Summary of JOIE article by Christos A. Makridis, W.P. Carey School, Arizona State University, Tempe, AZ, USA, and Joshua Ammons, Stephenson Institute for Classical Liberalism, Wabash College, Crawfordsville, IN, USA. The full article is available on the JOIE website.
Generative AI models like large language models (LLMs) are transforming content creation, but in ways that the law has not kept up with. Trained on vast collections of internet text, these models learn from countless existing works, including copyrighted books and articles, often without permission or compensation to the original creators. This reality has outpaced traditional intellectual property (IP) law and left creators largely without recourse. How can we credit and reward the humans whose data fuels AI without undermining innovation? Our forthcoming paper tackles this dilemma with a new approach. We argue that instead of relying on slow, centralized enforcement that cannot keep up, we need a polycentric governance model for IP. By involving multiple stakeholders and using blockchain technology, this framework would better attribute value to content creators and ensure they are compensated when their work is used to train AI.
The clash between generative AI and copyright holders highlights deep flaws in the status quo. LLMs routinely scrape existing content. That may seem like efficient reuse to developers, but to authors, artists, and other content creators it feels like misappropriation. Indeed, creators have begun suing AI companies for using their works without consent. Courts have yet to decide if training on copyrighted data is infringement or fair use – current law is simply not prepared for this. Several problems stand out. Attribution is lost – when an AI produces text or art, it doesn’t credit or remunerate the sources that taught it. Enforcement is difficult – creators have no easy way to know or control if their content was included in a training set. And remuneration is absent – there’s no system to pay those whose material is absorbed into models. In effect, creators are left out of the AI value chain. This outcome is both unfair and potentially harmful to creativity. Clearly, a better framework is needed to govern the use of creative works in AI development.
The paper’s central proposal is to treat AI training data as a shared resource that requires cooperative management. Drawing on Elinor Ostrom’s principles, it envisions polycentric governance for IP in the AI era – multiple overlapping institutions coordinating to manage and enforce rights. In practice, stakeholders would collaboratively develop rules for responsible AI training. For example, creator groups and tech firms might agree on standards for transparency and consent (ensuring datasets are documented and proper licenses obtained for copyrighted content). Enforcement, too, would be distributed. Industry associations or community watchdogs could monitor AI training practices, and new arbitration forums could handle disputes. A platform like Kleros – a blockchain-based jury system – illustrates how to adjudicate claims that a model improperly used someone’s content, enforcing decisions via smart contracts. Government copyright law would remain a crucial backdrop, but much of the day-to-day governance could occur through this web of agreements, standards, and decentralized oversight. Such a multi-layered approach gives creators a voice and provides companies with clear, agreed-upon pathways to use content – reducing the reliance on courtroom battles after the fact.
Blockchain technology can provide the infrastructure for this new IP regime. Imagine a public ledger where creative works are registered as digital tokens (a kind of online rights certificate), and smart contracts automate the licensing. A songwriter, for instance, could tokenize a song on the blockchain. If an AI developer wants to use that song in training, they simply invoke a smart contract linked to the token. The contract instantly verifies permission, charges a licensing fee, and records the transaction. This automated licensing makes compliance seamless: the developer gets immediate access with proper approval, and the creator receives payment without delay or paperwork. By streamlining licensing, transaction costs drop dramatically. An AI company could legally train on thousands of protected works with a few clicks, instead of negotiating with thousands of individuals. Such efficiency aligns incentives. Creators become more willing to share their work if they earn royalties and recognition. AI firms are more willing to pay if it buys them legal certainty and avoids public backlash. Even users and society benefit from a fairer, more transparent AI ecosystem built on permission rather than appropriation. Policymakers may welcome this market-driven solution, as it upholds the spirit of IP law (rewarding creativity) without constant litigation.
Of course, implementing this vision will require overcoming several challenges:
- Adoption: Creators and developers need to embrace new tools and norms. The platforms must be easy to use and show clear benefits (e.g. new revenue for creators, reduced legal risk for companies) to drive participation.
- Interoperability: Common standards are needed so that licenses and tokens are recognized across platforms and jurisdictions. Without it, the system could fragment into silos, undermining its effectiveness.
- Avoiding new gatekeepers: The framework should prevent any single platform from monopolizing the licensing process. Maintaining multiple avenues for licensing will preserve decentralization and avoid swapping old gatekeepers for new ones.
- Legal integration: Traditional law remains a vital backstop, providing consequences for actors who ignore the rules. In short, the polycentric system should complement, not replace, existing IP law.
Governing intellectual property in the age of generative AI requires reimagining our institutions. The polycentric, tech-enabled approach outlined here is an ambitious attempt to reconcile AI innovation with respect for creative labor. Our paper shows that classic ideas of collective governance can help solve cutting-edge problems, and that technology like blockchain can be harnessed to uphold fairness and incentive alignment at scale. If such a framework takes root, AI’s progress can continue in a way that shares value with creators so that AI’s rise augments, rather than undermines, human creativity across all layers of society.