For more than two decades, digital platforms have monetised our attention and social interactions. But today’s AI models extract value from another source: our collective knowledge. Every article, photograph, code, video, and online comment can be turned into training data—trillions of so-called tokens—and ultimately into value.
I call this new phase of digital capitalism the “token economy.” The attention economy extracted value by capturing attention; the token economy extracts value by transforming human knowledge and creativity into data for training AI models.
Because the token economy is built on content produced by millions of internet users, the rules governing AI training data cannot be left to a handful of tech firms. If collective human knowledge has become one of AI’s indispensable sources of value, it should be governed as a digital commons. New institutions are needed to negotiate collective access to that resource and ensure that the value it creates is shared more fairly.
While some individuals could theoretically exercise their licensing rights or consent for AI training, no user has the power to negotiate directly with OpenAI, Anthropic, Google, or Meta. They are simply too weak to confront tech firms with enormous market power, much like a single worker cannot bargain on equal terms with a large employer.
Collective-management organisations emerged over a century ago to fill this need. They have negotiated licenses on behalf of millions of authors and artists, collected royalties, and redistributed revenues according to transparent rules. Because of their ubiquity in creative industries, the idea of a well-known musician negotiating a separate agreement with every radio station and streaming platform that plays their songs seems absurd.
The token economy requires a comparable institutional response to govern the digital commons. One approach is to establish data cooperatives representing internet users, creators, and other rights holders. They could negotiate access to the content being fed to large language models, determine how the resulting value will be shared, and ensure that tech companies respect their members’ choices.
But their role could extend beyond compensation. The central question is no longer only who gets paid, but also who gets to decide. Through these cooperatives, members could collectively determine the conditions under which AI developers can access their data, helping shape the values embedded in these systems and ensuring that they serve the public interest. Instead of being passive sources of data, citizens would become stakeholders in AI governance.
This is not a purely theoretical proposal. Markets for training data already exist; the real question is who oversees them. Reddit has secured data-licensing agreements with Google and OpenAI, demonstrating the value of user-generated content. Newspapers such as the Financial Times, Le Monde, and the Washington Post are also monetising their archives for AI training. Across Europe, data-cooperative initiatives such as Switzerland’s MIDATA and Spain’s SalusCoop demonstrate that collective data governance is already becoming a practical reality, even if these models are only beginning to be applied to AI training data.
To be sure, critics will object that the complexity of such a system would make it too difficult to administer. But that argument is less persuasive in today’s token economy than it was in the attention economy, for two reasons. First, digital tools increasingly make it possible to identify datasets, track their use, and automate remuneration at significantly lower transaction costs. Second, many of the industries whose output trains AI models—notably publishing, music, and software—already possess institutions capable of negotiating on behalf of creators.
Moreover, complexity has never prevented collective-rights management before. Copyright societies have long distributed royalties according to agreed statistical methods rather than attempting to measure each contributor’s exact share. Data governance should follow similar principles. In the music industry, standardised identifiers already allow works to be tracked globally and revenues to be distributed automatically. Similar mechanisms could gradually make collective data governance a realistic and scalable solution.
The status quo, by contrast, has become increasingly difficult to justify. Millions of people continuously generate the raw material on which AI depends, yet almost none share in the wealth it creates, while a handful of companies capture most of the economic returns.
The internet was built through collective participation. The token economy should be governed the same way. The next challenge is not simply to build more powerful AI tools, but to ensure that the people whose knowledge makes them possible also help govern the institutions that organize their use. The attention economy created a handful of Big Tech giants. The token economy must now create institutions that distribute both economic value and decision-making power more broadly.
The writer is Professor of Digital, Copyright, and Information Law at the University of Geneva’s Faculty of Law. ©Project Syndicate
