socialjusticemunchkin:

shieldfoss:

socialjusticemunchkin:

shieldfoss:

socialjusticemunchkin:

shieldfoss:

socialjusticemunchkin:

nostalgebraist:

cancer, and the pleasure of the weed

Okay, can someone ELI5 why I’m supposed to pick “the pleasure of the weed” here?

Like, this isn’t some abstract theoretical toy, this Smoking Lesion, or as it would be better called, Exercise Genetics problem is actually a thing in my IRL

It has probabilistic Azathoth instead of absolute Omega, and I may be able to peek into the boxes before selecting thanks to modern technology, but even then it would seem that me being the sort of clockwork thing that has the property “one-boxes” would make me likely to be the sort of a clockwork thing that has the property “gets a million dollars” and being the sort of a clockwork thing that has the property “one-boxes even if the box turns out to be empty when peeking into it” would make it extra-likely?

Thus, I should be the sort of a person who likes exercise so I’d have the genes that make one like exercise and live long, even if I turn out to not have such genes, because the sort of a person who has good genes would make such a choice?

Then again, turning it the other way around into the Psychosis Weed problem (people with early psychosis are more likely to self-medicate) doesn’t make me interested in impacting my choice on whether or not to choose “the pleasure of the weed” to avoid psychosis-related genes retroactively.

One could argue that the question is different because self-medicating is caused by symptoms and thus choosing to have symptoms or not (yeah, good luck with that) would be the thing that matters while the choice to exercise is directly controlled by the exact neurochemistry the Exercise Genes are about, and I think that one is probably “the” reason for it. So using that logic I’d determine my choice in the Smoking Lesion based on the mechanism of action of the lesion.

On the other hand, it could be that I’m supposed to choose not exercising and I just inherently enjoy exercise because I have the good genes that make me live long and prosper and thus my neurochemistry is motivated to interpret the Exercise Genetics that way?

Because it is pleasure. Why would you not pick pleasure.

If being the sort of a person who picks that specific pleasure means I’m the sort of a person who gets cancer, then I don’t want to be the sort of a person who picks that specific pleasure.

If you have the desire, you have the unfortunate gene, no matter whether you give in or not.

But the question is determining whether or not I have the desire, as it’s not unambiguous. As a neurochemical process, my choice to engage in exercise is not separable from the desire to engage in exercise but rather a manifestation of it, thus in the absence of obvious markers that would establish the causal connection I don’t know what exactly my genotype is; but if I act in accordance with the choices that would be made by someone with the favorable genes, I’d expect it to be a manifestation of better chances of having received the favorable genes in the first place, as I would be less likely to have made such a choice if my genes were unfavorable.

From the top, just in case somebody else is reading along:

HYPOTHETICALLY: Gene A (Hence: GA) has a large measure of control both over the pleasure you get from smoking cigarettes AND over you chance of lung cancer. The reason we thought smoking caused cancer (Before the discovery of GA) was simply that the same people smoked and got lung cancer, but now we know cigarettes and lung cancer are not causally connected except through GA.

If so:

  • Causal Decision Theory says that if you find smoking enjoyable, you should do it because smoking doesn’t cause lung cancer.
  • Evidential Decision Theory says that you shouldn’t smoke, because then you will be the kind of person who has GA and will get lung cancer.

Now you say:

But the question is determining whether or not I have the desire, as it’s not unambiguous. As a neurochemical process, my choice to engage in exercise is not separable from the desire to engage in exercise but rather a manifestation of it, thus in the absence of obvious markers that would establish the causal connection I don’t know what exactly my genotype is; but if I act in accordance with the choices that would be made by someone with the favorable genes, I’d expect it to be a manifestation of better chances of having received the favorable genes in the first place, as I would be less likely to have made such a choice if my genes were unfavorable.

Not so.

If you didn’t know about GA, you would be correct – if you don’t know about GA and yet decide to avoid cigarettes, you are  probably the kind of person who doesn’t enjoy cigarettes, therefore probably the kind of person who doesn’t have GA, therefore probably the kind of person who won’t get lung cancer.

But you know about GA. You’re not avoiding cigarettes because you are “the kind of person who doesn’t enjoy cigarettes.” You are “the kind of person who doesn’t want lung cancer” which has zero correlation with GA, so your decision to avoid cigarettes is not evidence against the presence of GA in your genes.

The general insight to get here is the classic LW mantra: Probability is in the mind – the map, not the territory. By smoking or not smoking, you cannot change the actual territory of your genes.

