Search the Community

I decided to take a look at Carroll Quigley's "The Evolution of Civilizations" not because it is considered a prerequisite for discourse about civilization in these fora, but because, for some time, my own focus is on the abysmal state of the social sciences qua sciences. The social sciences are so abysmally unscientific that it is a revolutionary act of genius for anyone to bring to bear anything remotely resembling scientific method. Moreover, if one attempts to bring the social sciences into consilience with the larger body of human knowledge, one is attacked with religious fervor as evidenced by the treatment of E. O. Wilson by his Harvard colleagues in the 1970s over the nascent field of sociobiology. So Quigley was a revolutionary genius -- not so much because he offered anything fundamentally new, but simply because he spoke of a few obvious truths about science in a field of virtually universal deceit: the social sciences of the 20th century. Quigley's approach, however admirable given the horrid context, can too-easily lead one to accept premises of his which have subsequently shown themselves to be both scientifically inadequate and ethically vacuous. First, and foremost, the ethical vacuity on display in "The Evolution of Civilizations" is shared by the entire field of the social sciences. It may reasonably be summed up by comparing the ethics of medicine. In medicine, even if one has conducted double blind controlled studies of the safety and efficacy of a treatment (ie: one has established strong evidence the treatment causes beneficial effects) -- even then it is considered unethical to apply the treatment to human subjects without their informed consent. Accepting Quigley's proclamation that control experiments cannot be conducted in the social sciences to establish causality, the first duty of the ethical social scientist should be to denounce the use of his findings in a way that would violate the informed consent of human subjects in social engineering. Let me re-emphasize in stronger terms: Quigley, is not only not alone in this absence of ethics among social scientists, his posture is universally de rigueur. Nevertheless, those who hold Quigley up as an exemplar, however justified, have an ethical obligation to point out this ethical vacuity. Secondly, Quigley, himself, describes the social science equivalent of statistical mechanics -- averaging large numbers to make predictions. At the same time, he goes to great lengths in his discourse about "human nature" to emphasize that "culture" determines, for practical purposes, the outcome for statistically significant numbers of individuals. This is, essentially, the Boasian dogma of 20th century anthropology. It is upon this basis that we have seen the diagnosis of "institutional racism" held up as the "explanation" for statistical outcome differences between racial groups. This, in turn, has expended many trillions of dollars in social engineering projects spanning over a half century with outcomes that are, at best, questionable and, in any event, violate the scientific ethics of informed consent when treating human subjects, as described above. Having now made my essential critique of Quigley's otherwise reasonable premises, I want to point out what he got _very_ right in his presentation of scientific method, and how, with modern advances in universal intelligence based on mathematically defining Ockham's Razor in pursuit of automated science, we may be in a position to push beyond Quigley's limits. Ray Solomonoff essentially proved Ockham's Razor as essential to science in terms of computer theory and did so at the dawn of the computer age. However, over a half century into the computer age, we still aren't even beginning to exploring those implications in a practical way. Here's an obvious implication that should have been pursued almost from the outset in the 1960s: Whenever you have a dataset and are trying to come up with a predictive model, you have two basic options that avoid overfitting: Use the data you have, not to create the model but to test it. Approximate the data's Kolmogorov Complexity program as best you can so as to approximate Solomonoff Induction. #1 invariably ends up being impractical since you can't _really_ construct a model out of first principles. In any event, as you start to "consume" your data in tests of your models, you end up refining your models which gets you into the land of post hoc theorization thence overfitting as you consume more data. The best you can do is what Enlightenment philosophers came up with: Experimental controls -- which is to say, you have experimental setups, all identical except being treated in a slightly different way (including no treatment called "the control"). The social sciences have become the modern equivalent of a theocracy given their impact on public policy -- but social scientists haven't reached the level of scientific ethics required for them to insist that their theories not be taken as justification for imposing experiments on massive human populations as is required by Federal arrogation of social policy from the Laboratory of the States. If social scientists had anything worthy of being called "ethics" they would insist on devolution of social policy to the States and Federal support of migration of people to the States whose social policies they find mutually agreeable. This directly addresses the scientific need for experimental variation as well as the ethical need for informed consent when dealing with human experimental subjects. In the absence of such humility, the social sciences did have one other option: Data compression to approximate Kolmogorov Complexity. Note that I am not here talking about a general algorithm for data compression. I'm talking about a much simpler and obvious idea: Comparing theories by how well those theories -- losslessly -- compress the same datasets. And this is where I come to my perception of a "religious aversion" to Solomonoff Induction: Whenever I see arguments against the utility of Solomonoff Induction in the aforementioned role -- comparing theories by the size of executable archives of the same datasets -- they are _invariably_ (in my experience) strawman polemics. Yes, Kolmogorov Complexity is incomputable -- but that's not the argument. We're not trying to come up with a program to compress the datasets! There is a difference between a program that compresses the datasets and a program that DEcompresses the datasets (the latter being the approximation of the KC program). This difference is so obvious that its conflation in these arguments -- its _predicatable_ conflation -- is reminiscent of Orwell's notion of "Crime Stop": Selective stupidity to avoid violations of Ingsoc or the official ideology of The Party. There are other, less obviously stupid, strawmen that arise from time to time but these are almost invariably in the category of philosophical attacks on Cartesianism or the scientific method itself. While it is fine to have those philosophical arguments, it seems rather silly to hold up practical application of Solomonoff Induction on that basis as virtually the entire structure of technological civilization is Cartesian.