= 27d567bc2d48d9f40e9ccc20ea370b15)
Dmytry comments on Calibrate your self-assessments - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (117)
re: flirting, your underlying system cares for maximizing max. expected pay-off, it doesn't care for maximizing accuracy. Biologically, you get giant utilons loss for missing a mating opportunity (Think about it! For your underlying system, that's a huge fraction of worth of your life! It's the utility loss comparable to being forced to play Russian roulette with live bullets, several times in the row!), and microscopic utilon loss for being rejected. Your underlying system, though, doesn't trust your cognition with the correct mating odds - if it tells your cognition correct odds, you end up not trying, and not reproducing.
Our underlying system assigns massive utilon loss for being rejected. Far more than either current reproductive maximisation incentives or our own practical hedonistic interests would assign. ('Think about it'...)
You mean "hedon loss", right? Yeah, we do hate being rejected, but I'm not sure we should.
I mean the utilon calculation done by the crudely specified 'underlying system', which we infer based on observed behavior and reported experience. That's not quite the same as "hedon loss" but it is also a very different thing to "utilon loss" as could be described from the perspective of an anthropomorphized agent that seeks to maximise inclusive genetic fitness.
Yes, the thing that adds up to this!
Which doesn't do any utilon calculation, anyway. :-)
(Edited to add more links.)
Nah, it assigns massive punishment to the spokesman part of the brain coz it failed. Then the spokesman, in spare time, complains about it.
If it assigned utilon loss in the end, he wouldn't have been trying to talk to women despite getting rejected. Half of 'omg i dont have any consistent utility system' is the spokesman who's not deciding, whining about getting beaten by underlying system for failures. edit: i do believe in some localization of 'consciousness' in the sense that the part close to the speech centre is to some extent independent and unaware of what is going on in the rest. Brain is a distributed computing system, with non negligible spatial separation of components, and very non-negligible lag. I programmed that sort of systems, you have components working in partial ignorance of the state of other components. And if you are bandwidth limited you see how you can maximize the ignorance to free up bandwidth.
edit: Okay, to summarize for ya: you know how dogs are trained professionally? With treat and punishment method? Now consider one part of your own brain internally training other part of your own brain in precisely same way. With feelings of pain and feelings of reward. Positive and negative feelings are not utility, feelings are instrumental value that trains neural network to perform a task, adjusted for maximum efficacy (ideally).
The more or less mainstream 'just so' story here paints a somewhat different picture - and it is one that strikes me as fairly credible as far as these things go. As these things so often do it consists of an appeal to the Environment of Evolutionary Adaptation where, it is said:
In short, our instincts are calibrated for environments where a sexual advance being rejected has a far greater consequence than what it does for most people here and now.
A lot of the time guys don't end up trying to talk to women when doing so would seem to be the best way to satisfy their desires (in the 'maximise utility sense').
There is no such thing as environment of evolutionary adaptedness. Look at your foot. It's suboptimal in environment of evolutionary adaptedness and is only explainable by how it got there (from going on the tree and shortening important bones there).
I do agree that there are adaptations to that environment, but with this many free variables up to speculation on what the environment was, and on the method in one exactly adapts, and how much, one can explain literally anything, including anything that could later be shown to be product of conditioning. On top of this, you have this huge majority of people NOT taking rejections too seriously, which you can undoubtedly also explain by EEA in some other way, because it can explain everything. In fact I seen a zillion explanations of why and how we do one-night stands, and why women choose confident men for one night stands, out of some other EEA.
WRT the utility, pain is not utility. Pain is a value that trains your neural network not to do same stupid stuff again. If i do, or don't, spray with a water bottle my dog for chewing stuff up, (ideally) has nothing to do with utility of the items it is chewing up and everything to do with how much I think would get her not to chew stuff up, and the positive reinforcement not working for this task very well. Apply same principle to yourself feeling sort-of-pain for any internal reasons, and here you go.
WRT the brain being a distributed computing system with non-trivial lags, it is a plain fact, WRT distributed computing system's part having incomplete picture of the whole, that is plain how you get the distributed systems to work optimally. It doesn't give freedom to explain everything, in fact it has trouble explaining consciousness and pervasive sense that you are not a distributed computing system. Ditto goes for facts about 'how do you train a piece of brain'. With stick and carrot, that's how.
