Commons solve innovation---viewing AI as a public resource to be shared, rather than enclosed via digital commons is key.

In most simple terms, the commons are the natural and cultural resources that are accessible to all members of society. What is typical about them is that they are held in common, instead of being owned privately (Bollier 2014). Public debate about the commons has become more mainstream due to environmental degradation and has been popularised—amongst others—by the frst female winner (2009) of the Nobel Memorial Prize in Economics, Elinor Ostrom. Her work includes Governing the Commons (Ostrom 1990), in which she refutes the Tragedy of the Commons thesis (Hardin 1968). She has inspired thinking about the design and organisation of cooperative alternatives beyond markets and states.

Ostrom, together with her colleague Hess, has also worked on extending the debate about commons to knowledge. Hess and Ostrom (2007) approached knowledge as a complex ecosystem that operates as a common, similar to what Benkler (2006) theorised as commons-based peer production. In a similar vein, others have been working on the concept of digital commons, which refers to the communal ownership and distribution of informational resources and technology (Birkinbine 2018). Taking the ideas of knowledge and digital commons together opens up opportunities to inquire about alternative structures for AI ownership and governance.

We are confronted with intense competition and concentration in AI capitalism, a situation similar to what has been labelled the enclosure of the commons. According to Bollier (2014), the latter refers to a situation in which corporate interests appropriate our shared wealth and turn it into expensive private commodities. This is happening also in the digital sphere, whereby platforms control access to data and increasingly enclose the digital world within their private sphere. Resisting this—by pushing for alternatives—can be done by stressing the importance of data and AI as public goods, produced by society and its members (Taylor 2016; Viljoen 2021). The important task then is to explore how the commons can be reclaimed.

While thinking about the commons has its roots in radical political economy, there is a disagreement about what the end goal of its project should be. Some position the commons as an emergent value system that has the potential to transform or even replace capitalism (Broumas 2017), while others perceive the value of the commons in how it can respond to the excesses and exploitative tendencies of capitalism (De Angelis 2017). As such, the commons are not per se a replacement of capitalism but rather something that can co-exist and couple with capital circuits through the commodity firm.

4.2 Data commons

How can we think about the commons in the context of AI capitalism? First of all, we need to conceptualise the data commons. Bria (2018) defnes data commons as a shared resource that enables citizens to contribute, access and use data as a common good, without or with limited intellectual property restrictions. Instead of considering data as a commodity or capital (Sadowski 2019), it can be thought of as a collective resource (Viljoen 2021). As such, it can empower citizens and help them solve shared—common—problems.

The bigger picture of negotiation and agreements around data commons is part of calls for a New Deal on Data (Bria 2018). A report of the Decode project2 explains what such a deal on data could entail (Bass et al. 2018): First, there is a need to push for more transparency, accountability and trust in data projects; Second, individuals should be given more control and people should be empowered to decide how their data is collected and used; and, Last, it should be an important ambition to unlock more value of data as a common good while protecting people’s privacy and encouraging fair terms of use.

Of course, there are questions how to practically organise this. A lot of inspiring work on the data commons proposes solutions in terms of data infrastructure and data trusts (Coyle 2020). A new data infrastructure should help dealing with institutional and regulatory aspects of how data can be shared, what standards and policies should be set up and which organisations and communities should be involved in contributing to and maintaining this data infrastructure. One approach for an innovative data infrastructure has been developed and trialled in several countries: data trusts. Data trusts can exist in many forms and models but the general principle is that they sit between an individual generating data and a company or institution wanting to use that data (Delacroix and Lawrence 2019). In this system, control over data is transferred to a third party, which can use the data for pre-defned purposes. Data trusts can use data from diferent sources and allow to steward data use for all. Important in its governance is data solidarity, meaning that corporate and public data shareholders share the benefts and risks of data access and production (Bunz and Vrikki 2022). Coming up with a system for sharing and giving access to data does not only beneft society; it is also necessary for AI innovation (Hall and Pesenti 2017).

4.3 Compute capacity for the commons

Compute capacity is the second element of a commons approach, as an alternative to the power concentration of AI capitalism. Some even position computing infrastructure as part of the data commons itself (Grossman et al. 2016). I discussed already how crucial computing power is for the development of AI. Only Big Tech (and some elite universities) have the resources to upgrade their infrastructure—contributing to an AI compute divide (Ahmed and Wahed 2020)—while leading AI companies collect rent from and keep control over what is happening on their compute infrastructure (Srnicek 2019). As an alternative, investments in common/public compute capacity could help society becoming less dependent on the private infrastructure of Big Tech.

While the corporate sector often claims that public investment stifes innovation, (Mazzucato 2013) debunks this myth and actually argues that the radical technologies behind, for example, the iPhone (e.g., GPS, touch screen display and Siri) were all backed by government funding. Another example is Google’s search algorithm, which was publicly funded through the National Science Foundation (NSF).

The first supercomputers were used by universities (in the US and the UK) and governments should consider pooling (more) resources to invest in (national or international) compute capacity that will drive the future of AI. Common investment in AI compute capacity will also help to democratise AI (Riedl 2020), meaning that more people and organisations can be involved in developing AI systems. This is particularly relevant for quantum computing, which is considered crucial for revolutionary breakthroughs in the future of AI—the so-called quantum AI (Taylor 2020). Public/common investment in computing infrastructure could also mean a de-commodifcation of compute capacity and create a new public service that can be made available to society, accessible to different organisations, companies and interest groups.