Imagine if you could:

  • Reporter: Minister, is it true that you’ve just banned cigarettes?
  1. Minister: Yes, absolutely. The filthy things cause lung cancer.
  • Reporter: There is literally a gene that causes lung cancer, you’re banning utility for no good reason. What say you in your defense?
  1. Minister: AH BUT You see: The two are correlated! By forcing people to avoid cigarettes, we are forcing them to be the kind of person who doesn’t have the cancer gene!
  • Reporter: U wat m8?

Okay, so once the confusion on the original formulation (unambigous information) has been cleared, is the “weak” form of Smoking Lesion as invalid as the “strong”?

So let’s say that GA does somewhat impact the enjoyability of smoking, but not massively; a lot of people with GA don’t like smoking, and many people without do. It is somewhat strongly correlated though.

Then, let’s assume that I kind of enjoy smoking, and could take it or leave it if there is something actually significant, but without it I’d slightly prefer to smoke.

Then, let’s take into account that when I decide whether I smoke or not, I’m not a pure bayesian reasoner of perfect emptiness, but instead the biased piece of clockwork I am, doing some amount of reasoning and some amount of justifying the things my brain already has decided it wants to do. Newcomb’s problem is pretty much all about reasoning, but the question of whether I smoke or not is basically mostly system 1 because the margin is a lot fuzzier.

Thus, the release of endorphins I get from smoking is likely to have a tiebreaker role. In this situation I learn about GA and decide against smoking. On the margin this suggests that I’m less likely to carry GA because if I enjoyed smoking more than I do I’d choose to smoke. Since I do not know my genes I don’t know whether I carry it, but if I’m the sort of a person whose biased decision on the matter comes out against smoking, ending up being the sort of a person who doesn’t smoke is favorable.

Now a few points where it breaks down a bit:

Obviously the trick doesn’t work if I haven’t ever smoked, because then there is no link between my decision and my enjoyment of smoking. And the trick doesn’t work if I know my genes, so there is a highly suspiciously specific window where it could work if it could work at all.

Some kind of a “conservation of evidence” effect might be strictly limiting the effect it could theoretically have to be below the marginal utility from smoking, rendering it impossible even in the vague cases. In fact this sounds highly probable based on my utterly unscientific mathematical intuition; it would make a significant amount of sense for it to work like that.

Furthermore, the causal link in the chain of decision-making has to be from not enjoying smoking in the first place to not smoking, as any interference from decision theory is unrelated to the amount of enjoyment I get from smoking; this seems like it would tie very strongly to the above.

In other words, if there is no causal link from smoking to cancer, cancer concerns should not influence my decision to smoke, for me to be a legit non-GA-carrier.

So it basically boils down to “do what you want, aka. what you would’ve done without hearing about GA” which further boils down to “you’re lucky if you don’t enjoy smoking”. This seems like it would tie to the conservation of evidence because whether or not there is a decision to be made is already giving me evidence on my GA status; if everyone with GA gets cancer, and the prior is 50% and scales linearly with enjoyment of smoking so that all who love smoking are 100% GA and those who hate it are 0% GA, and I’m totally ambivalent about smoking it suggests that my cancer risk is 50%; if I subtly prefer smoking so that it feels like there’s anything that could be influenced, my cancer risk is higher already and letting GA concerns impact my decision is a strictly utility-losing proposition. If I’m non-GA, there is no temptation to smoke in the first place so the decision is moot; if I’m GA the desire is already there and its size is equivalent to the difference from prior, so if my preference for smoking is equivalent to 55% that’s what my cancer risk is and that’s what decision-theoretic trickery can’t impact. Yes. So the “weak” form is also invalid and GA should not impact my smoking decision. Question dissolved. Thank you everyone.

Now the part that has a potential to drive one quite nuts is:

Considering that reverting to a pre-GA status is not possible, how would one take action to correct for the bias injected by the GA thought experiment? If I used to smoke occasionally, then stopped because of GA concerns, and then found that smoking didn’t feel so fun anymore after all; should I try to smokehack to increase my enjoyment of smoking to compensate for the disutility of GA having slightly rewired my brain so that if my cancer risk is 60% but my smoking enjoyment is only 50%? (Assuming that everything else is equal; opportunity costs etc. are accounted for and so on)

I should probably not think too much about it and just enjoy the fact that my brain is like “but exercise is fun, we don’t want to reduce exercise” because it suggests that I’m likely to have good genes and get to enjoy not only exercise but also better health. It all is supposed to add up to sanity after all…

1: Yes, the weak form has the same weaknesses. Short version: Evidentiary decision theory relies on agents who do not update on evidence. You can replace all absolutes in my last post (Will/won’t) with 90% / 10% and still get the same outcome.

2: I am back in bed w/ ankle pain (Who even gets gout at age 31?) and on my tablet again so I’m not writing a longer answer to this one.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s