Das hat eine neue Studie herausgefunden, die sich vor allem auf unveröffentlichte Daten stützte, auf deren Grundlage sie nicht trainiert wurde Chat-GPT schneidet besser ab als menschliche Ärzte bei der Diagnose von Krankheiten. Da es viele Open-Source-KIs wie Chat-GPT gibt, bedeutet dies, dass diese Technologie kostenlos zur Verfügung gestellt werden kann.
Wie lange dauert es, bis diese AI Doctors sowohl kostenlos als auch weltweit über Smartphones verfügbar sind?
Manche Amerikaner halten es vielleicht für unmöglich, dass ihr Land ein solches Ergebnis hervorbringen könnte, aber es könnte durchaus von irgendwo im Rest der Welt kommen. Es gibt mehrere NGOs oder sogar UN-Organisationen, die die Finanzierung eines solchen Projekts unterstützen würden. Alternativ könnten viele europäische Länder mit einer allgemeinen Gesundheitsversorgung diese zunächst für ihre eigenen Länder starten und sie dann als Open Source für den Rest der Welt bereitstellen.
All dies scheint sehr bald passieren zu können – die Technologie ist bereits da. Das UBI scheint noch in weiter Ferne zu liegen, und es ist kaum vorstellbar, dass es vor Ende des Jahrzehnts in den Mainstream gelangt. Wie könnten kostenlose Universaldienste die Debatte darüber verändern, wenn BGE in großem Umfang diskutiert wird?
Will universal services, like free AI healthcare, come before UBI? Recent advances suggest the AI technology to enable it is already here.
byu/lughnasadh inFuturology
9 Kommentare
Can’t wait for my AI doctor to hallucinate and inject 17 lethal doses of morphine straight into my bloodstream
What makes you think that the capitalists who own the servers, code, and energy production will use AI for egalitarianism and not wealth acquisition and power consolidation?
Good luck. Ai medicine will probably be more expensive than the crappy medical system we currently have. Corporations are not here to help the world, they are here to make money. Sometimes helping people is side effect but it’s always profit first. I’m not saying corporations are immoral or evil, but stop deluding yourself.
Yep. AI is going to be cheaper than your doctor. Sure. Get back to me when you have to pay to use ChatGPT per prompt instead of the big tech companies eating the cost for it.
Do not forget how profit is made. It starts with attention. In fact, many things start with attention. Attention of viewers, the public, even market attention. Doesn’t matter if you want a job, a date, a customer, its all starting with attention.
Attention = Profit <– Thats what I want you all to remember when listening to or reading about AI advancements. Everything which garners attention has potential for profit. Especially on the internet.
Hype and sensational reporting about overly complicated shit is always going to be common. I see a ton of this with AI. Also with security. Especially with AI security to boot.
I work in these spaces, trust me we are not nearly as close to anything as the articles state because the authors do not truly understand. I dont necessarily blame them maliciously either, mind you.
AI has a *MUCH* higher error rate than the triple nines ( 99.9% ) success rate or uptime that we are accustomed to in general information technology. Also, we are not nearly as close to AGI as articles will lead you to believe. Don’t forget, Attention = Profit … and fear / negative emotions garner more attention more readily than most things.
Right now LLMs ( Large Language Models, like chatgpt, claude, etc ) still ‚hallucinate‘ a lot. They are kinda like a 95 year old with dementia. Sometimes they will think that toasting a hat for breakfast is appropriate. This is because they do not actually think, they guess as to what text output will cause a human to give them a ‚thumbs up‘ ( a positive approval ) of the output.
There are ZERO exceptions to this, the error rates for LLMs are out of tolerance. I regularly see all manner of folks over rely on AI.
They are a tool, a flawed tool. Just like doctors are not perfect, LLMs are far from perfect as well. There are risks. All risks are mitigated through multiple layers ( steps ). Right now the best architectures involve layered risk mitigation.
Okay cool story bro – so what does all this mean?
In other words…
LLMs and AI are TOOLS, they will be work accelerants and productivity multipliers. They will be treated and integrated into our organizations like any staff member or tool, part of a team which is part of a bigger system that has double checks and balances. I tell my clients, and IT people I train, that AI should be treated like an Intern / junior employee at this time.
The distribution of information and skills across humanity takes time, less time than tech deployment. This means we will need to wait a bit for humans to create methods, teach those methods, and for others to learn those methods to practically utilize these new tools.
1. You will first see AI assisting medical professionals, to help catch human error or reduce human labor required to hone in on key details.
2. Then you will see AI taking more lead, and humans doing the double checking ( Quality Assurance, Quality Control )
3. Way later on, you will finally begin to see AI that is robust enough to be reliably doing delicate work.
Most of the media leads you to believe that 3 will come first, and is happening now. It is not, and it will not.
Where can I be wrong? Laws, regulations, economics ( profits… ) etc. But from a reliable engineering and architecture perspective you should think of AI in the way I describe. I am only speaking on the engineering of things. Human behavior and socioeconomics are a whole ’nother can of worms
Note: written fast, I’m sure redditors will point out any errors made and I’ll try to double back and correct them
Diagnosis isn’t the part I’m concerned with not affording, it’s the everything-else. Unless AI is printing my capsules and performing my surgeries for free too, it hardly matters
More to the point, an AI can’t face consequences when it misdiagnoses me. At least a human doctor has the wherewithal to be skeptical of their own judgment and check their work, if only because getting it horribly wrong could ruin their careers
Who is clamoring for AI-led healthcare besides AI investors who are desperate for a payout? Anyone who’s played around with an LLM for an hour knows how limited it is and how easily it fucks up
In case anyone has failed to notice, AI is in the ‚marketshare‘ phase. It’s ‚cheap‘ because it’s not very good, and they’re trying to build a market for it. Almost all AI’s are largely under private control, by and large the means to run them are too. So in the next 5/10/15/25/50 years, they’ll start monetising it. I can see better then human AI being extremely expensive; with a host of cheap AI taking over lower paying jobs.
Here’s the thing with a capitalists society, it will always capitalise on a crisis or a changing marketplace. Be it buying Gov Services running them well for a few years before jacking up prices massively, or running a business at a loss for decades to grow a dependence on it before jacking up prices massively.
Probably still a ways off because this study was based off of clinical vignettes, not directly analyzing patients.
Like, while the other steps in the process seem solvable, I think are underestimating the difficulty of data collection, ongoing monitoring, making sure the patient understood and is following directions, detecting if there’s been an input error, etc.
It also doesn’t solve the politics around medicine. Like, you don’t even need AI to make doctor expertise available to the general public basically free, at least at point of service. Plenty of countries do that already. Making it cheaper is certainly helpful, but there’s an elephant in the room technology isn’t going to be able to fix.
LLMs aren’t really good for high skill things like medical diagnoses. The paper you use as a source even states that as it thesis:
>**Findings** In a randomized clinical trial including 50 physicians, the use of an LLM did not significantly enhance diagnostic reasoning performance compared with the availability of only conventional resources.
>**Meaning** In this study, the use of an LLM did not necessarily enhance diagnostic reasoning of physicians beyond conventional resources; further development is needed to effectively integrate LLMs into clinical practice.
The LLM itself is good at answering test questions, because that is what it is designed to do. It is trained off of written tests. It hasn’t been tested or shown to work when testing with actual patients.
You’re making a claim with no evidence and then extrapolating upon it.