this post was submitted on 09 Jun 2025
190 points (99.0% liked)
Technology
71585 readers
4688 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Sigh. Unfortunately there's a lot of misinformation around this topic that gets people riled up for no reason. There's plenty of research in healthcare decision making since Paul Meehl (see Gerd Gigerenzer for more recent work) that shows using statistical models as decision aids massively compensate for the biases that happen when you entrust a decision to a human practitioner. No algorithm is making a final call without supervision, they are just being used to look at situations more objectively. People get very anxious in healthcare when a model is involved and yet the irony is humans alone make terrible decisions.
I'd like to know what specific steps are being taken to remove the bias from the training data, then. You cannot just feed the model a big spreadsheet of human decisions up to this point because the current system is itself biased; all you'll get if you do that is a tool that's more consistent in applying the same systemic skew.
There is an implicit assumption here that models are being 'trained', perhaps because LLMs are a hot topic. By models we are usually talking about things like decision trees or regression models or Markov models that put in risk probabilities of various eventualities based on patient characteristics. These things are not designed to mimic human decision makers, they are designed to make as objective a recommendation as possible based on probability and utility and then left down to doctors to use the result in whichever way seems best suited to the context. If you have one liver and 10 patients, it seems prudent to have some sort of calculation as to who is going to have the best likely outcome to decide who to give it to, for example, then just asking one doctor that may be swayed by a bunch of irrelevant factors.
[Citation needed]
If these things were being based on traditional AI techniques instead of neural network techniques, why are they getting implemented now (when, as you say, LLMs are the hot topic) instead of a decade or so ago when that other stuff was in vogue?
I think the assumption that they're using training data is a very good one in the absence of evidence to the contrary.
Because it's sensationalist reporting that is capitalising on existing anxieties in society.
The MELD score for liver transplants has been used for at least 20 years. There are plenty of other algorithmic decision models used in medicine (and in insurance to determine what your premiums are, and anything else that requires a prediction about uncertain outcomes). There are obviously continual refinements over time to models but nobody is going to use chatGPT or whatever to decide whether you get a transplant.
https://onlinelibrary.wiley.com/doi/pdf/10.1002/hep.21563
https://onlinelibrary.wiley.com/doi/pdf/10.1002/hep.28998