In my experience as a subject of hypnosis, I always have a background thought that I could choose to not do/feel the thing, that I choose to do/feel as I'm told. I distinctly remember feeling the background thought there, before choosing to do, or letting myself feel, the thing I'm told. It is still surprising how much and ho many things that are usually subconscious can be controlled through it, though.

If your theory leads you to an obviously stupid conclusion, you need a better theory.

Total utilitarianism is boringly wrong for this reason, yes.

What you need is non-stupid utilitarianism.

First, utility is not a scalar number, even for one person. Utility and disutility are not the same axis: if I hug a plushie, that is utility without any disutility and if I kick a bedpost, that is disutility without utility, and if I do both at the same time, neither of those ends up compensating for each other. They are not the same dimension with the sign reversed. This is before going into the details where, for example, preference utilitarianism is a model where each preference is its own axis, and so is each dispreference. Those axes are sometimes orthogonal and sometimes they trade off against each other, a little or a lot. The numbers are fuzzy and imprecise, and the weighting of the needs/preferences/goals/values also changes over time: for example, it is impossible to maximize for sleep because if you sleep all the time, you starve and die and if you maximize for food then you die of eating too much or never sleeping or whatever. We are not maximizers, we are satisficers, and trying to maximize any need/goal/value by trading it off against all the others leads to a very stupid death. We are more like feedback-based control systems that need to keep a lot of parameters in the good boundaries.

Second, interpersonal comparison are between hazardous and impossible. Going back to the example of preference utilitarianism, people have different levels of enjoyment of the same things (in addition to those degrees also changing over time intrapersonally).

Third, there are limits to the disutility that a person will endure before it rounds off to infinite disutility. Under sufficient torture, people will prefer to die rather than bearing it for any length of time longer; at this point, it can be called subjective infinite disutility (simplifying so as to not get bogged down in discussing discounting rates and limited lifespan).

Third and a halfth, it is impossible to get so much utility that it can be rounded off to positive infinity, short of maybe FiO or other form of hedonium/orgasmium/eudaimonium/whatever of the sort. It is not infinite, but it is "whatever is the limit for a sapient mind" (which is something like "all the thermostat variables satisfied including those that require making a modicum of effort to satisfy the others", because minds are a tool to do that and seem to require doing it to some, intra- and interpersonally varying, extent).

Fourth and the most important point to refute total utilitarianism, you need to account for the entire distribution. Even assuming, very wrongly as explained above, that you can actually measure the utility that one person gets and compare it to the utility that an other person gets, you can still have the bottom of your distribution of utility being sufficiently low that the bottom whatever% of the population would prefer to die immediately, which is (simplified) infinite disutility and can not be traded for the limit of positive utility. (Torture and dust specks: no finite amount of dust specks can trade off for the infinite disutility of a degree of torture sufficient to make even one single victim prefer to die.) (This still works even if the victim dies in a day, because you need to measure over all of the history from the beginnning of when your moral theory begins to take effect.) (For the smartasses in the back row: no, that doesn't mean that there having been that level of torture in the past absolves you from not doing it in the future under the pretext that the disutility over all of history already sums to infinity. Yes it does, and don't you make it worse.)

But alright. Assuming you can measure utility Correctly, let's say you have the floor of the distribution of it at least epsilon above the minimum viable. What then? Job done? No. You also want to maximize the entire area under the curve, raising it as high as possible, which is the point that total utilitarianism actually got right. And, in a condition of scarcity, that may require having not too many people. At least, having the rise in amount of people being slower than the rise in distributable utility.

didn't we use to call those "exokernels" before?

I'm curious who the half is and why. Is it that they are half a rationalist? Half (the time?) in Berkeley? (If it is not half the time then where is the other half?)

Also. The N should be equal to the cardinality of the entire set of rationalists you interacted with, not just of those who are going insane; so, if you have been interacting with seven and a half rationalists in total, how many of those are diving into the woo? Or if you have been interacting with dozens of rationalists, how many times more people were they than 7.5?

There was a web thing with a Big Red Button, running in Seattle, Oxford (and I think Boston also).

Each group had a cake and if they got nuked, they wouldn't get to eat the cake.

At the time when the Seattle counter said that the game was over for 1 second, someone there puched the button for the lulz, but the Oxford counter was not at zero yet and so they got nuked, then they decided to burn the cake instead of just not eating it.

• Aumann's agreement theorem says that two people acting rationally (in a certain precise sense) and with common knowledge of each other's beliefs cannot agree to disagree. More specifically, if two people are genuine Bayesian rationalists with common priors, and if they each have common knowledge of their individual posterior probabilities, then their posteriors must be equal.

With common priors.

This is what does all the work there! If the disagreeers have non-equal priors on one of the points, then of course they'll have different posteriors.

Of course applying Bayes' Theorem with the same inputs is going to give the same outputs, that's not even a theorem, that's an equals sign.

If the disagreeers find a different set of parameters to be relevant, and/or the parameters they both find relevant do not have the same values, the outputs will differ, and they will continue to disagree.

"the problem with lesswrong : it's not literally twitter"

Thank you so much for writing this! I remember reading a tumblr post that explained the main point a while back and could never find it again -because tumblr is an unsearchable memory hole- and kept needing to link it to people who got stuck on taking Eliezer's joking one-liner seriously.