Lack of detailed understanding of drug mechanisms and potential risks of AI in clinical trials

Signals over time

3 observed signals across 1 sources, tracked for 2 days. Confidence: low.

What people are saying Observed

• “>I ask nearly every doctor/researched in the medical field: if you had a AI-created drug that tremendously improved cancer treatment outcomes for your patient, would you hesitate to prescribe it because nobody understood how it worked? I have yet to hear "yes, I would hesitate", most people say "it would be cruel to deny a person a treatment that worked". You think that way because you're already working on an unfounded assumption, namely that the AI found a cure for cancer and that the drug has already went through clinical trials. What you don't consider is that e.g. a cancer researcher might have produced a million AI created drugs and one of them is a potent drug for cancer and 999999 other drugs are useless, actively harmful or even deadly. Now you go to to the same guy and say "You have an AI-created drug that tremendously improves cancer treatment outcomes. Why would you hesitate to prescribe it, because nobody understands how it works?" And suddenly you are injecting 999999 useless AI created treatments into people to find the one cure that works.”

  Hacker News · frustration

• “> [...] would you hesitate to prescribe it because nobody understood how it worked? I have yet to hear "yes, I would hesitate", most people say "it would be cruel to deny a person a treatment that worked". I feel like the way you frame the question makes answering it very easy but oversimplifies the reality. In a situation where an AI suggests a treatment to you without explanation and it appears to work, probably you can't even determine that the improved outcome is due to the drug or due to randomness (yes there are really rare cases where people recovered from stage 4 cancer). Especially if it's a personalized drug. If in this hypothetical world "it works on the patients I've tried it on" is good enough to try the drug, and you're willing to potentially cause harm if it doesn't work, then what's the line between science and pseudoscience/quack medicine? Every treatment surely "appears to work" in at least some cases. So wouldn't it be cruel to deny the patient a try at every last treatment that has ever worked for someone in the past, including practices like bloodletting, acupuncture, etc? Would ”

  Hacker News · frustration

• “I think the best thing to say is that by thinking through the example, sharing it with people in the field, and then steelmanning it, he might have had some good points: 1) pharma would love to understand the mechanisms of drugs in detail but we don't have that. AI could help. 2) there is a possibility that a clinical trial run by AI could hack the trial get something approved when it wasn't safe or effective.”

  Hacker News · complaint