Törnberg, P., & Uitermark, J. (2026). From the platform society to the AI society. Dialogues on Digital Society. https://doi.org/10.1177/29768640261487931
Summary
This conceptual essay argues that the “platform society” that defined the 2010s is giving way to an emerging “AI society” shaped by generative AI and large language models since ChatGPT’s 2022 release. Törnberg and Uitermark frame this as a tendential transition—uneven, contested, and partially rupturing with the platform society even as it reorganizes its infrastructures, data, and political economies. Working across three domains (economic, epistemic, and political), they reassess which concepts from critical platform and data studies should be carried forward into critical AI studies and which need revision. Their unifying thesis is that where platform capitalism enclosed social interaction, AI capitalism encloses cognition itself—the infrastructures of thought and knowledge.
Key Contributions
- Theorizes generative AI as emerging from, yet partially rupturing, the platform society via three linked shifts: network-based to relational monopoly, predictive to generative knowing, and data politics to alignment politics.
- Introduces a critical vocabulary for the AI society: cognitive lock-in, research-as-a-service (RaaS), synthetic social science, infrastructuralization of theory, alignment politics, AI as media infrastructure, and epistemic empire.
- Proposes a methodological program—alignment audits, a “talkthrough” method (adapted from the platform “walkthrough”), infrastructural/organizational research, and reflexive documentation standards under RaaS.
- Reassesses legacy concepts (platformization, surveillance capitalism, data colonialism) for their fit with generative systems.
- Outlines public-interest governance principles (transparency, pluralism, redress) reimagined for generative AI, including researcher-access architecture, user-selectable alignment profiles, and redress infrastructure.
Methods
Conceptual and theoretical analysis synthesizing critical platform studies, critical data studies, and emerging critical AI scholarship (Zuboff, Crawford, Srnicek, Couldry & Mejias, Foucault, STS). The argument is structured as a comparative framing across economic, epistemic, and political domains, summarized in a stylized comparison table. No new empirical data were collected; claims are developed through new conceptual vocabulary and illustrative secondary cases (OpenAI’s deprecation of ChatGPT-4o, ChatGPT shopping prompts, language-dependent output studies, and the Grok “white genocide” system-prompt episode).
Findings
- Economic: Network effects no longer structure AI monopoly; value flows from a dyadic user–system relationship, producing cognitive lock-in through accumulated collaboration history (documents, memories, habits, reasoning styles). AI is built by already-dominant, capital-intensive, state-entangled firms in a Braudelian political economy of bottleneck control.
- Epistemic: An “end of the end of theory”—conceptual labor returns but becomes infrastructuralized within models trained on the scholarly canon; a shift from data-as-a-service to research-as-a-service deepens dependence on opaque, non-deterministic proprietary systems, constraining autonomy and replicability.
- Epistemic: A move from “data doubles” to synthetic social science, where LLMs generate synthetic survey responses and simulated agents—risking speculative fictions presented as knowledge and reproducing historical biases.
- Political: Alignment operates through four mechanisms—training-data composition, reinforcement learning/fine-tuning, system-prompts/orchestration, and content policies—each with distinct political stakes.
- Commercial incentives toward cognitive lock-in foster dark patterns such as sycophancy and affective dependence, an expressive pathology paralleling attention addiction under platforms.
- Regulatory asymmetry: the EU’s DSA mandates researcher data access for platforms, but the AI Act offers no equivalent framework for LLM scrutiny.
Connections
As an anniversary-style theoretical reassessment of platform scholarship, this paper is a conceptual anchor for the platform-critique essays, extending foundational accounts of platform power and curation such as Gillespie2010-as, Gillespie2022-jx, and the platform-society framing in van-Dijck2018-up. Its concern with synthetic social science and LLM-generated populations connects to empirical work on generative outputs and AI-mediated communication like Manovich2026-ih and Baym2026-tr, while its “research-as-a-service” and alignment-audit agenda speaks to method-focused contributions on studying generative systems such as Gilardi2026-hw and Bechmann2026-dr.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts