What are you going to train it off of since basic algorithms aren’t sufficient? Past committee decisions? If that’s the case you’re hard coding whatever human bias you’re supposedly trying to eliminate. A useless exercise.
A slightly better metric to train it on would be chances of survival/years of life saved thanks to the transplant. However those also suffer from human bias due to the past decisions that influenced who got a transpant and thus what data we were able to gather.
And we do that with basic algorithms informed by research. But then the score gets tied and we have to decide who has the greatest chance of following though on their regimen based on things like past history and means to aquire the medication/go to the appointments/follow a diet/not drink. An AI model will optimize that based on wild demographic data that is correlative without being causative and end up just being a black box racist in a way that a committee that has to clarify it’s thinking to other members couldn’t, you watch.
“Treatment request rejected, insufficient TC level”
A Voyager reference out in the wild! LMAO
Had to be done. It’s just too damn close not to.
The death panels Republican fascists claim Democrats were doing are now here, and it’s being done by Republicans.
I hate this planet
That’s not what the article is about. I think putting some more objectivety into the decisions you listed for example benefits the majority. Human factors will lean toward minority factions consisting of people of wealth, power, similar race, how “nice” they might be or how many vocal advocates they might have. This paper just states that current AIs aren’t very good at what we would call moral judgment.
It seems like algorithms would be the most objective way to do this, but I could see AI contributing by maybe looking for more complicated outcome trends. Ie. Hey, it looks like people with this gene mutation with chronically uncontrolled hypertension tend to live less than 5years after cardiac transplant - consider weighing your existing algorithm by 0.5%
Tho those complicated outcome trends can have issues with things like minorities having worse health outcomes due to a history of oppression and poorer access to Healthcare. Will definitely need humans overseeing it cause health data can be misleading looking purely at numbers
I wouldn’t say definitely. AI is subject to bias of course as well based on training, but humans are very much so, and inconsistently so too. If you are putting in a liver in a patient that has poorer access to healthcare they are less likely to have as many life years as someone that has better access. If that corellates with race is this the junction where you want to make a symbolic gesture about equality by using that liver in a situation where it is likely to fail? Some people would say yes. I’d argue that those efforts towards improved equality are better spent further upstream. Gets complicated quickly - if you want it to be objective and scientifically successful, I think the less human bias the better.
Everyone likes to think that AI is objective, but it is not. It is biased by its training which includes a lot of human bias.
Creatinin in urine was used as a measure of kidney function for literal decades despite African Americans having lower levels despite worse kidneys by other factors. Creatinine level is/was a primary determinant of transplant eligibility. Only a few years ago some hospitals have started to use inulin which is a more race and gender neutral measurement of kidney function.
No algorithm matters if the input isn’t comprehensive enough and cost effective biological testing is not.
Well yes. Garbage in garbage out of course.
That’s my point, this is real world data, its all garbage, and no amount of LLM rehashing fixes that.
Sure. The goal is more perfect here, not perfect.
Yeah. It’s much more cozy when a human being is the one that tells you you don’t get to live anymore.
Human beings have a soul you can appeal to?
Not every single one, but enough.
I still remember “death panels” from the Obama era.
Now it’s ai.
Whatever.
everything republicans complained about can be done under Trump twice as bad, twice as evil and they will be ‘happy’ and sing his praises
Say what you will about Will Smith, but his movie iRobot made a good point about this 17 years ago.
(damn I’m old)
Hasn’t it been demonstrated that AI is better than doctors at medical diagnostics and we don’t use it only because hospitals would have to take the blame if AI fucks up but they can just fire a doctor that fucks up?
I believe a good doctor, properly focused, will outperform an AI. AI are also still prone to hallucinations, which is extremely bad in medicine. Where they win is against a tired, overworked doctor with too much on his plate.
Where it is useful is as a supplement. An AI can put a lot of seemingly innocuous information together to spot more unusual problems. Rarer conditions can be missed, particularly if they share symptoms with more common problems. An AI that can flag possibilities for the doctor to investigate would be extremely useful.
An AI diagnostic system is a tool for doctors to use, not a replacement.
Studies have also shown that doctors using AI don’t do better than just doctors but AI on its own does. Although, that one is attributed to the doctors not knowing how to use chatgpt.
Do you have a link to that study? I’d be interested to see what the false positive/negative rates were. Those are the big danger of LLMs being used, and why a trained doctor would be needed.
has it? source?
I would rather have AI deciding it than bank account balances.
What do you think the AI would be trained on?
See also: UnitedHealthCare
