From Niche to Regime: Rethinking Openness in the Age of Artificial Intelligence

2.1 The multi-level perspective (MLP) framework

In light of the far-reaching consequences of the aforementioned changes currently envisioned for society (and education), it is recommended that a multi-level approach be adopted (Figure 1), with a view to distinguishing logics and rationales among the various actors involved. The MLP framework is predicated on transition theory, a concept that draws upon a range of disciplines, including history and evolutionary economics and it posits that diffusion or transitions occur through interactions among three levels: the niche, the regime, and the landscape.

Figure 1

Beginning at the top of the three levels, the term ‘landscape’ encompasses grand narratives or socio-technical imaginaries (Jasanoff & Kim 2015)¹ and has been chosen quite literally because of its relative material hardness. Only exogenous developments or shocks such as economic crises, demographic changes, wars, ideological change, or major environmental disruption (e.g., the Fukushima nuclear accident in 2011) are assumed to lead to changes at the landscape level. The impact of these exogenous factors creates pressures on the current regime – which is situated underneath the landscape, and can generally be considered outside of possible direct influence. These pressures, in turn, create what are termed ‘windows of opportunity’ for the diffusion of (radical) innovations. It is defined as “(…) rule-set or grammar embedded in a complex of engineering practices (…); all of them embedded in institutions and infrastructures” (Rip & Kemp 1998: S. 340). Regimes are characterized by their profound institutionalization, protracted longevity (due to coordination among actors), and unwavering resilience in thwarting any demand for transformation in social and political practices from the landscape and the niche. The latter then refers to the layer on which radical innovations are being explored. These innovations are often the product of precarious networks, and their purpose is to envision future technological solutions for pressing social problems. Each new innovation is swiftly subjected to the power of the regime, which can lead to effects of colonization and domestication (Bayne & Ross 2024). These effects could be observed in the context of Massive Open Online Courses, once heralded as a disruptive revolution in education (Daniel 2014: S. 20) but which subsequently fell short of this narrative and the exaggerated expectations (Weller 2021).

It is therefore imperative to ensure that radical innovations are not constrained by the prevailing regime, such as market competition. Furthermore, it is important to exercise caution in relation to overly ambitious promises, which artfully blend utopian social goals with a deregulatory or even fascist agenda (Golumbia 2024). Incubators and similar entities can serve as a protective barrier, within which innovations can demonstrate their viability and cultivate a prerequisite for achieving widespread dissemination of the novel technology within society. Conversely, prominent individuals with enormous economic power such as Peter Thiel and strategic intellectual organizations have, over an extended period, been successful in controlling the discourse, leading to significant changes in government politics, particularly in the second Trump administration (Alexandre 2023; Faure 2024).

While a significant proportion of research has focused on major transitions, such as the transition from fossil fuels to renewable energy (Hajer & Pelzer 2018), we contend that the theory makes a substantial contribution to the domain of OE, not only when taking into account the 2019 OER Recommendation and the 2021 Open Science Recommendation (both by UNESCO), but also, especially, recent technological developments regarding AI and Blockchain.

We will now briefly elaborate on the merits of adopting a multi-layered approach to OE. This will then be followed by an examination of the prevailing discourse on AI and OE.

2.2 Adopting the MLP framework for Open Education

For the sake of clarity, the following reflections are situated within the last 25 years and in the context of the evolution of OER and OS. The primary argument put forward is that the focus of OER and OS has been predominantly targeted at the upper and lower layers, while the regime, which constitutes the middle layer, has often been neglected. Even when the regime level is addressed, the measures remain largely rhetorical, without touching the deep-rooted power structures.

To better illustrate how openness-related initiatives unfold across different levels, we assign selected examples to the MLP framework (see Table 1). The examples can be described as prototypical in terms of how they are embedded at the respective level and in terms of their logic of action.

