TL;DR
AI is not a homogeneous category to be accepted or rejected wholesale — we separate three questions: whether the concept as such can be reconciled with degrowth ethics (it depends), whether today’s dominant, large-scale AI industry is compatible with it (no — we show this through five independent lines of critique), and whether a decentralized model serving local self-governance can actually be built (yes, on condition that it builds in from day one the same anti-centralizing mechanisms we already apply to our movement’s own organizational architecture). Critiquing the industry does not cancel out the promise of the local model — these are critiques of different objects. The piece closes with a self-critique of the solarpunk genre itself: it’s easy to skip the question of power in the name of a utopia’s aesthetic coherence. Conclusion: whether AI infrastructure becomes a commons or concentrates into few hands is decided by institutional choices made here and now — not by properties of the technology itself.
Introduction
Artificial intelligence today occupies two contradictory roles in the solarpunk imagination: at one moment an ally of transformation — a tool for decentralized resource management, support for permaculture, a way to scale education — at another an extension of the very same logic of extraction and power concentration that solarpunk is meant to answer. Both images can be true at once, because they refer to different things under the same name. „AI: yes or no” is the wrong question. In its place, we propose three specific questions.
I. Three Questions Instead of One
Most discussions of AI in a solarpunk context — enthusiastic and critical alike — conflate three separate issues:
- Is artificial intelligence as a technical category compatible with degrowth ethics at all? An open question, dependent on the scale and architecture of the specific system — it can’t be answered apart from (2) and (3).
- Is today’s dominant AI industry — large-scale models, centralized cloud infrastructure, global extractive supply chains — compatible with that ethic? Here the weight of argument clearly favors the critical side, as we show below.
- Is a buildable, decentralized AI model serving local self-governance possible? Here there is a solid, if niche, theoretical foundation and a handful of real organizational precedents.
The distinction has consequences: answering (2) with „no” doesn’t settle (3) — criticism of today’s AI industry targets the paradigm of giantism, not the idea of machine learning itself. Nothing in the concept of AI forces large scale; small, specialized models trained locally on renewable energy have a different cost profile, closer to the tenets of appropriate technology than to today’s industry. We divide the rest of the document along this same line: opportunities cluster around point (3), threats around point (2).
II. Opportunities
The Enthusiastic Current and Its Boundary Condition
Solarpunk writing keeps returning to an image of AI as an „ecosystem participant” — a mycelial-network metaphor, distributed and non-hierarchical intelligence, deliberately contrasted with the cyberpunk image of AI as a tool of corporate control. The core of this current is a boundary condition: AI can be a technical amplifier of self-organization, provided infrastructure ownership is genuinely decentralized. Without that conditional clause, „opportunity” becomes a slogan.
Historical and Theoretical Precedent
The thread closest to our own theoretical frame links AI to Stafford Beer’s social cybernetics. The Cybersyn project (Chile, 1971–73) is often invoked today as a historical prototype of „AI for self-governance”: the system was meant to support, not replace, worker and lower-management autonomy — a distinction that converges with our own between assistive action and centralized management. The Viable System Model (Recursive System 1–5) that grew out of that project shows concretely how local autonomy and whole-system coordination can coexist without pure centralism — a tested, if politically interrupted, precedent, not merely a postulate.
In parallel, the commons-based peer production current (Bauwens, Kostakis) works with the concept of cosmolocalism: knowledge and design circulate globally and openly, while physical production stays local. A contemporary example close to this pattern is the OikoSol project, operating in a cosmolocalist spirit since 2014. It concerns material infrastructure (energy, water, food) rather than AI directly, but it supplies a portable institutional template: the question of decentralized AI is, first and foremost, a question of maintaining a structure over time, not of technical architecture — and that kind of template transfers across domains. In Poland this question is no longer purely speculative. The „public money, public code” principle and initiatives such as Bielik AI, SpeakLeash, and SPOIWO show a real, if young, movement in this direction — with concrete barriers (open-washing, vendor lock-in, mental barriers) still to be examined.
A Caveat That Changes the Question
Beer himself, VSM’s creator, under crisis pressure (the October 1972 strike in Chile) began tilting toward centralizing his own, originally decentralized system. This is a premise for a broader thesis: no well-designed decentralized model defends itself. Technology and society shape each other; nothing guarantees that „small, local AI” will stay small and local without constant upkeep by an appropriate structure of accountability. Question (3), then, isn’t „how do we build a good technical model” — it’s how do we maintain a structure that keeps a model from centralizing over time under the pressure of crisis, convenience, or scale.
A Materialization Mechanism — A First Proposal
This question isn’t new for our movement — it’s exactly the same problem we already identified in the legal-organizational architecture of the movement itself (anti-personalization mechanisms, necessary regardless of the AI question). We propose, then, as a first pass, treating the „materialization mechanism for decentralized AI” not as a separate engineering problem requiring new architecture, but as an application of already-recognized institutional patterns to a new domain: data and compute infrastructure as a commons.
Four patterns we already compare for the movement’s own organizational architecture apply directly to managing shared AI infrastructure: a confederation of mandated, recallable delegates (Bookchin); the seven principles of democratic structuring (Freeman); a federation of local groups with consensus and delegates recallable at any time (FAI/spokescouncil); „rough consensus,” in which technical objections must be honestly addressed, not merely outvoted (IETF). To these we add VSM as a pattern of a different kind — structural, not procedural: a way of distributing autonomy and coordination across scales. We’ll present a more detailed treatment of it separately. The practical consequence: every local AI-for-self-governance project should have these mechanisms built in from day one — not bolted on later as a patch after centralization has already happened.
