One useful analytic trick I've developed is of inverting propositions: "human interactions in general don't scale" gives us "how is it that human interactions
can and
do scale?"
Because one frame of history is to look at it as the story of precisely that scaling: family units to clans, clans to tribes, tribes to warlords or kingdoms, kingdoms to empires, the emergence of democratic republics (or of communist states, if you subscribe to an alternate arc), increasing scales of militaries, of religions, of commerce, of academia, etc., etc.
Each of these can be considered as networks, and there are some well-known models of network value.
The naive Metcalfe model states that network value grows with the square of the nodes.
A superior alternative was proposed by Ben Tilly (btilly at HN) and Andrew Odlyzko, which is that the value grows as n * log(n), that is, additional nodes add value, but comparatively less over time.
I've extended that to include a constant cost function, that is, one that's uniform for the network (at least on average), though it may change over time:
V = n * log(n) - k * n
This means that the value of the network grows
so long as k n is less than n * log(n). Put another way,
k constrains the maximum size of the network.
Your network will grow if you can reduce k and keep it small.
One way this manifests is as hygiene factors. Through the mid-19th century, the maximum size of a city was limited by its reproductive rate less its death rate plus net in-migration. As deaths typically outpaced live births (from disease and accidents, generally), cities needed to sustain large in-migration simply in order to maintain constant size. And the limit for 19th-century London was about 1 million people, roughly the size of ancient Rome at its peak. Deaths from epidemics such as cholera would claim tens of thousands of victims per year.
The solution was public health and hygiene, particularly the establishment of both clean fresh-water supplies, and of removing sewage waste. That latter was somewhat accidental: sewers built to drain away storm water were connected to by individual households as those installed flush toilets. The sewerage and storm water were drained away far to the east at the Thames Estuary.
Further advances came with food purity regulations, pasteurisation of milk, improved preservation and canning, refrigeration, increased bathing and handwashing. In the case of New York City (which saw a similar set of improvements) 85% of the reduction in mortality from 1850 to 2015 had occurred by 1920, which is before the introduction of modern antibiotics, most vaccines, organ transplants, medical imaging, and cancer treatments.[1][2] I'd first seen this pointed out by Laurie Garrett in the 1990s.
The further increases in the late 20th century largely come not from medical technology but rather access to medical care, and show far more strongly in under-served population (minorities generally, Black women, and especially Black men) than among the White population.[3] More recently Robert J. Gordon made this a major point of his analysis in The Rise and Fall of American Growth, noting an almost complete halt to medical progress (as measured by outcomes) beginning in the 1960s / 70s. (2015).
Put another way: one of the key reasons a site such as Facebook can grow to 5 billion MAU is because it tamps down hard on the systemic costs imposed by each additional member. In particular I've noted a general progression of decline as sites scale, with consistent patterns at roughly order-of-magnitude scales: 10, 100, 1k, 10k, 100k, 1m, and beyond. Networks have a number of characteristics: scale or size, topology (e.g., point-to-point, star, tree, mesh, hybrid), speed (the rate at which actions or messages propagate), etc. It's far easier to maintain control over a star (broadcast) network than a mesh (fully p2p) network, as the central node is the sole originator of content, and mediates all interactions. As I've describe elsewhere in this thread, hierarchical, modular or divisional structures tend to scale far better than monolithic ones. They do of course have their own limitations.
The key takeaway though is that scaling interactive networks, of any type, is largely a cost-minimisation function.
________________________________
Notes:
1. Graphically illustrated in "The Conquest of Pestilence in New York City" <https://1.bp.blogspot.com/-uTWEATUzgxk/TXQoTibILtI/AAAAAAAAA...> <https://economicspsychologypolicy.blogspot.com/2011/03/conqu...>
2. A surprising number of anti-cancer chemotherapy treatments can be traced to the chemical warfare compounds of World War I. <https://medicine.yale.edu/ycci/clinicaltrials/learnmore/trad...>
3. See: <https://www.usnews.com/news/blogs/data-mine/2015/01/05/black...> and "The gap between blacks and whites was seven years in 1990. By 2014, the most recent year on record, it had shrunk to 3.4 years, the smallest in history, with life expectancy at 75.6 years for blacks and 79 years for whites." <https://www.nytimes.com/2016/05/09/health/blacks-see-gains-i...>