4.4 A commons approach to AI human capital

While not often considered as part of the data commons, an argument can be made about common investment in AI human capital too. Having an upgraded computer infrastructure is one thing, AI human capital—the AI talent and human resources that are necessary to develop AI innovations—is as important.

Given the high level of specialisation, success in research on machine/deep learning is dependent on people who have accumulated large expertise through formal training (e.g., PhD) or years of applied work (Ahmed and Hamed 2020). As a result, there is a growing gap between the increasing demand for AI expertise and the limited supply, resulting in a talent scarcity (Metz 2017).

A commons approach to AI human capital would, for example, include to provide more funding for public IT services and universities allowing them, respectively, to reduce outsourcing and facilitate more research labs to keep their faculty members instead of being recruited by larger, corporate, organisations with deep pockets.

Investment in public infrastructure and resources can support commons-based economies and models of organisation which allow to depart from an incentive structure focused on value creation rather than value extraction (Kostakis and Bauwens 2014). However, this depends on new regimes in terms of ownership, control and governance.

First, a central aspect of envisioning an alternative political economy of AI is rethinking ownership. Regulation is often proposed as a strategy to limit the market/monopoly power of Big Tech (Posner and Weyl 2018). Competition and antitrust law, for example, could be used to break up the AI/tech giants. However, such a strategy might be counterproductive, as the power of, for example, social media platforms is that they connect everyone in society. Common ownership might be an alternative approach that could be more productive (Kostakis and Bauwens 2014). There is a solid case for placing the technologies producing AI in public and collective ownership. It would mean that communities have more control over how AI is produced and how the public can benefit from its services. The end goal is to have a digital infrastructure that is available to and provides advantages for a broad range of stakeholders in society, not just the AI behemoths.

Second, related to ownership is the aspect of promoting common governance. The goal here is the democratisation of AI and this requires the decentralisation of power, back in the hands of the public (Posner and Weyl 2018; Riedl 2020). If we consider AI as a GPT, which will alter the structures of society, we need to make sure there is democratic oversight and control. After all, we have installed regulators that have the power to protect the interests of citizens in other sectors, such as postal services, electricity, broadcasting and telecommunication. The services provided by AI are so crucial in everyday life, making it necessary that society has a greater say about it.

Inspiration for alternative structures in terms of ownership, control and governance can be found in the platform cooperativism model (Scholz 2017), which allows involvement from multiple stakeholders in the ownership, development and management of platforms.

Finally, we need to come up with a new vocabulary when thinking about AI systems and how they deliver benefits to society. Instead of corporate discourses portraying AI as Tech for Good, boosting innovation and entrepreneurship, it makes sense to perceive AI infrastructures as a computational utility, subject to democratic control (Mosco 2017). Dyer-Witheford and colleagues (2019) elaborate on this and push for considering AI as a communal utility. This means that communities and workers should be involved in determining what sort of work should or should not be automated, and thus call for a genuine determination by the general intellect in the design of AI. In this general intellect, collective cooperation and knowledge become a source of value (Terranova 2000). The proposed principles of common ownership and governance should be central in developing AI as a communal utility.

5 Concluding remarks

This article analyses AI capitalism. I discuss the contradictions between visions of AI as a General Purpose Technology (GPT) generating benefts for society at large and the reality of AI capitalism, characterised by commodifcation, extraction and a power concentration. These aspects are behind the unstoppable expansion of tech platforms and monopolisation in the feld of AI. This leads to a winnertake-all scenario, in which AI giants follow a strategy of enclosure: controlling access to data, talent and compute capacity and enclosing more of the world within their private sphere (Couldry and Mejias 2019). As a result, few players have the power to make decisions about how AI is developed and used and what its impact on society will be. It prevents the culture of AI (Elliott 2019) being accessible for all (Verdegem 2021).

Analysing the political economy of AI makes clear we need to rethink AI ownership and governance, and how data but also human resources and computing infrastructure can be shared and made accessible to a broader range of stakeholders in society. To challenge the power concentration of Big Tech, we need to come up with alternatives.

In this article, I propose the framework of the commons for helping to imagine alternatives in the context of AI capitalism. The commons is an approach that allows us to rethink collective ownership and governance regimes and empower more groups in society to have control over and a say in the development of AI and how the beneftis of AI can be shared with society at large. Instead of the GPT AI resulting in Big Tech as winners and most of us losers, a commons approach allows more groups in society to be among the winners of AI innovation.

The movement towards data and AI commons should be combined with more general initiatives to push back against the commodifcation of and extraction in everyday life. Establishing a more equal and inclusive society depends on the democratisation of tech and the involvement of diferent types of stakeholders (Posner and Weyl 2018). Citizens, civil society and other organisations should have a more prominent role in the development and management of AI, which could take form along the lines of the multi-stakeholder cooperative model (Standing 2019). The latter refers to a model of organisation (co-ops) that allow for governance by representatives of diferent stakeholder groups within a certain organisation.