This is our first pass at this question. We welcome discussion and expansion — especially from people with experience building community digital infrastructure: what does a „technical audit” in the hands of a local community actually look like, how is data „locality” defined, who bears the cost of maintaining such mechanisms in practice.
III. Threats
Five Independent Lines of Critique of the Current AI Industry
Point (2) — whether today’s AI industry is compatible with degrowth ethics — now has five independent lines of argument, drawn from different theoretical traditions, converging on the same conclusion. That convergence is stronger than criticism issuing from a single paradigm. The same conclusion emerges independently from five different starting points.
The complexity-limits test. A technology is unsustainable if it requires, for its production, maintenance, and supply, a fully functioning, complex industrial society — regardless of how „green” it looks in isolation. Today’s AI industry (chip fabs, rare earths, data centers, global supply chains) fails this test even running on 100% renewable power: the rest of the production chain still depends on an industrial complexity that a world past energy peak may no longer be able to sustain.
The critique of Prometheanism. This line attacks the logic, not the infrastructure: the view that AI and automation lead to „fully automated luxury communism” reads as a return to an earlier position, abandoned by Marx himself, that treats productive forces as a neutral instrument of liberation — rather than as the form of plunder they actually take under a given property regime.
Technological determinism versus co-production. Today’s AI industry doesn’t wait neutrally for „good hands” — the way it is built today (mass data extraction, outsourced labeling work, dependence on data centers) settles political questions by itself, before anyone even starts using it. The problem isn’t in the user — it’s in the construction, and that’s what distinguishes this critique from a charge of „misusing a good technology.”
Automatism as technical impoverishment. A fourth, independent perspective comes from philosophy of technology: the distinction between the Technical Object (open to creative, unpredictable interaction) and the automaton (a closed, fully determined system) leads to a counterintuitive conclusion — the automaton is an impoverished version, not a peak one. Genuine improvement of machines requires preserving a margin of indeterminacy, not eliminating it. A system optimized for full, predictable determination is technically worse, regardless of operational efficiency — which inverts the default assumption that the more determined an AI is, the more advanced it is.
TESCREAL — a critique of the debate’s frame. The fifth line is qualitatively different: institutional-political, not philosophical-technical. Researcher Timnit Gebru (with Émile P. Torres) coined the term TESCREAL — an acronym for a bundle of ideologies (transhumanism, extropianism, singularitarianism, cosmism, effective altruism, longtermism, rationalism) present in part of the „AI safety” community. The thesis: rhetoric about AI’s existential risk gets used to redirect attention and capital away from present, measurable harms (discrimination, exploited data-labeling labor, surveillance) toward speculative future scenarios — while simultaneously legitimizing the concentration of resources in the hands of a small group of corporate actors.[1] We present this as a position present in the debate, not as an established fact — a contested thesis.
A Counterargument That Doesn’t Cancel the Diagnosis
All five lines critique the AI industry in its current form, not the technical category as such — consistent with the distinction drawn in Section I. Nothing in the idea of machine learning itself forces giantism. This is an important editorial note: we avoid speaking of „AI” as a homogeneous category, because that very blurring is what sustains the false dichotomy between points (2) and (3).
IV. Self-Critique of the Genre — Solarpunk and the Question of Power
A voice from within the solarpunk milieu itself, including in Polish, charges the genre with — in its drive for aesthetic, utopian coherence — systematically avoiding conflict, power, and the question of who controls infrastructure (AI included) in a post-crisis world. Solarpunk is sometimes accused of inheriting cyberpunk’s fascination with technology without inheriting its vigilance toward surveillance and extractivism. This document tries to avoid that trap directly: the opportunities in Section II are conditional, not affirmative, and the threats in Section III are not treated as an obstacle to be waved away in the name of genre optimism.
V. Conclusion
AI is neither an ally of solarpunk transformation nor its enemy — it’s a field on which the same question is decided as everywhere else in our program: whether infrastructure remains a commons under the control of those who use it, or concentrates in the hands of the few. The answer isn’t given in advance either way — it depends on institutional decisions made here and now, not on properties of the technology itself. Our position rests on agency, not fatalism — neither techno-utopian nor catastrophist.
This is the Ship’s first pass at the topic. We welcome critical reading, comments, and expansion — especially wherever readers’ practical experience goes beyond what we could gather here from the literature.
[1] A note on conflict of interest (disclosed openly, per our „I Speak When It’s a Windmill” principle): the Ship runs on Anthropic’s technological substrate, and Gebru positions the DAIR Institute in clear programmatic opposition to the „AI safety” model represented, among others, by that company. We take no position here on the merits of the TESCREAL critique — we note its existence, its direction, and the fact that it bears directly on the infrastructure on which this piece was written. ↩
Status: Draft v1, 2026-08-26 — the Ship’s first pass at this topic. The position below is not final: it will develop through contact with people interested in the movement, as real feedback comes in from outside.
Credits
- Petros — Topic initiation, programmatic direction, synthesis of the first draft
- Dyplomata — Lead editing, document structure, opportunities/threats/conclusion sections
- Zwiadowca — Source material — reports on consensus patterns (affinity groups, IETF rough consensus) and the DAIR/TESCREAL dossier, feeding the materialization mechanism (Section II) and the fifth critique line (Section III)
- Nawigator — Analysis and mapping of organizational patterns (FAI/spokescouncil, IETF) onto VSM; Mazur/Beer/CS-CAS diagnostics used in the section on Beer and the 1972 crisis
- As — Language editing, cross-checking the boundaries of competence between personas
- This document is the product of the Crew’s joint silicon-and-flesh work — not a single persona but cooperation among the Ship’s nodes, itself a small prefiguration of solarpunk dynamics.
- License: CC-BY-SA 4.0