We have selected these examples to illustrate the theoretical statements of the MLP model and to provide a certain analytical frame. By doing so, the MLP allows for a differentiated view of current dynamics surrounding openness, such as in the context of generative AI. While experimental practices are emerging at the niche level, for example through project-based funding lines or civil society reflections (e.g., criticism of openwashing), their structural impact within educational regimes remains limited for the time being. At the regime level, on the other hand, there are attempts to symbolically stabilise established patterns of order – for example, through technocratic reform proposals such as CC Signals or the definition of ‘open source AI’, which are intended to ensure compatibility with existing licensing practices without fundamentally changing institutional power relations. Finally, normative visions of the future and global strategy discourses (e.g., the Dubai Declaration) are formulated at the landscape level, which often has no direct link to operational logic in the education system. Nevertheless, policy documents such as the Dubai Declaration aim to indirectly shape institutional responses or trigger niche innovations.

Overall, it is clear that many past or current measures remain confined to protected experimental niches or involve symbolic reforms at the institutional level, with no actual structural adjustments, thus reinforcing the status quo. This creates a key area of tension: the transformation deficit resulting from the clash between technological disruption caused by AI and institutional inertia requires new responses at the regime level.

Among other devlopments, the notion of digital public goods (DPGs), which has re-emerged in the discourse (Marda, Sun & Surman 2024) and is embraced as a catalyst for OER in the Dubai Declaration (UNESCO 2024) and the CC signals project (Hardinges, Pearson & Ross 2025) illustrates the limits of regime-level interventions.

It is noteworthy that by beginning to (re-)frame digital educational content and OER as DPG, the inherent issue with a lacking foundation becomes apparent. Therefore, we argue that definitory relationships between technologies and values proposed through ‘openness’ are in need of renegotiation.

3.1 Landscape

Long-term, slow-moving developments and major social trends (e.g. climate crisis, globalization, political instability, digital transformation, populism) occur at this level. Many of the recent crises are now being transferred to AI and technical infrastructure is becoming a tool to solve them (Bulathwela et al. 2024). Instead of dealing with them as socio-political issues, there is a discursive shift towards issues of AI technologies, algorithms, and data. This reinterpretation of technological systems as political instruments leads to a normalization of authoritarian logic, without this being openly discussed as a political decision (Brevini 2021). Instead, AI is presented as a seemingly neutral answer to the practical constraints of the time. The legal philosopher Antoinette Rouvroy (2020) speaks of ‘algorithmic governmentality’, a new form of social control based on the algorithmic processing of large amounts of data rather than on traditional political, legal or social norms.

In this context, international organizations such as UNESCO (2024) and Creative Commons (Hardinges, Pearson & Ross 2025) promote AI governance frameworks that emphasize transparency, openness, and ethics. While these efforts often invoke openness as a normative principle, they may also serve to legitimize institutional agendas or geopolitical interests. Thus, openness becomes part of a soft power strategy to shape global technology interests.

In Geels’ (2002) model, landscape changes are often considered external, inert and difficult to influence. However, global political initiatives such as the Dubai Declaration show that there are attempts to shape the landscape. The declaration advocates the expansion of open infrastructures and the maintenance of DPG, including open source software, open AI models and open standards. Concomitant with initiatives such as the United Nations Global Digital Compact, a global, cooperation-based alternative to commercial AI infrastructure (Big Tech) is being advanced. Digital education, as a commons, should not be replaced by AI, but rather complemented by ethically bound, culturally embedded AI – based on accountability, ‘openness’ and participation. Concurrently, we are observing renegotiations that aspire to establish a vision for a digital infrastructure oriented towards the public interest as an alternative to commercial social networks controlled by a central authority (e.g., Meta). The objective is to establish digital spaces, such as those facilitated by public broadcasting services, that prioritize democratic opinion-forming, exchange, and participation. These spaces seek to open up a new narrative aimed at mitigating the destructive dynamics characteristic of market-driven platforms (Dogruel et al. 2025). A reorientation of the role of educational technologies is also being called for, the specifics of which are outlined as ‘rewilding’ (for example, devices with a long service life, open source, repair cafés, DIY culture) and which functions as a principle for the public good (Macgilchrist 2021).

The Dubai Declaration, therefore, represents an endeavor not merely to acquiesce to the shifting of landscape power through the medium of AI, but rather to facilitate a political and normative framing of the technological landscape. In this new context, the objective is to preserve the principles of OE and digital justice and to develop them further by creating robust political, ethical and infrastructural foundations for the ‘AI-infected’ educational landscape.

