Teaser

Long before “foundation models,” Manuel Castells argued that we live in a network society where power flows through programmable networks and the codes that route communication. Read from today’s AI moment, much of his analysis lands with force—though some parts need updating for a world of centralized model stacks and compute bottlenecks.

Methods window

Assessment target: BA Sociology (7th semester) — Goal grade: 1.3 (Sehr gut).
Approach. Conceptual reconstruction (space of flows vs. places; programmability; communication power; identity forms) and evaluative comparison with present AI developments (platformization, generative models, regulation). APA short style in text; publisher-first links below.
Theory anchors. Castells (2010; 2009; 2015; 2001).

---

What Castells got right (strong matches)

1. Power as communication power.
   Castells claimed that contemporary power operates by coding and routing communication across media networks (Castells 2009). AI now performs exactly that: ranking, labeling, and generating messages at scale, while policy “grammars” and recommender codes institutionalize attention. (global.oup.com)
2. Programmable networks.
   He argued networks are programmable—goals and rules can be rewritten by those who control infrastructure (Castells 2010). Today’s AI stacks (APIs, model checkpoints, safety layers) are re-programmable layers that steer what travels furthest. (Wiley)
3. Mass self-communication.
   Castells anticipated user-driven broadcasting—people producing to networks, not just consuming from mass media (Castells 2009). Generative tools lower entry costs further, extending mass self-communication into mass self-generation. (global.oup.com)
4. Networked movements and identity work.
   His account of networked social movements and project/resistance identities maps well onto 2010s–2020s mobilizations; AI now accelerates translation, subtitling, and routing of identity narratives (Castells 2015). (Wiley)
5. Space of flows over space of places.
   The primacy of flows (remote coordination, cloud infrastructures) remains a good first approximation for AI’s development and deployment regimes (Castells 2010; 2001). (Wiley)

---

What partly matches (needs refinement)

1. Decentralization vs. platform concentration.
   Castells’s open-ended “programmability” partly assumed diverse actors could reprogram networks. In AI, compute, data, and distribution are highly centralized (few cloud providers; app-store gates). The concept still holds—but requires an explicit political-economy layer about market power and who controls the switches (Castells 2009). (global.oup.com)
2. From access to routing rights.
   He pushed the debate beyond connectivity toward routing and amplification. That diagnosis fits; yet today legal regimes (DSA; EU AI Act) are becoming co-programmers by mandating transparency, risk controls, and appeal channels—an institutional dimension that only sat in the background of the original theory (Castells 2009; Regulation (EU) 2022/2065; Regulation (EU) 2024/1689). (EUR-Lex)
3. Space of flows vs. places under material constraints.
   Castells highlighted “flows,” but AI’s material anchors—data centers, energy grids, supply chains—re-assert places (zoning, chips, power). The framework adapts if we treat places as choke points inside flows rather than their negation (Castells 2010; 2001). (Wiley)
4. Identity and authenticity.
   His optimism about self-communication needs a caveat: synthetic media stretches authenticity and provenance problems. The logic of identity politics still follows Castells; the signal quality now depends on provenance infrastructure not theorized in early work (Castells 2015). (Wiley)

---

What likely does not match (or must be rethought)

1. Durably plural programmability.
   In practice, AI programmability is often enclosed—guardrails, datasets, and weights sit behind proprietary boundaries. Programmability exists, but not equally across actors; without governance, switching/programming power skews elite (Castells 2009). (global.oup.com)
2. End of mass media.
   Castells emphasized the shift to self-communication; today, AI has re-centralized distribution via a small set of model and platform gateways. Mass media didn’t end; it merged with platforms and model intermediaries that gate training data and output reach. (Update needed relative to Castells 2001; 2010.) (OUP Academic)
3. Regulation as afterthought.
   Where Castells foregrounded network dynamics, binding regulation now actively shapes AI architectures (risk tiers, transparency, redress). Any contemporary Castellsian analysis must treat the EU AI Act and DSA as constitutive code for communication power. (EUR-Lex)

---

A Castells-inspired agenda for AI sociology

• Follow the switches. Map who owns APIs, training pipelines, app stores, and cloud interconnects—that’s where communication power condenses (Castells 2009). (global.oup.com)
• Trace code paths. Document how inputs are routed and transformed—source → model → policy → ranking—to surface where programmability really lies (Castells 2010). (Wiley)
• Bridge places into flows. Equip edge institutions (schools, libraries, city hubs) to translate, mirror, and cache civic content so identities can travel through flows (Castells 2015; 2001). (Wiley)
• Treat law as code. Read the DSA and EU AI Act as network switches that distribute attention and constrain risk. Evaluate how compliance grammars alter routability for different actors. (EUR-Lex)

---

Transparency & AI disclosure

Co-authored with an AI assistant (GPT-5 Thinking). Human lead: Dr. Stephan Pflaum (LMU Career Service). Workflow: outline → theory reconstruction → match/partial/mismatch analysis → APA checks → QA. No personal data used. Limits: interpretive synthesis; models can err. Contact: contact@sociology-of-ai.com. Post_id: sai-2025-11-07-castells-essay.

Check log

• Argument structure ✓ • Matches/Partials/Mismatches ✓ • Law-as-code dimension ✓
• APA short style in text ✓ • Assessment target ✓ • Disclosure ✓

---

Literature (APA, publisher-first links)

• Castells, M. (2010). The rise of the network society (2nd ed.). Oxford, UK: Wiley-Blackwell. (Wiley)
• Castells, M. (2009). Communication power. Oxford, UK: Oxford University Press. (global.oup.com)
• Castells, M. (2015). Networks of outrage and hope: Social movements in the Internet age (2nd ed.). Cambridge, UK: Polity. (Wiley)
• Castells, M. (2001). The Internet Galaxy: Reflections on the Internet, Business, and Society. Oxford, UK: Oxford University Press. (OUP Academic)

Regulation (context):

• European Union. (2024). Artificial Intelligence Act (Regulation (EU) 2024/1689). Brussels: EUR-Lex. (EUR-Lex)
• European Union. (2022). Digital Services Act (Regulation (EU) 2022/2065). Brussels: EUR-Lex. (EUR-Lex)