Aviv Tamar uses the POMDP 'tiger problem' to challenge Victor Taelin's claim that extra information never harms rational agents

The problem is that “better decision making” depends on medically approved research with that new data (which will come but take years).

The main point you have to understand is that medicine is bad for you, unless you really need it. Doing an unnecessary treatment (pills, surgery, radiation) has often as bad as an outcome as not doing a necessary one.

@VictorTaelin I have a ton of respect for your work. Read this and tell me what happened. Where and how information theory failed us. What we should do differently. https://jamanetwork.com/journals/jama/fullarticle/900666

@decompiled_dev mds will see a cost problem and rather than thinking "let's reduce the costs to save more patients" they're like "let's kill more patients to reduce the costs"

@avsa ultrasound has no radiation

@VictorTaelin if you have a better "decision making" process please quit larping and save some lives thank you 👍

No but it you find a false positive “lump” some people might decide to do a biopsy or start radiation treatment.

Sure we can use AI to reduce false positives but in human trials it takes time.

I am excited about mid journey’s scanner - but these things aren’t to be dismissed.

@keycane @VictorTaelin That’s called a public swimming pool

@VictorTaelin Every medical test has a big $ cost to it though. Its never free to get information in their context.

@boysbasbleu the guys at Midjourney are making incredible progress in that front, you might check them out!

Yes once we find ways to accurately without false positives and no harmful side effects we should increase testing, but that is not the present we are living in. Instead we have imperfect tests, imperfect understanding of how disease or the human body works. So we should play the hand we are given, that means don't expose people to unnecessary procedures, especially not if they can cause significant drawbacks. Which why there is information that is harmful to know, and information which isn't worth getting. Because you wouldn't perform exploratory surgery on a healthy person just to make sure they are healthy, because afterwards they will definitely not be healthy anymore.

Sure doctors are bad at math (the usual type I / type ii error on positive test result exercise), but you’re expanding the somewhat reasonable position of “point in time testing of broad swaths of the population does not on net improve outcomes” to “more data is bad” The quoted tweet (so cheap longitudinal scan collection becomes de rigeur) is great, though.

@VictorTaelin Rational agent is doing all your heavy lifting. Most people aren't rational agents.

@VictorTaelin @avsa Being put into a vat filled with water somebody else might have pooped into might not be better than radiation.

@VictorTaelin Brave of you to think patients are receptive to a lecture about decision theory when someone tells them maybe the blob is cancer maybe not

@VictorTaelin Not related to the medicine part, but I can see a scenario where more noisy information can be degrading to output quality when the system that does the processing has limited capacity.

@VictorTaelin Blind test brings often no information due to sensitivity and specificity. But once you tell someone "you've been tested positive for cancer", even if the actual chance is 1 / 1000 you'll have to do further invasive test to confirm resulting in more harm on average.

@VictorTaelin Yes, humans are sub-optimal decision makers. decisions about the value of information have to be pragmatic, not idealistic. resources in medicine are limited and interventions carry risks in and of themselves. the game theory is much more complicated than this strawman

@VictorTaelin Yes, but patients are not rational actors and doctors also have incentives to not act rationally. The lack of clinical data will initially make the value of these scan close to zero even if in reality they are reliable.

@VictorTaelin Entropy goes down - DPI