openinfra.hypotheses.org
Beyond the Paradox of Open: Alek Tarkowski on AI and the Future of Digital Commons
What happens when openness, intended to democratise access to knowledge, instead reinforces concentrations of power? In this interview, **Alek Tarkowski of Open Future** discusses how generative AI is intensifying the _paradox of open_ , turning knowledge commons into invisible resources for commercial AI systems. He explores why new forms of reciprocity, public AI and independent digital infrastructures are essential to making openness sustainable and equitable.
**From the perspective of your own background and work, what do you associate with the “politics of open infrastructures”?**
Politics of open infrastructures are the sum of processes of negotiating what openness means, how it functions, and what are its affordances in a changing digital environment. I would hope for these politics to be mainly the result of intentional governance, of the shaping of open ecosystems by their stewards and stakeholders. And Lessig’s maxim “code is law” holds true today: the strongest and most interesting governance efforts in this space are infrastructural in character.
Efforts to shape openness have in recent years been most strongly expressed not through licensing frameworks or institutional manifestos like the one that shaped the Open Access movement – but by infrastructural choices. Unfortunately, the strongest case in point is a case of open washing: the emergence of open weights model sharing in AI development. Amid a lack of clarity as to the standard for open sourcing AI models, the dominant openly available AI infrastructure is based on constrained sharing of resources, combined with a rhetoric of openness. And efforts to develop public and open AI infrastructure – alternatives to Big AI Tech that are crucial as new AI ecosystems emerge – are constrained by lack of public infrastructure. You cannot have a data space, or a data commons if you have to rely on hyperscaler infrastructure.
Politics of open infrastructures also involve a re-negotiation of openness in the face of concentrations of power. They are a reaction to a crisis of openness that we first observed in the case study of the YFCC100M dataset – when 100 million CC licensed images were used to create a training resource for machine vision systems. This was an impressive case of reuse of openly shared resources, and at the same time a case of extraction of value from the commons. And a case that signalled what was to come, with the training of AI systems on the open web, and the ongoing dynamics where value extracted from the commons feeds concentrations of power related to web and AI platforms, without any reciprocity.
A response to this crisis is a pressing need, and the key political question for advocates of openness and stewards of the commons. Unfortunately, as of today, they seem to lack a capacity to coordinate a response; but also there are tensions, often unacknowledged, that limit coordination. One such fault-line runs between open source and other fields of openness. Open source development is well aligned with digital sovereignty narratives, while broadly understood knowledge commons are left out of scope. This leads to fundamentally different narratives for code and non-code infrastructures. Open code, understood as critical infrastructure, fits well into narratives stressing competition goals or needs of digital sovereignty. Knowledge commons, in turn, are not appreciated as such infrastructure. Changing this requires a policy narrative that pays attention to epistemic independence as also critical for ensuring sovereignty. I believe that it is key, in face of how AI technologies are transforming the public space, to acknowledge this aspect of sovereignty.
It might also be a case of a lack of the collective capacity to imagine a way out of this crisis, into new modes of openness. I am convinced that there is no return to the “traditional” forms of open sharing, as they were developed at the turn of the century – and the need to chart blueprints for these modes is urgent.
**Open Future coined the term “paradox of open” to describe how openness can enable rather than prevent the concentration of power. How has the rise of generative AI changed or intensified this paradox?**
We first used the phrase in an essay titled “Paradox of Open”, which Paul Keller and I wrote in 2021. Back then we were mainly thinking about the impact that platformization of the web had on the commons. The first case to which we applied this analytical lens was the YFCC100M dataset mentioned above – concerned machine learning technologies. We published our case study in autumn of 2022, two months before ChatGPT was launched. Since that time, it indeed became clear that the paradox has intensified.
It is now well understood that open resources played a foundational role in the development of generative AI models. Wikipedia, the flagship of copyleft sharing, is the single most important source of high quality, open data for language models. And the web, enabled by open protocols and built on a norm of public accessibility, has been by now scraped in bulk, and fed into models that require ever larger troves of data. The scenario of openness as an enabler of concentrations of power is playing out, in front of our eyes.
We have been returning to this paradox, and Open Future’s purpose can be understood as that of solving this paradox. Especially since free and open licensing, the core governance mechanism, is increasingly proving useless for governing AI-related content use. Models have been trained with disregard for licensing conditions (usually under a presumption of fair use), copyleft mechanisms most probably break due to the material conditions of model training, and AI interface design treats attribution as unnecessary friction.
And we are seeing a specific shape of threats to open sharing that emerge under the conditions of the extreme concentrations of power, combined with the emergence of machine web traffic and content use: the dark commons. The knowledge commons are becoming an invisible substrate that resources AI systems, which in turn do not reciprocate, or even acknowledge the role of these commons. This is not a case of enclosure, or of ecosystem capture, but rather a capture of value from a functioning ecosystem that for now remains open; and of collective labour that becomes hidden.
