By Leander Stähler*
Generative “artificial intelligence” (AI) tools have challenged the flow of digital content. As a crucial regime for practically all creative processes, various concerns thus emerge in copyright law, e.g. regarding the inputs used to train AI tools, the level of similarity between outputs and trained inputs, as well as the core issue of whether AI should fall within the scope of copyright law in the first place. These are all crucial legal questions accompanying the development of tools falling under the broader umbrella of AI, each interacting with forms of creative expression. In the short term, some actors in the creative industry are already securing lucrative deals (and some disputes are still developing, see critically, Pope). In the long term, some are questioning the sustainability of the fabric of copyright law, and its underlying notions. This contribution instead takes a look at copyright concerns in the medium-term, discussing the tools offered by Creative Commons (CC) licensing tools as a viable option for bringing together the heterogeneous needs of creators in the context of European Union (EU) copyright acquis.
It is understandable that many artists and creative industries are feeling the proverbial rug of copyright being pulled from under them. Especially in the EU, authors enjoy a high level of protection which secures revenue streams. As the EU is moving towards a new era of bespoke AI regulatory tools under the AI Act, which includes rules regarding synthetic content created by AI systems (Art. 50(2) AI Act) and rules for providers of “general-purpose AI models” (GPAI) who are required to put in place a copyright policy (Art. 53(1)(c) AI Act), none fundamentally amend copyright law as such. Although these rules have already been critically received, artists and the creative industries are taking a closer look at the tools available to protect themselves, with nearly 1000 participants taking part in the process of developing a GPAI “Code of Practice” (see the first and second drafts).
At the same time, public interests depend on the greater availability of copyright-protected content online. The EU has long maintained that economic interests of creators need to be reconciled with the goal of promoting ready access to information. Accordingly, the EU copyright acquis includes within it a suite of exceptions and limitations on the rights of rightholders. Indeed, exceptions and limitations to copyright serves as a crucial benchmark for user’s Fundamental Rights as recognised by the EU Charter. Such exceptions and limitations, including the adopted exceptions for purpose of “text and data mining” (TDM) under the Copyright in the Digital Single Market Directive, may pursue aims diverging from those of rightholders. Therefore, although TDM exceptions have been a recent target for the scorn of creative industries, they need to be viewed in their contribution towards the initial goals of furthering the EU’s competitiveness and scientific leadership, and their overarching role for research. Ultimately, including also in the context of specific innovative projects, AI will need to navigate both sides of copyright.
How can the concerns of the creative industries as well as of the public be taken seriously? Clearly, there exists a possibility that content published online may be used for the compiling of datasets, including for research purposes. Such datasets may be collected, serve as the basis for training AI models, and resulting outputs of AI models may be deemed infringing. This raises questions with thus-far-ungeneralisable answers under copyright law.
Nevertheless, part of the answers may be provided by licensing practices. As some of the affected content contained in such datasets may be published under CC licenses, a suitable approach needs to be addressed through the CC ecosystem. CC-Plus, a protocol to the CC suite of licenses that allows for additions not addressed by the core licenses, can play a crucial role. In general, CC licenses can come with stipulations that must be respected by users, an authorisation by an exception or limitation notwithstanding. Some CC licenses may already be interpreted to prohibit the use of content for the collection and training of AI models in certain instances, such as CC-BY-NC or CC-BY-ND. CC-Plus offers a pathway for rightholders to exploit and share content beyond the permissions of a CC license. This provides rightholders with the opportunities to reap the benefits for certain – including AI-related – use cases, while ensuring that content can still be generally accessed by the public for permitted uses.
CC as an organisation has also closely followed the concerns of many authors regarding the impacts of generative AI and possible responses the CC community can spearhead. CC licenses may eventually be due an update, yet we may still be far from an AI-specific parameter (CC-BY-AI?) to the core licenses. CC-Plus can thus provide not just an intermediate solution for ensuring content remains available, e.g. for non-commercial purposes, but potentially provide a starting point for expressing AI-specific terms and appropriate licensing practices for AI training datasets. As a scholarly discussion on CC and generative AI is emerging, previous work already demonstrates the malleability of the CC-Plus protocol.
CC licenses, and content published under them, will remain part and parcel of the internet’s landscape. CC licenses have already blazed a commons-based trail and have allowed many authors to share their works in a manner that enhances access and re-usability. How open-access content can continue to contribute to a more just and egalitarian digital society is a fundamental question that will need to accompany forthcoming copyright discussions. As certain artists hope to fight back against AI, a blanket condemnation of CC may go amiss, especially as this could deprive access to content for uses unrelated to AI. In addressing the material power imbalances in the “open” AI ecosystem, discouraging the use of CC licenses could be a step towards a ‘dead internet’. Working with CC and the CC-Plus framework can be a viable approach in negotiating the obstacles ahead.
Article published originally on the 25th of February, here
This article gives the views of the author(s), and does not represent the position of CiTiP, nor of the University of Leuven.
“CC Icon Statue” by Creative Commons, generated in part by the DALL-E 2 AI platform. CC dedicates any rights it holds to this image to the public domain via CC0.
