How copyright laws adapt to Generative AI
With the advancement of generative artificial intelligence (AI) technologies, copyright issues have become increasingly relevant. A recent analytical report by the RAND Corporation highlights key aspects of this challenge, including the protection of AI-generated works, the legality of training AI on copyrighted materials, and differences in legislative approaches across countries, Kazinform News Agency correspondent reports.
Key challenges related to artificial intelligence and copyright span several primary areas.
AI Training
The use of copyrighted materials for training AI remains a contentious issue. At its core lies the question of whether such practices can be justified under existing legal frameworks.
AI model training often falls under the "fair use" doctrine, allowing the use of copyrighted materials without permission in certain cases, particularly when serving transformative purposes, like data analysis. However, many rights holders argue that these practices undermine the licensing market and harm their commercial interests.
Competition and copyright disputes
The impact of generative AI models on the market and creative industries is becoming increasingly evident. These technologies can produce works that compete with those of human creators, replicating artistic styles or imitating distinctive features of well-known creations.
Legally, two key issues remain contentious: the permissibility of training AI models on unlicensed data and the legitimacy of generated content that closely resembles original works. Particular attention has been drawn to cases where models reproduce recognizable elements, such as artists’ signatures or famous characters (e.g., Mickey Mouse).
Generative AI could be used to mimic the style of artists, such as singers, illustrators, or writers. While style itself is typically not protected by copyright, using specific works to create imitations could lead to legal claims.
Proponents of AI technologies, on the other hand, emphasize their social and economic benefits. They argue that these models drive progress in art, medical research, and technology, boosting productivity and accelerating innovation.
Copyright Law
Countries approach the protection of works created using AI and the regulation of AI training on copyrighted data differently.
United States
In the U.S., copyright applies exclusively to works created by humans. Fully AI-generated content without significant human involvement does not qualify for copyright protection. For instance, if a user directs AI to compose a poem in the style of a famous artist, the resulting expressive elements, produced solely by the AI, are likely unprotectable.
However, when AI is used as a tool and the human creator contributes substantial creative input, the work may qualify for protection. Simple uses of AI tools, such as spellcheck or photo filters, do not affect copyright eligibility, while basic commands to generate AI-created works generally fall outside copyright protection.
The legal stance on works between these extremes, where assistance and generation blur, remains unclear. Importantly, under U.S. law, the user of the AI tool, rather than its programmers, is considered the author of any qualifying work.
European Union
The EU follows a similar approach to copyright, recognizing only humans as authors. However, the European Court may consider exceptional cases where AI is used in complex creative processes.
The EU has introduced an opt-out system, allowing rights holders to explicitly prohibit the use of their materials for AI training. For research and academic purposes, text and data mining (TDM) is permitted without consent, but rights holders can enforce restrictions for commercial use.
EU legislation requires AI developers to publish detailed summaries of the content used for model training, including major data sets, databases, and other sources, along with explanations of their use. While this strengthens the protection of authors' rights, it may complicate market entry for AI developers in the EU.
United Kingdom
Initially, the UK considered broad exceptions for text and data mining in commercial contexts but faced opposition from creative industries. As a result, TDM is now limited to scientific research purposes only.
China
China recognizes copyright for AI-generated works, including those created with simple prompts or instructions. Сopyrightable works must be original and reflect intellectual achievement, among other things.
A Chinese court case involved AI-generated images where the defendant used and removed the watermark from works shared by the plaintiff. The Beijing Internet Court has ruled that the plaintiff's images created using generative AI are original.
The plaintiff actively designed key elements of the images, such as characters and their representation, through prompts. Additionally, the iterative process of refining the images further emphasized originality through aesthetic decisions.
Singapore
Singapore includes a broad exception for text and data mining (TDM), allowing both commercial and non-commercial entities to use copyrighted works for computational data analysis. A key condition for this exception is that users must have "lawful access" to the works they intend to analyze.
Other countries
Japan provide TDM exceptions to the general protection of rights holders provided by copyright law. In Latin America legislation is still in development. Kazakhstan adopts a principle similar to that of the U.S., recognizing only human authorship for copyright protection. However, there are currently no specific laws regulating AI use in the country.
