Re your link "Divide and conquer: intermediate levels of population fragmentation¹ maximize cultural accumulation":²
> "In this paper, we review the literature from various fields that suggest that, under some circumstances, increased connectedness can decrease cultural diversity and reduce innovation rates."
Related from social epistemology: the Zollman effect,³ which is when an epistemic community diverges from truth (or converges slower) when you increase their connectedness (or rate of sharing information). Computer simulations don't do it justice imo. The sims I've seen also don't seem to measure the notion of innovation-rate, as separate from "convergence to truth". E.g., in an actual community, you may wish to increase the precious few sources of noise/chaos you have, bc you can't discover anything novel without breaking up order from time to time.
If we look at the brain as an analogy (risky analogy tho this is), there's the Critical Brain Hypothesis,⁴ which suggests that the brain's homeostatic mechanisms are designed to keep the network in a constant state of "criticality, at the border between order and chaos, bc that tends to be the optimal balance between exploiting current best models (ie consolidation) vs exploring new ones (ie noise).
Reasons I expect cultural evolution to be much much less efficient (along some dimensions at least) than eg the brain or even DNA include:
- Cultural evolution has a very close analogy with language evolution, and people are reluctant to use new words unless they already expect other ppl to know those words. Thus, new words usually require common knowledge of them to emerge before the word gets used (sorta analogous w "epistasis" in the genetics context). There are some ways around this constraint (eg via a centralised node for generating common knowledge, like Hollywood, the government; or via specific dimensions along which innovation is viewed positively, like frivolous memes or designated topics du jour), but they tend to be inefficient and narrow.
- Sentences (or other units of cultural evolution) are already extremely chunked/high-abstraction relative to the neural assemblies that generated them. Thus the dimensionality on which cultural evolution can generate variety is much lower, and this leads to fewer smooth paths down the loss-landscape. Thus you're more likely to get stuck in spurious local minima, or get unlucky and traverse down a very long and very slow incline.
- Both of the above points lead to technical debt (which I call "dependency debt" in my notes, to make clear that it's a very general problem not restricted to programming), which is when the cost of refactoring complex (inadequate) equilibria increases superlinearly as the complex system evolves via smth like hill-climbing.
¹ In social epistemology context, I usually say "bubbliness" instead of "fragmentation", bc the former is a cuter word, and it associates to "social bubbles" which is a notion we already have some lived experience with.
> Thus, new words usually require common knowledge of them to emerge *before* the word gets used ...
Another thing this makes me think about: sometimes to even form a new community, you need a large, connected group of people to draw from. You need to find people to form a community around something, and a lot of niche interests might not even exist if people couldn't find each other via the internet.
Yeah, "epistasis" is a generalisation of "catch-22". When the beneficial trait of a specific new mutation depends on the existence of another specific mutation and vice versa, the species is exceedingly unlikely to evolve that trait bc it would require both genes to mutate in a single individual at once.
Scientific paradigms evolve like this, I think. Interest in a particular research direction depends on the scientist's ability to converse about that direction w other scientists. So progress is dimensionally constrained by "conversation distance". Most of the time. Unless your interest is unusually socially-independent, and you can write the Sequences to help ppl catch up. Happens.
Disclaimer: Wikipedia uses the term "epistasis/epistatic" a bit differently (seems like a bad way to do the ontology imo…), but I'm not sure what else to call it. I'm referring to the same concept from a different perspective afaict.
Re your link "Divide and conquer: intermediate levels of population fragmentation¹ maximize cultural accumulation":²
> "In this paper, we review the literature from various fields that suggest that, under some circumstances, increased connectedness can decrease cultural diversity and reduce innovation rates."
Related from social epistemology: the Zollman effect,³ which is when an epistemic community diverges from truth (or converges slower) when you increase their connectedness (or rate of sharing information). Computer simulations don't do it justice imo. The sims I've seen also don't seem to measure the notion of innovation-rate, as separate from "convergence to truth". E.g., in an actual community, you may wish to increase the precious few sources of noise/chaos you have, bc you can't discover anything novel without breaking up order from time to time.
If we look at the brain as an analogy (risky analogy tho this is), there's the Critical Brain Hypothesis,⁴ which suggests that the brain's homeostatic mechanisms are designed to keep the network in a constant state of "criticality, at the border between order and chaos, bc that tends to be the optimal balance between exploiting current best models (ie consolidation) vs exploring new ones (ie noise).
Reasons I expect cultural evolution to be much much less efficient (along some dimensions at least) than eg the brain or even DNA include:
- Cultural evolution has a very close analogy with language evolution, and people are reluctant to use new words unless they already expect other ppl to know those words. Thus, new words usually require common knowledge of them to emerge before the word gets used (sorta analogous w "epistasis" in the genetics context). There are some ways around this constraint (eg via a centralised node for generating common knowledge, like Hollywood, the government; or via specific dimensions along which innovation is viewed positively, like frivolous memes or designated topics du jour), but they tend to be inefficient and narrow.
- Sentences (or other units of cultural evolution) are already extremely chunked/high-abstraction relative to the neural assemblies that generated them. Thus the dimensionality on which cultural evolution can generate variety is much lower, and this leads to fewer smooth paths down the loss-landscape. Thus you're more likely to get stuck in spurious local minima, or get unlucky and traverse down a very long and very slow incline.
- Both of the above points lead to technical debt (which I call "dependency debt" in my notes, to make clear that it's a very general problem not restricted to programming), which is when the cost of refactoring complex (inadequate) equilibria increases superlinearly as the complex system evolves via smth like hill-climbing.
¹ In social epistemology context, I usually say "bubbliness" instead of "fragmentation", bc the former is a cuter word, and it associates to "social bubbles" which is a notion we already have some lived experience with.
² https://splittinginfinity.substack.com/p/links-6
³ https://jonathanweisberg.org/post/zollman/
⁴ Great youtube video about Critical Brain Hypothesis: https://www.youtube.com/watch?v=vwLb3XlPCB4&list=PLgtmMKe4spCMzkiVa4-eSHVk-N4SC8r9K&index=13
> Thus, new words usually require common knowledge of them to emerge *before* the word gets used ...
Another thing this makes me think about: sometimes to even form a new community, you need a large, connected group of people to draw from. You need to find people to form a community around something, and a lot of niche interests might not even exist if people couldn't find each other via the internet.
Yeah, "epistasis" is a generalisation of "catch-22". When the beneficial trait of a specific new mutation depends on the existence of another specific mutation and vice versa, the species is exceedingly unlikely to evolve that trait bc it would require both genes to mutate in a single individual at once.
Scientific paradigms evolve like this, I think. Interest in a particular research direction depends on the scientist's ability to converse about that direction w other scientists. So progress is dimensionally constrained by "conversation distance". Most of the time. Unless your interest is unusually socially-independent, and you can write the Sequences to help ppl catch up. Happens.
Disclaimer: Wikipedia uses the term "epistasis/epistatic" a bit differently (seems like a bad way to do the ontology imo…), but I'm not sure what else to call it. I'm referring to the same concept from a different perspective afaict.