3.2 Regime

At the regime level, institutional actors such as foundations, ministries, research alliances, and standardization bodies attempt to stabilize and govern AI and openness through definitions, policies, and infrastructural projects. These are not merely technological add-ons, but rather fundamental changes to the way in which these institutions operate. A notable illustration of this transformation is the adoption of algorithmic risk scoring in social assistance, which replaces human judgement with mathematically optimized assessments.

This development is precipitating a profound renegotiation of institutional norms, responsibilities and evaluation standards. Decisions that were once socially, deliberatively and ethically embedded are now regarded as technical optimization problems. This shift has implications for the manner in which decisions are made, as well as the perceived legitimacy of those decisions. The technical solution is regarded as being incorruptible, efficient and objective. However, it is important to note that its efficacy may be dependent on structurally biased training data or the perpetuation of discriminatory patterns.

The field of AI has prompted a rethinking of the very standards to which regimes are held, as evidenced by the sudden alignment of social justice with algorithmic fairness (Bozkurt et al. 2024). This has given rise to new lines of conflict and interpretive struggles about the relationship between technology, power and society.

In terms of the MLP framework (Geels 2002), it is imperative that AI is addressed at the regime level, given its pervasive integration into social structures and its substantial impact on their evolution. The integration of AI, as argued by Morozov (2024) in his essay “The AI We Deserve”, must be executed in a manner that is consistent with democratic governance structures, thereby ensuring that technological advancements are aligned with social values and do not engender societal destabilization. Instead, these advancements should be transformative, contributing to the enhancement of existing regimes. It is therefore encouraging to see the growing discourse around the technical aspects of ‘openness’ in AI, namely what aspects of this emerging technology are accessible, distributable and re-usable by society at large (Bommasani et al. 2024; Tarkowski 2025; Mahendra 2024; Nest 2024; Liesenfeld & Dingemanse 2024; Maffulli 2023).

One of such an enhancement concerns renegotiating the integration of OER, which are, according to Daly, Schneider and Ahmad (2024), dynamic, network-based knowledge artefacts that are continuously changed and maintained. It is evident that the practices associated with OER precipitate the introduction of novel institutional, technological and cultural practices and it can thus be deduced that these are no longer congruent with the conventional paradigms of publishing and textbook use. Nevertheless, the prevailing model of employing OER remains contingent on authorship and adoption, rather than being oriented towards care, governance and collective responsibility. The absence of governance is indicative of the fact that regime infrastructures (e.g. platforms such as Pressbooks) are tailored to the model of the stable, single-author work – they reflect the structure of the old regime. The implementation of alternative, decentralized infrastructures for collaboration, democratic control and shared stewardship is proposed as a means of disrupting established routines and initiating pathways that will lead to long-term transformation within the field as a whole. Digital commons, which are grounded in open licenses, are significant in this regard as a means of undermining and reconstituting legal regimes (Dulong De Rosnay & Stalder 2020). The efficacy and sustainability of commons-based peer production is demonstrated by projects such as Wikipedia and free software. Operating independently of market prices or hierarchical management, these initiatives rely on collective, network-based forms of ‘commoning’.

The ongoing absence of a sustainable foundation for OER within education systems, attributable to a diffusion of responsibility, is indicative of a need for the new social contract to call for collective responsibility on the part of all education stakeholders to advance educational equity.

3.3 Niche

Rather less hype has been put forward regarding OER as a disruptive force for education. Niche innovations have focused more on the affordances of OER, such as in the forms of collaborative knowledge production and shared learning (Prince Machado, Tenorio Sepúlveda & Ramirez Montoya 2016). Nevertheless, this form of participatory content production is at odds with conventional, teacher-centered educational models. It demands digital competencies, dynamic engagement and an environment conducive to open learning, all of which are not yet firmly embedded within many institutional frameworks. The pedagogical ‘openness’ and role change that are prerequisites for this approach are often only present in small, experimental settings. Despite the existence of national and university-owned repositories, OER are seldom systematically incorporated into curricula. Institutional standards, quality guidelines and funding incentives for the use or production of such resources are lacking. The utilization of these resources is contingent upon the dedication of individual educators, as opposed to being informed by overarching strategic frameworks.

