this post was submitted on 28 Jul 2023
154 points (99.4% liked)
Science
1224 readers
5 users here now
This magazine is dedicated to discussions on scientific discoveries, research, and theories across various fields, including physics, chemistry, biology, astronomy, and more. Whether you are a scientist, a science enthusiast, or simply curious about the world around us, this is the place for you. Here you can share your knowledge, ask questions, and engage in discussions on a wide range of scientific topics. From the latest breakthroughs to historical discoveries and ongoing research, this category covers a wide range of topics related to science.
founded 2 years ago
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I feel for you and anyone suffering with a meat allergy, but I dunno how much I'd trust AI for any serious purposes after seeing the garbage it can spit out.
Seriously, I've managed to get AI to write me instructions on how to inflate a phone and how to shave alligator hair. Rather that say "I'm sorry, that doesn't make any sense, but here are some related topics", instead it literally wrote out actual instructions for that nonsense LOL!
So yeah, I have no reason to trust AI for anything serious, it's about an ignorant joke of a language model is all it really adds up to.
That's specifically for a LLM which would probably not be the best AI base for medical uses.
People still don't understand that AI is an all encompassing term like "tool" and not a single thing.
Just like we use thousands of vastly different and specialized tools, in a decade we'll be surrounded by medical AI, engineering AI, accounting AI, design AI, research AI, life coaching AI, etc.
Right now we have a few LLMs and generative AIs, but that's like having a pen and a spray gun.
Of course you wouldn't ask any of them for a medical diagnosis.
In a use case like this, AI would be less about a final diagnose and more about getting the doctor or patient pointed in the right direction, especially with rare cases that few doctors are aware of. You no longer need to visit a hundred specialists in the hope of finding the one person who's seen something similar to your case before.
Agree in this case AI is just WebMD symptom checker but with the ability to take in infinite data points and narrow it down with prompting questions and hopefully being able to upload images for further diagnosis.
Yeah, I'm not talking about a language model AI. But rather something like the stuff the insurance companies are using to assess risk -- they take a lot of data in and cluster them together. Humans are sometimes really bad at recognizing patterns if you don't have enough data. A pattern that goes: "oh, all these people in this region with this specific digestive problem spatially maps to this insect" is the sort of thing ML should be good at. But where it will be really good is in turning proteins into diagnosis: "if this protein is detected in the blood in an general scan, combined with symptoms, then diagnose X" -- right now you only get tested for the things the doctor orders. Even more promising yet: with enough data, the AI should figure out which proteins actually do specific functions in the body, which will advance the research side (see, for example, Alphafold).
Pattern recognition is something modern techniques are very good at.
ChatGPT isn't that. It also isn't intelligent and doesn't know anything. It's basically a jacked up parrot blindly throwing words together.
Pattern recognition is only worthwhile when it isn't based on some randomized neural network.
Speaking from experience, I've written my own AI like software that actually has a known null state and well defined training process.
In this application, AI would really just be a fancy search engine the doctor can use to look for things he doesn't already know about.