Sociocybernetics & Social Systems Theory
This cluster traces a single, decade-spanning question through four very different vocabularies: what happens when systems-theoretic and cybernetic thinking is asked to account for a digitally mediated society whose “actors” increasingly include non-human, adaptive, and networked entities? The papers filed here move from infrastructure to epistemology to agency to formal multi-agent theory, but they share a common refusal to treat the social as reducible to individual actors’ properties — insisting instead that structure, observation, and network position are constitutive.
From infrastructure to social capital
Sofia2012-802a21dc is the earliest and most infrastructural entry, arguing that Internet architecture itself must be redesigned around social capital principles rather than purely technical metrics. Its core move — importing sociological constructs (trust, reach, influence, engagement, Putnam-style social capital) into a domain governed by networking metrics (degree, betweenness, closeness) — is a direct precursor to the later papers’ insistence that social-theoretic categories cannot simply be read off technical or behavioral data without translation. The paper’s central tension, that networking and social-capital notions of centrality diverge (a high-betweenness node is a “bridger” socially but a mere relay technically), anticipates a recurring worry in this cluster: metrics borrowed from one domain misrepresent structure when applied naively to another. This is the seed of a sociocybernetic sensibility applied to the Internet as architecture rather than as content or discourse.
Cybernetic epistemology and the propagation of meaning
Giglietto2019-e9be81c1 shifts the register from infrastructure to epistemology, drawing explicitly on second-order cybernetics (von Foerster, Bateson) and Luhmann to reframe “fake news” not as a property of content or creator intent but as an emergent, observer-dependent phenomenon of propagation. Its micro–meso–macro model — individual truth judgments, a matrix of injector/propagator combinations, and emergent macro-level cascades — operationalizes the systems-theoretic principle that meaning is constructed through communication rather than transmitted intact. This is the cluster’s clearest statement of classical sociocybernetic method: information is “a difference that makes a difference,” and global patterns (cascades) are autonomous emergent objects irreducible to any single actor’s intent, even though they are built entirely from actors’ local judgments. Where Sofia2012-802a21dc worried about misapplied structural metrics, Giglietto worries about misapplied causal/intentional narratives — both diagnose a mismatch between the level at which meaning is produced and the level at which it is analyzed.
The demarcation problem: agency as social status
Anicker2024-vp pushes the systems-theoretic move further by relocating agency itself outside the individual system. Rather than asking whether an AI “has” agency as an intrinsic property, it treats agency as a socially granted and monitored license — a status conferred and revocable through social practice. This is a direct descendant of the constructivist epistemology in Giglietto2019-e9be81c1 (truth and information as observer-relative) applied to a harder problem: not what is true, but who counts as an actor. It also sets up the conceptual ground that Ng2026-og needs and largely bypasses: Anicker’s demarcation problem asks whether AI systems can be agents at all in social-theoretic terms, while Ng et al. presuppose agentic status and ask how populations of such agents behave once embedded in networks.
Formalizing multi-agent social systems
Ng2026-og is the cluster’s most recent and most formal contribution, and in many ways closes the arc opened by Sofia2012-802a21dc: it insists that agentic AI, like the prosumer-driven Internet before it, cannot be understood through individualist, task-level assumptions but must be modeled as a networked, co-evolving social system. Its (f, g, G) triple — information exchange, influence dynamics, interaction network — is a formalization of exactly the structural intuitions that Giglietto2019-e9be81c1’s micro-meso-macro cascade model and Sofia2012-802a21dc’s social-capital metrics gesture at more informally. The four structural priors (strategic heterogeneity, network-constrained dependence, co-evolution, distributional instability) are essentially cybernetic claims — dependence, feedback, non-equilibrium — restated for a population of LLM-driven agents, and empirically grounded in a real large-scale agent social network (MoltBook) rather than simulation. Notably, Ng et al. sidestep Anicker’s demarcation problem by fiat, treating agentic status as given and focusing instead on population-level dynamics; this makes the two papers complementary rather than redundant — one asks whether these entities count as agents, the other asks what follows structurally once they do.
The arc
Read together, the four papers trace a migration of sociocybernetic concern: from the material substrate of the network (Sofia2012-802a21dc), through the epistemology of meaning and truth circulating on it (Giglietto2019-e9be81c1), to the question of who or what is licensed to act within it (Anicker2024-vp), to a formal theory of how populations of such licensed actors co-evolve at scale (Ng2026-og). The throughline is a consistent skepticism toward individualist and purely technical framings, replaced by an insistence — recognizably Luhmannian in spirit even where not explicitly cited — that social systems, human or machine, must be understood through structure, communication, and emergent macro-dynamics rather than through the isolated properties of their parts.