From niche to regime: Rethinking openness in the age of artificial intelligence

Synthesis: Deimann and Lukács (2026) ask why 25 years of openness advocacy have not changed how educational institutions actually work, and whether generative AI will repeat the pattern. The paper is a theoretical essay. No data were collected, the evidence is a set of policy documents and secondary sources assigned to analytic positions, and no coding frame or case study is involved. The device is Geels' multi-level perspective, which separates slow landscape pressures and grand imaginaries from the institutionalized regime in the middle and from protected niches below. Read through it, OER, open science and the open AI debate share one structure: effort accumulates at landscape and niche level, the regime absorbs or ignores it, and interventions aimed at the regime stay rhetorical. The contribution is diagnostic and prescriptive, and the claim that the regime is the neglected layer is illustrated rather than demonstrated.

Key Findings

1. It reframes openness as a contested imaginary shaped by many actors rather than a property a system either has or lacks, explicitly rejecting the binary of commercial against emancipatory openness.
2. It argues that OER and open science work has concentrated on the landscape and the niche while the regime, the middle layer of institutional rules and infrastructures, has been neglected or addressed only rhetorically.
3. Its mapping table places the 2024 Dubai Declaration at landscape level, the Opening up Education initiative and Creative Commons' CC Signals at regime level, and temporary German OER funding projects and openwashing at niche level.
4. It treats openwashing, a term it takes from Vincent (2019), as the paradigm case of regime logic dressed in landscape language, with the company behind ChatGPT as its clearest example.
5. It argues that the prevailing OER model remains organized around authorship and adoption rather than care, governance and collective responsibility, and that platforms such as Pressbooks encode the stable single-author work.
6. It proposes an ethos of openness, built on participation, equity, critical engagement, collective stewardship and digital empowerment, as the connective tissue between levels and as the criterion for judging niche innovations.
7. It proposes open infrastructure (open-source tools, shared data, community-governed platforms, open model development) combined with context-specific policy negotiated inside institutions, against top-down mandates.

Three levels and a missing middle

The framework does the analytical work, so it is worth stating plainly. Landscape is the layer of grand narratives and socio-technical imaginaries, material in the sense that only exogenous shocks, wars, demographic change or ideological shifts move it. Regime is the rule set embedded in engineering practices, institutions and infrastructures, characterized by longevity and by resistance to demands for transformation. Niche is where radical innovation is explored in precarious networks, and every niche innovation is quickly exposed to colonization and domestication by the regime.

The argument about OER is positional. Attention went to the upper and lower layers, to declarations on one side and funded projects on the other, while the regime that decides what institutions actually do was left alone. When it was addressed, the measures were symbolic. Their illustration is the MOOC, announced as a disruptive revolution and then absorbed, falling short of the narrative and of the exaggerated expectations. The same sequence is now visible in AI, where an open source AI definition and licensing projects aim at compatibility with existing practice, not at redistribution of institutional power.

The scope matters for the claim. The authors confine the analysis to the last 25 years, which covers the 2019 OER Recommendation, the 2021 Open Science Recommendation and the Dubai Declaration, and they add the 2022 release of ChatGPT as the event that pushed both communities into the current debate. The implication they draw is that the pattern is repeating, with a technology that moves faster than the institutional response.

Openness as an imaginary rather than a property

The paper's sharpest conceptual move is to treat openness as something actors compete to define. It cites the Open Knowledge Foundation's argument that the term is politically loaded, privileging certain world views and reinforcing power relations where there is no conscious orientation to the public good. It adds the older charge that openness can be instrumentalized by the state as soft power, not only appropriated by commercial interests, and it notes that AI is routinely portrayed as a fixed object with an inherent nature, which makes reflection on its use harder than it should be.

Applied to AI, the definitional problem becomes concrete. Openness could attach to training data, to weights or to code, and the three are not equivalent. The paper treats the 2024 Open Source AI Definition as a genuine step and openwashing as the reason the step was needed. It also sees a limit in the digital public goods revival. The idea has re-entered the discourse, and the Dubai Declaration and CC Signals have taken it up, but the paper reads the move as an illustration of the limits of regime-level intervention rather than an escape from them.

