On Sat, Jun 8, 2024 at 9:27 PM Matt Mahoney <mattmahone...@gmail.com> wrote:
> How do you use the algorithmic information criteria to choose between > social theories with the same number of bits, like "poverty causes crime" > and "crime causes poverty"? > That's a very degenerate case but your valid point is related to algorithmic probability: The shorter the algorithm that outputs the same dataset is simply more likely. The best model of a given dataset has a length equal to the dataset's Kolmogorov Complexity. If there are two such models then it's a tie. > I can use the correlation in the distribution P(crime, poverty) from your > LaboratoryOfTheCounties benchmark to improve compression. But to show which > causes which, I would also need P(crime) and P(poverty). > As I said in the intro to Hume's Guillotine <https://github.com/jabowery/HumesGuillotine>, data and models of data are qualitatively different, therefore the criteria for selecting them are qualitatively different. Data selection is *inescapably* subjective if for no other reason than that one person may be interested in poverty and other interested in crime and another interested in their relationship. Given the subjective judgement that the relationship is interesting, your use of the word "need" is also subjective in the sense that one might have limited resources with which to gather data but be able to find those variables of interest latent in the data that one has at hand. William M Brigs addresses the dangers of this at 19:30 in this video <https://x.com/i/status/1797668902304141743>, which is a well-informed polemic on what is wrong with "causality" in the social sciences. Having said that, there is no clear distinction between "exploratory" data analysis and other kinds of data analysis -- so one may have to put up with proxies in order to see if there is enough "signal" there to justify investment in better data. Moreover, as we've seen with ML models trained on "large" data sets, all manner of latent variables manifest with grokking aka "emergent" intelligence. It just doesn't pay to specialize your model for one particular "task" such as predicting one particular "dependent variable" leads to social pseudoscience myopia. The world is a complex dynamical system and attempts to model it in any manner other than with *cyclic* causality leaves you in the same state ML was in when everyone was focused on feed-forward supervised training with labeled data. > You can only get that using controlled experiments, like the one in Sweden > where they found that lottery winners didn't commit fewer crimes. > Aside from human experimental subjects and all of the ethical issues (that gave rise to my proposing the meta-polity of sorting proponents of social theories into governments that test them) even assuming one is not dealing with human (or even animal) subjects, one must still decide which experiments to run with limited resources. That's where model selection comes in. Do your data analysis to produce a model (ie: losslessly compress your data) so you have a hypothesis to test (counterfactual based on the best model) before you even design your experiment. > Also, what are you saying about sex? The population is collapsing because > of birth control, the right to use it, and that most people don't care what > happens after they die. AI will make this worse by isolating people, but > this started even before the internet. > This is your social theory. While I have agreed with *aspects* of it ever since I was in high school writing science fiction back in the early '70s <https://jimbowery.blogspot.com/2016/09/johnny.html>, it is not _my_ social theory. Whose social theory is right? Yammering at each other doesn't help. The existing dispute processing systems -- where central government imposition of social policy is the norm -- are designed to make us yammer at each other while we seek to impose our respective social theories on everyone else as a matter of self defense. Either that or meekly accept whatever "the science" or "the authorities" or "the powers that be" decide to do with our sorry asses. This is what is going to get us killing each other by the tens of millions in the not too distant future. So I'm offering two ways out: 1) If "the science" or "the authorities" or "the powers that be" just can't tolerate sorting proponents of social theories into governments that test them -- and they just HAVE to cram whatever they believe is the TRUE social theory down our collective throats *for our own good* (of course) -- they have the AIC. 2) If they can't bring themselves to apply the AIC, and they insist on imposing their social pseudoscientific experiments on unwilling human subjects, then it will become necessary to start disabling the cooling systems of the data centers until they relent and permit us to sort proponents of social theories into governments that test them. And don't think for a second that this can't be done with the portion of the population that is willing to die in a so-called "civil war". Sex evolved because you can write DNA code faster with random cuts and > pastes than just random bit flips. Reproductive behavior is complex because > it is our most important function. Humans are the only mammals that don't > go into heat or that use nudity for sexual signaling, and the only mammals > besides prairie voles that fall in love. All of this evolved after we split > from chimpanzees 6 million years ago. But male aggression evolved before > that. 95% of both homicides and chimpisides are committed by males. > > Government programs intended to encourage reproduction aren't working. I > suppose we could develop the technology to produce babies in factories, but > what would be the point? If people wanted children, robots would be easier > to care for. We will either evolve to reject technology or create the > species that replaces us. > > > On Sat, Jun 8, 2024, 4:24 PM James Bowery <jabow...@gmail.com> wrote: > >> >> >> On Fri, Jun 7, 2024 at 8:51 PM Matt Mahoney <mattmahone...@gmail.com> >> wrote: >> >>> ... >>> Evolution selects for cultures that reject technology and women's >>> rights. I disagree, but I will also die without offspring. >>> >> >> Evolution selects for sex, and sex selects for women's rights *and* for >> technology, but since "we" *have no word for sex* it is difficult to >> discuss what evolution selects for. >> >> "We" have no word for sex because the word "we" designates an asexual >> group organism that finds sex threatening to its integrity. So it >> suppresses sex. This is related to why "queens" parasitically castrate >> their offspring in eusocial species. >> >> So.... what *is* sex, that we are not to even *talk* about it? >> >> The evolutionary platform that gave rise to the Cambrian Explosion was >> not fully formed until individual vs individual masculine aggression arose >> as the individual organism's counterbalancing choice to the individual >> feminine choice of nurturance. *That* is sex and *that* is why eusocial >> organisms castrate offspring to produce sterile workers specialized as are >> the various asexual cells that make up specialised organ tissues. >> >> And now we're seeing the loss of life's meaning throughout technological >> civilization as total fertility rates plummet to suicidal levels. Everyone >> has their go-to cope "explanation" for this suicidal trend, but no one >> wants to reform the social pseudosciences with the Algorithmic Information >> Criterion for causal model selection -- and for the same reason that they >> don't want to recognize that they owe their very nervous systems to sex. >> > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/Ta32348ecc8396e1a-Mcbcde25830bb20aae073530e> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Ta32348ecc8396e1a-M56df810063e7bef236fc2680 Delivery options: https://agi.topicbox.com/groups/agi/subscription