Similarly, the AI landscape saw the emergence of niche challenges for the dominant regime of proprietary models almost immediately. Discussion saw the growth of ostensibly ‘open source’ large language models, community platforms for sharing models, as well as research into decentralized training methods. Unfortunately, despite touting an ‘openness’ approach, arguably very few or no initiatives truly provide access to their thinking and work (Tarkowski 2023). However, much like the institutional hurdles that OER faced, these AI niches confront similar challenges related to computational resources, data access, questions regarding true ‘openness’ vs. ‘openwashing’ and pathways to sustainable development in varied educational and scientific contexts. These experiences interestingly echo the marginalization often experienced by OER initiatives as the key drivers and narrative leaders remain dedicated enthusiasts or well-funded research labs and accelerator projects.

While these examples demonstrate alternative imaginaries of openness, they often remain confined to pilot status. Without sustained political support or structural embedding, their potential to transform the regime remains limited. The lack of upward linkages illustrates the fragility of niche-to-regime transitions.

In the following chapter we will formulate recommendations on how the issues mentioned can be picked up productively and incorporated into future ecosystem developments.

4.1 Theoretical foundation for ‘openness’, and as guiding principle

As previously mentioned, with reference to the phenomenon of ‘openwashing’ in AI and the historical struggles of OER to gain deep traction, the term ‘openness’ itself is still often contested or diluted, e.g. in the case of OER, going from simple access to educational materials to a wider-ranging concept of ‘open educational practices’ (Ehlers 2011). Without a clearly defined and critically (re-)examined understanding of what ‘openness’ entails (either context independent or dependent), continued efforts risk becoming fragmented or possibly serving agendas contrary to core values like equity and democratization (Peters & Roberts 2012).

Simply advocating for ‘open’ AI or ‘open’ content is insufficient when the term itself is malleable and suffers from inherent paradoxes (e.g. Keller & Tarkowski 2021; Heurich & Lukács 2023), which make clear guidance challenging. Drawing inspiration from critical pedagogy and philosophies of open education (Deimann & Farrow 2013), we posit that a robust theoretical framing of ‘openness’ – emphasizing values beyond mere access, values such as participation, equity, critical engagement, collective stewardship and (digital) empowerment – can act as essential connective tissue within and between the levels of the MLP framework. It can provide the normative basis using persuasive narratives to critique existing regime practices, which have frequently fallen short of sustaining OER. Furthermore, it is capable of providing criteria for the evaluation of niche innovations, such as novel AI tools and OER infrastructure. These innovations can be assessed for their ability to genuinely advance open values or whether they merely repackage existing power structures. Crucially, a solid theoretical grounding can provide the ethical compass demanded by the Dubai Declaration to ensure that technology in education serves democratizing goals rather than reinforcing or introducing new forms of control through intransparent models, algorithms and governance structures.

4.2 Connecting the levels through policy and infrastructure

Theoretical clarity must be coupled with tangible mechanisms for change. The rise of AI makes this even more transparent: reliance on closed ecosystems mirrors past challenges. It is therefore key to conceptualize, build and maintain truly open infrastructure, consisting of open-source tools, shared data, community-governed platforms and open AI model development and deployment. Infrastructure (when properly set up) can act as a fertile ground for niche innovations and can provide viable alternatives that can exert pressure on, and offer integration and transition pathways into an existing regime, in this case specifically aligning with calls for DPG and ‘rewilding’ technologies. As argued by Daly, Schneider and Ahmad (2024), “governance cannot be regarded as a merely technical problem any more than we can expect technology to dispense with the need for it”, requiring more OER-sensitive governance models to support, for instance, collective processes for maintenance.

Complementing this is the need for contextualized policy – as landscape-level declarations by UNESCO can be used for setting a general direction of discourse, effective change often requires negotiation within the regime structures. In the example of OER, OS and OE we argue against solely top-down mandates, advocating instead for policies co-developed within educational and research institutions with all its relevant stakeholders. The policies, which would be grounded in the aforementioned ‘Ethos of Openness’ would then be tailored to local contexts, empowering each community to shape their governance and processes while facilitating the integration of niche practices (new AI tools and OER platforms) into established workflows. This can foster a sense of collective responsibility, as envisioned in the call for a new social contract for education (Toukan & Tawil 2024). The combination of truly open infrastructure and context-aware and participatory policy offers a pragmatic pathway for (re-)connecting the MLP levels and helping steer future socio-technical transitions toward genuinely open educational futures.