WRT maximizing utility, people of course suck at maximizing utility. But if you see utility of a belief for Yvain on the likehood of rejection, the best belief for him is that rejection is unlikely, a mis-calibrated one, because rest of the brain is mis-calibrated too. Evolution, by the way, doesn't make ideal systems. Nothing affects the prediction that Yvain is better off being overconfident about women. The rejection lasts for, maybe, day, the successful relationship where he DID dump his interests early (to cause early rejection to avoid wasting time), lasts for lifetime. He hyperbolically discounts for time. To act more optimally with hyperbolic discounting, he also has miscalibrated expectations. If he calibrates his expectations but does not get rid of hyperbolic discounting (which is much harder), he'll act even less optimally.
Most people do take rejections too seriously. In fact, for this reason, I recommend 'Rejection Therapy' (deliberately asking for stuff that you will not get) as an excellent personal development technique. (We tried it on the rationality boot camp last year.)
Particularly when it comes to mating based rejection people nearly universally experience anxiety grossly out of proportion to what is actually at stake - which is mostly the inconvenience of having to go and try again with someone else.
Well, I guess there's different definitions of 'most people' (rationality boot camp?) and 'too seriously' . The too seriously, is when you take it so seriously that you end up alone.
edit: there's also the 'too seriously to have 15 children by 10 women' kind of too seriously, on that i'd agree. With regards to anxiety being grossly out of proportion, that's meant to be compensated for by over-confidence. Yvain is not generally over confident. Just in those cases (flirting).
"\> 70% of single individuals of a suitable mating age in the world" is both a grossly conservative estimate and satisfies any remotely reasonable usage of 'most people' in the context.
Did you just try to spin that as an insult to those in my reference class? If I recall correctly all but two of the twenty participants were as of that time in at least one relationship. The lesson that should be taken is that even those who are already successful can serve to benefit from recalibrating their aversion to rejection downwards towards the optimal level.
That's certainly what you can expect in the extreme case. More often, however, people simply end up with fewer experiences with fewer people and must be satisfied with relationship arrangements that are perhaps less than they could have been. Or, at the least, must counter the aversive emotions that may otherwise have been an inconvenience while going after what they want despite their inhibitions.
If I were take the narrow group of my own friends and acquaintances and your suggested symptom "forever alone" the experiment would be biased in the other direction. Very few are single and an absolutely sickening proportion has gone and got themselves outright married.
[emphasis added]
Selection bias much?
There certainly could be, in such a class. Mind you I'd extend the 70% estimate to include all other individuals too - it'd just be slightly less of an understatement.
Nah, no insult meant beyond the sample being biased, which I trust we are all rational enough here not to take as an insult. I think it is fair to guess that you have mean IQ well over 100, which too is enough to ruin applicability of experiments.
I agree, actually. But see, there's the example of big problem for smart individuals in general: you do have that hyperbolic discounting, and you do have anxiety, you can't think them away, you must train them away, and that doesn't even fully work. Then you have your wild overestimate of the probability of success when some conditionals are met, and you can think this away easier.
Even if combined with the former it makes a fairly solid strategy, implementing a strategy on tweaked biases. See, there's also environment of 'cultural adaptedness', or 'memetic adaptedness', if you wish, and its at least hundred years back, and hundred years back it is NOT safe to hit on strangers unless some conditions (them flirting) are met. And it still works pretty well now. Not ideally - thanks to anti-murder laws, it is much safer to hit on strangers now - but it works. edit: actually, scratch that. Only in 5..10% of population it is perfectly safe to hit on strangers, and even there, you have a ton of harassment laws so even though you aren't likely to be beaten up, you can be screwed over.
edit: and for full disclosure, I also sucked at hitting on strangers. Its extremely uncomfortable. The point is that the overconfidence after conditions (flirting) are met is compensating, and makes a two way conditional implemented on biases: if (not flirting ) don't proceed ; if ( flirting) do proceed ; Hmm. To think about it maybe we came up with some truth here from disagreement. You can implement simple agents (similar to game AIs i can code) by combining biases, if you can tweak biases. And that is a plausible way how evolution can implement logic without wiring up individual neurons. The results are ultra messy though and have a ton of strange side effects, and become deregulated when one tries to get rid of some of the biases.
You should by now be aware that the claim (that I had previously assumed to be completely uncontroversial) is nothing to do with people at a particular training program (which related only to experiments with a solution) but rather with humanity in general. It isn't presented as the outcome of my own experiment but rather as a matter of both common and expert knowledge.