I wouldn't say that, but it's certainly one where the consequences are high and doctor-in-the-loop seems highly desirable for the foreseeable future, especially since most patients are not going to be knowledgeable enough to realize the shortcomings of an LLM and make an informed decision about whether they're OK taking an LLM-prescribed medication.
That's a pretty thin defense when things do go wrong.
Never said that at all.
Statements like this is effectively saying "Nothing will go wrong".
> but they can always add a no liability clause to their ToS
Then why should I trust the output at all?
However, the entire point was to reference the IBM training manual that states: "A computer can never be held accountable, therefore a computer can must never make a management decision".
If the bot cannot be held accountable, it shouldn't be allowed to make decisions.
Again, not at all. There's very clear context and scope for the pilot as presented by the article. Doctors involved approved 8 treatments as appropriate for recommendation in this manner, for mild to moderate acne. And the first 100 cases will be reviewed before the recommendation is finalized.
Even with all that, a "bad" recommendation can still be made at a later time (and there are still views to catch them), but if that case does happen then there's negligible harm. The chances and scope of harm from using public roads (something done regularly by billions worldwide) vastly outweigh this, and unless you can come up with at least one scenario where AI can cause non-negligible harm in the pilot context, you're just fearmongering.
> Then why should I trust the output at all?
Nobody is forcing you to.
If the bot cannot be held accountable, [...]
There is no dependency relationship between accountability and decision making. But, again, it's up to individuals if they want to use this service.
Just in case anyone else has the gut "oh ffs" reaction to this.
You don't have to go to medical school to know what bad acne looks like, or how to treat a UTI. We should be working on tiering prescriptions based on ease of diagnosis and risk.
https://www.justice.gov/opa/pr/opioid-manufacturer-purdue-ph...
https://www.propublica.org/article/purdue-settlement-leaves-...
I'm curious if this will hold up, the people of Utah rejected that mega data center iirc, maybe they feel differently on this one because of what private equity has done to rural healthcare
big /s