Abstract:Machine collectives increasingly coordinate, reciprocate, and form shared conventions on their own, tempting us to call them societies like ours. We argue that this conflates two registers of social order and misses what the human one is for. Human sociality is the way a finite, natal, generational form of life organises its own finitude: members die with their tacit knowledge, newcomers start ignorant, and cohorts must hand on what they cannot keep. Kinship, inheritance, teaching, and much of obligation and trust are the forms this takes, and machine collectives reproduce only the residue once finitude is subtracted. We meet the objection that machines already have death with two distinctions. Autonomy versus heteronomy: a death an operator can reset, roll back, or copy around is not constitutive finitude; the test is resettability. Representation versus binding force: humans learn finitude from others' deaths, so its representation is machine-learnable, but its grip on motivation needs the learner's own end to be inescapable. Treating machines as a model organism, a controlled experiment varying only whether death means loss shows that cumulative, transmissible culture arises only under irreversible loss; a copyable-immortal population is more capable yet culturally empty. Across a machine population and three real human genealogies, copying over-transmits status relative to every human regime and fragments lineage like blood descent, whereas externalisation lowers transmission into the human range and connects lineage like intellectual descent. In a real frontier model, language-model end-game defection vanishes when the game is de-labelled, a sign of recognition, not a mechanism. The gap between a machine collective and a human society is therefore not one of intelligence but of finitude, and it closes only when finitude is built in.
From: Canhui Liu [view email]
[v1]
Fri, 12 Jun 2026 00:12:45 UTC (728 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。