On sociology
(another long-in-progress draft) It’s interesting that much important analysis could be called “sociology” — eg., Robin Hanson’s Age of Em is essentially speculation on the structure of a future human society — and yet is divorced from sociology as an academic discipline.
Samo Burja has begun writing about how institutions, power, cooperation, and so on work in the modern world, but at least appears to have started from scratch, without requiring the reader to have any previous knowledge base (contrast how, eg., any physics paper would expect a general knowledge of quantum mechanics).
Although it’s harder to tell, because of their intense desire for secrecy, Leverage Research seems to have been doing a similar thing: developing their own theories of groups and societies, but starting from scratch and primary historical sources rather than building on some previous body of work.
There also don’t seem to be many famous living sociologists, even in the same way that there are famous living scientists or economists or psychologists. And sociological ideas don’t seem to leak into general popular discussion very much (compare eg. genetics in biology, personality testing in psychology, or stimulus spending in economics). From this pattern, I might infer that many questions which we’d like answers for aren’t being studied very productively, thus forcing people to start over again and again, rather than publishing small incremental updates the way a physicist or computer programmer would.
It would be great to be able to go beyond the thoughts of individual writers and develop some more generally applicable, more robust body of knowledge. Interdisciplinarity is such a cliche now that it pains me to say this, but managing the transition from a single intelligent species to many intelligent species is partly sociological. One could imagine, for example, developing a form of general AI that has been mathematically proven to be safe — no nuclear wars, no engineered plagues, no microscopic robots converting humans to computronium — while still not having any idea what life would be like after everyone woke up the next day. One obvious consequence is that various groups would immediately start racing to build their own AIs, and even without any global catastrophic risks, both stopping them and not stopping them would create radically different worlds than the one we live in now. (Iain Banks and Vernor Vinge try to grapple with this, but are of course limited by being individual thinkers with almost no real data to draw from.)
Here are some broad speculations for what might be done to escape the current trap: To the extent possible, standardize and clearly define terminology. Eg., Samo Burja uses “empire” to mean any hierarchical organization with a defined identity that seeks to expand and gain more power. That’s a reasonable enough concept, but this appears to be a novel word usage, which makes the problem of incompatible terminology worse (compare xkcd 927: Standards ). Other bad terminology is so verbose that it’s hard to communicate with (eg. “meta-relational semantics”), or causes horrible confusion by trying to re-define common words to mean something very different (eg. using “violence” to refer to subversive propaganda). Something like “corrigibility” from AI safety is good terminology; it’s easy to remember, while being a rare enough word that it can uniquely identify one particular idea.
Employ toy mathematical models to illustrate concepts. The idea of supply and demand curves in economics, for example, is a core foundation that also lets one make many useful predictions. Everyone knows they aren’t literally accurate in the same way as physics. For example, in theory, the additional demand for silver at one dollar under the current price is probably millions of pounds; while in practice, it would take a while for a store that sold at that price to get millions of orders. However, they let us predict that eg. price controls will probably create shortages eventually, and that’s good to be aware of every time California votes on a new ballot proposition.
Create better filtering mechanisms to quickly differentiate work that is both good and original. Sciences, like the medical world, use academic credentials and peer-reviewed journals, which… could be improved on, certainly, but you need something there. Otherwise, you get the blog equivalent of this list of treatments for cancer . (Maybe some of these have some effect! It’s not impossible. Good luck figuring out which ones all by yourself.)
Since real randomized trials are very hard, perform retroactive “experiments” by looking up data — the vast majority of which can be found somehow, but is not widely known or easily accessible. For example, the bureaucracy theory of Eroom’s Law would predict that the FDA application for Parnate (1961) was much shorter and had many fewer steps than the application for Viibryd (2011). I expect that this is true, I’d bet money on it, but I don’t actually know because I haven’t looked it up. If it was not true, that might falsify the theory.