The term dark commons draws an analogy with dark (or ghost) kitchens: production sites for food delivery platforms that retain the element of food production, while discarding everything else. These exist in a state of precarity, they are restaurants with sociality and cultural value removed, resembling the horrific factories of the early industrial period. I hope that the analogy with the current crisis of the digital commons is clear.
The most interesting response to the rise of generative AI actually precedes the rapid growth of these technologies. Wikimedia Enterprise was initially launched with search engines in mind. It is an infrastructural solution that attempts, at the same time, to control harmful externalities of machine traffic, and to establish a norm of financial reciprocity. Other experiments in data sharing, such as the Mozilla Data Collective or Harvard’s Institutional Data Initiative, also shift the locus of control from legal rules to infrastructural conditions.
And the risk all these experiments face is precisely that of enclosure, enacted by the stewards of the commons themselves, as a defensive measure. Solving the paradox of open means finding the narrow path between remaining prone to capture and giving in to the urge to close the commons.
**Which ideas or initiatives (also from Open Future!) do you consider most promising for making open infrastructures and digital commons more sustainable and reciprocal?**
First of all, there is a need to frame openness in terms of financial contributions to the commons, and reciprocity. This points to a blind spot in the politics of open: the question of how the production and sharing of resources is sustained. At the turn of the century, open sharing frameworks emerged with a vision that was at best naive: that this sustainability can be taken for granted. Even the debate on commercial and non-commercial sharing, which was so close to these questions, was traditionally framed in relation to legal matters rather than economic considerations.
There is a need to acknowledge the groundbreaking character of projects like the Goteo crowdfunding platform, which already in 2011 investigated the sustainability of the commons, and built infrastructure for financial contributions. The project has been extremely successful in Spain, but gained neither traction nor replication abroad. Similarly, the success of Wikimedia as a platform needs to be understood also with regard to its distributed, civic model of sustainability, based on individual donations to the platform. Finally, new financial frameworks – like WIkimedia’s Enterprise API – are some of the most important innovations that I have been observing.
Following this line of thinking, one of our lines of work – led by Paul Keller – explores new frameworks for financial reciprocity. The starting point for this work is a shift of attention from infringement of individual works – the traditional focus of copyright advocacy – to appropriation of the information commons as a whole. This means that solutions need to go beyond forms of compensation like individual (or even collective) licensing of individual rights, and also concern redistribution towards broadly understood information commons. This includes not just stewards of the knowledge commons, but also public service media, journalistic organizations or cultural heritage institutions. This can be seen as a post-copyright framing, based on a proposal for a levy, triggered by market deployment of AI systems.
The concept of public AI: democratically governed alternatives to dominant AI systems, is another bold idea that we have been exploring. It starts with an understanding that the concentrations of power causing the paradox of open cannot be addressed just by restructuring the commons; particularly as for now successful defences usually lead to enclosure, creating an ideological dilemma for the commons. And also with a realization that the threat of a dark commons requires action at the level of AI systems and interfaces: the commons will be made visible only through alternatives to the existing systems. And these alternatives need to have public AI characteristics: be open, have public interest functions, and remain under democratic control.
The key lesson from our work on solving the paradox of open is that the commons can be meaningful and sustainable only if it rests on an infrastructural foundation. Today, the digital commons often depend on private infrastructure: open code is shared on GitHub, datasets on Hugging Face, and many efforts to develop commons platforms struggle to either sustain independent infrastructure, or are weary of reliance on hyperscalers. Under these conditions, even well-designed governance of shared resources can be insufficient to secure a sustainable commons. Public digital infrastructure (meaning infrastructure that is accessible, standards-based, and stewarded for public benefit) should be a core building block for the commons.
_Alek thanks Paul Keller for his feedback._
Alek Tarkowski is the Director of Strategy at Open Future and a sociologist by training. For two decades he has worked on openness and the digital commons, as an advocate, a movement builder and a researcher. His current work looks at how commons-based approaches can shape the governance of AI and the new digital ecosystem.
Read more: https://paradox.openfuture.eu/
Alek Tarkowski
View all posts
Katja Mayer
View all posts
* * *
The text only may be used under licence Creative Commons Attribution Share Alike 4.0 International. All other elements (illustrations, imported files) are “All rights reserved”, unless otherwise stated.
* * *
OpenEdition suggests that you cite this post as follows:
Alek Tarkowski, Katja Mayer (October 14, 2026). Beyond the Paradox of Open: Alek Tarkowski on AI and the Future of Digital Commons. _Politics of Open Infrastructures_. Retrieved October 21, 2026 from https://openinfra.hypotheses.org/569
* * *
* * * * *