The essay's own political framing is not concealed. It warns against promises that blend utopian social goals with a deregulatory agenda, and its conclusion names a fascist takeover by Big Tech as the harm to prevent. That is an argumentative commitment, stated as such, and it shapes which documents count as evidence of capture.

The regime that OER never dislodged

The institutional analysis is where the paper is most specific. OER are described, following Daly, Schneider and Ahmad, as dynamic and network-based artifacts that are continuously changed and maintained, which makes them incongruent with publishing and textbook paradigms. Practice, however, still runs on authorship and adoption. Governance for maintenance, curation and collective responsibility is absent, and the authors read that absence as a diffusion of responsibility across stakeholders rather than an oversight by any one of them.

The infrastructural version of the argument is that tools shape what is possible. Pressbooks and comparable platforms are built for a stable single-author work, so they reproduce the old regime even when the content is openly licensed. Against that the paper sets Wikipedia and free software as evidence that commons-based peer production can be effective and sustainable, operating without market prices or hierarchical management. At niche level the record is thinner still: OER are rarely embedded in curricula, quality guidelines and funding incentives for use or production are missing, and uptake depends on individual enthusiasm.

The conclusion extends the point beyond OER. Technology is not a sufficient catalyst for transition, a lesson the authors draw from 40 years of research on educational technology, and they observe that students reach for YouTube and Wikipedia while instruction remains largely teacher centered. Their blunt version of the risk is a question: why would anyone need OER when a chatbot can produce ready-made content.

Ethos first, infrastructure second

The recommendations are sequenced deliberately. First, a theoretical foundation for openness, values beyond access, so that the term can serve as a normative basis for critique and as a test for whether a niche innovation advances open values or repackages existing power. Second, infrastructure: open-source tools, shared data, community-governed platforms, open model development and deployment. Third, policy developed inside institutions with their stakeholders instead of handed down, on the argument that landscape declarations set direction while regime negotiation produces change.

What the framework cannot do is tell you when a transition has occurred. The paper applies no indicator to any jurisdiction, compares no country with another, and does not specify what evidence would show that a regime had shifted rather than absorbed a reform. Its examples are described as prototypical, chosen to illustrate the model, and the selection is not defended against alternatives. The result is a well-constructed argument about where to intervene, with the effectiveness of intervention left open.

What this means for practice

Instructors appear in this paper mainly as the people who carry OER when institutions do not, and the implied advice is uncomfortable: individual dedication is precisely the failure mode described. The usable move is to read any open tool or resource against the paper's three questions. Which level is this intervention actually operating at? Does the license reach the layer that matters for this system, the content, the weights or the code? And does it change what the institution does, or only what it says? For a teacher deciding whether to build a course on an openly licensed model, the essay's warning is that openwashing at the model layer is as likely as it was at the content layer.

Limitations

• No data, sample or method in the empirical sense. Documents are assigned to the three levels by the authors' judgment, and no coding scheme, second coder or inter-rater check is reported, so the placement of any single item is an interpretation.
• The level assignments are contestable in ways the paper does not argue. Classifying CC Signals or the Opening up Education initiative as regime rather than niche is asserted, and a different assignment would change the diagnosis.
• The paper's account of regime resistance is general. It names no institution, ministry or platform operator as a case, and offers no mechanism by which the recommended ethos would enter decision-making beyond the claim that stakeholders should co-develop policy.
• The political framing carries claims that the analysis does not test, including the characterization of open AI initiatives as largely failing to provide access to their work and the closing warning about Big Tech, both of which rest on cited commentary rather than on inspection of the systems named.

Citation

Deimann, M., & Lukács, B. (2026). From niche to regime: Rethinking openness in the age of artificial intelligence. Journal of Interactive Media in Education, 2026(1), article 4, 1-12.