Stronger point: since we are at Less Wrong, think Bayes Theorem. In this case, a “true positive” would be cancer leading to death, and a “false positive” would be death from a medical mishap trying to remove a benign cyst (or even check it further). Death is very bad in either case, and very unlikely in either case.
P (death | cancer, untreated) - this is your explicit worry
P (death | cancer, surgery)
P (death | benign cyst, untreated)
P (death | benign cyst, surgery) - this is what drethelin is encouraging you to note
P (benign cyst)
P (cancer)
My prior for medical mishaps is higher than 0.5% of the time, but not for fatal ones while checking/removing a cyst near the surface of the skin. As drethelin’s #2 notes, this is not binary. If it is not a benign cyst, you will probably have indicators before it becomes something serious. Similarly, you have non-surgical options such as a cream or testing. Testing probably has a lower risk rate than surgery, although if it is a very minor surgery, perhaps not that much lower.
If the cyst worries you, having it checked/removed is probably low risk and may be good for your mental health. But now we might have worried you about the risks of doing that (sorry) when we meant to reduce your worries about leaving the cyst untreated.
In general if you list everything you can think of and give it probability scores, you ignore unknown unknowns.
For medical interventions like surgery unknown unknowns are more likely to be bad than to be good.
As a result it’s useful to have a prior against doing a medical intervention if there no strong evidence that the intervention is beneficial.
Maybe we need to visualize surgery different. I used to think about it like replacing a part in a car. Why not just do it if the part is not working too well.
Maybe we should see it as damage. It’s like someone attacking you with a knife. Except that the intention is completely different, they know what they are doing, their implements are far more precise and so on, so the parallel is not very good either, I am just saying that “recovering from an appendicitis” could be at least visualized as something closer to “recovering after a nasty knife fight” than to “just had the clutch in my car replaced”.
Stronger point: since we are at Less Wrong, think Bayes Theorem. In this case, a “true positive” would be cancer leading to death, and a “false positive” would be death from a medical mishap trying to remove a benign cyst (or even check it further). Death is very bad in either case, and very unlikely in either case.
P (death | cancer, untreated) - this is your explicit worry P (death | cancer, surgery) P (death | benign cyst, untreated) P (death | benign cyst, surgery) - this is what drethelin is encouraging you to note P (benign cyst) P (cancer)
My prior for medical mishaps is higher than 0.5% of the time, but not for fatal ones while checking/removing a cyst near the surface of the skin. As drethelin’s #2 notes, this is not binary. If it is not a benign cyst, you will probably have indicators before it becomes something serious. Similarly, you have non-surgical options such as a cream or testing. Testing probably has a lower risk rate than surgery, although if it is a very minor surgery, perhaps not that much lower.
If the cyst worries you, having it checked/removed is probably low risk and may be good for your mental health. But now we might have worried you about the risks of doing that (sorry) when we meant to reduce your worries about leaving the cyst untreated.
In general if you list everything you can think of and give it probability scores, you ignore unknown unknowns. For medical interventions like surgery unknown unknowns are more likely to be bad than to be good.
As a result it’s useful to have a prior against doing a medical intervention if there no strong evidence that the intervention is beneficial.
Maybe we need to visualize surgery different. I used to think about it like replacing a part in a car. Why not just do it if the part is not working too well.
Maybe we should see it as damage. It’s like someone attacking you with a knife. Except that the intention is completely different, they know what they are doing, their implements are far more precise and so on, so the parallel is not very good either, I am just saying that “recovering from an appendicitis” could be at least visualized as something closer to “recovering after a nasty knife fight” than to “just had the clutch in my car replaced”.
What do you think?
Why do you think we need to do so?