@vgr

I sometimes wonder if we’re missing something fundamental about the interplay of decentralization and feature growth. Something like this syllogism: Major premise: Growth tips of complex technologies are always centralized for a while Minor premise: Complex technologies are always growing in feature space Conclusion: There is always centralization in complex technologies This would make the feasibility of full decentralization the same thing as the possibility of functional escape velocity and the feasibility of true ossification. I mean it seems like a path-dependent arbitrary thing… rpcs, sequencers, did servers, the US treasury market, there’s always *something* acting as a centralization bottleneck. The specific thing might be random, but not the fact that there is a bottleneck. Geometric intuition: “Progress” could be defined as the median vector of all the directions people want to go. Project positions onto it and some actor will always be “first” (even in a murmuration of starlings). If all actors vary at least a bit (and diversity/plurality is good, remember), some arbitrary feature of the leader will seem like a bottleneck temporarily. And even the smallest homogenizing feedback force (memetic envy, preferential attachment, hyperstitional beliefs from predictions) will make that arbitrary feature seem necessary. If you zoom out. Every new feature that starts must start with an n=1 implementation. It cannot evolve at infinite velocity so it must stay n=1 for a while. And even while its n=few, preferential attachment type feedback and random superior adaptive fit will ensure one instance dwarfs others. I think “sufficient decentralization” is ill-defined. What it should mean if I’m right: 1. No feature stays the centralization bottleneck too long 2. No two new features from the same originator become centralization bottleneck in a row You can have more than one centralization bottleneck at a time btw. When multiple actors monolpolize multiple features each of which is necessary for the whole. A kind of Mexican standoff crossed with a vetocracy, where a lot of actors can shut down or even destroy the system but no subset of actors can guarantee it will stay live. This might be worse than centralization with a SPOF. The US right now is not just a vetocracy, but one with a destruction-capable gridlock due to mechanisms like budget ceiling approvals. The deepest reason might be that our need for shared meaning causes us to converge towards homogeneous purposes and consensus definitions of progress. Variety and divergence always prevent full convergence in the long term, but in the short term there is always a convergent direction, a leader, and a centralized SPOF.

4 replies

1 recast

16 reactions