GenAI: Balancing innovation and copyrights & the challenges of Chile’s AI Bill

In the past two years, since OpenAI launched their GPT model to the users, Artificial Intelligence (“AI”) has experienced fast and exponential growth, especially in the field of Generative AI (GenAI). In this sense, on May 21, 2024, the Chilean government proposed an AI Regulation Bill, to regulate AI development. 

However, this legislation introduces changes to Copyright Law (N° 17.336), that could lead to potential conflicts between AI software developers, Intellectual Property right holders, and policymakers. This article explores the conflicts arising from this bill, specifically its implications for copyright law, and the difficult balance between innovation and the protection of the right holders.

The role of the copyrights in the development of GenAI

Generative AI has transformed industries by producing new data, text, images, audio, and video from vast datasets. These models rely on data training to function effectively, which involves analyzing large volumes of data, often protected by copyright laws. The use of copyrighted material without explicit permission raises several concerns about infringement and fair use.

Chile’s proposed AI bill introduces an exception to the Chilean Copyright Law, allowing the use of large datasets for data mining, provided it doesn’t result in direct commercial exploitation of copyrighted material. 

This exception is designed to support innovation by enabling developers to access copyrighted works for non-commercial purposes, particularly for AI training. The practical challenges that GenAI developers could face in securing licenses from every right holder whose works are included in massive datasets cannot be overlooked. The large volume of data required to train GenAI models makes it almost impossible to obtain individual permissions from all copyright holders, potentially putting developers at risk of infringement claims.

However, this could lead to tension with copyright holders, who may feel that their rights are being undermined without adequate compensation or even control.

The transparency requirement dilemma

One of the central requirements in AI regulation, as reflected in Chile’s AI bill, is transparency. Developers are required to disclose the datasets and methodologies used to train AI models. This is vital for ensuring fairness, avoiding bias, and maintaining accountability in AI systems. However, the demand for transparency raises concerns for developers about the protection of their trade secrets.

Many AI developers consider the algorithms, datasets, and techniques they use as proprietary information. Requiring them to disclose too much could expose valuable intellectual property of their core business. This creates a tension between complying with transparency requirements to foster trust and protecting trade secrets, which are essential to maintaining a competitive edge.

This conflict is not unique to Chile. In Europe, the AI Act also mandates transparency, and many developers have raised concerns about revealing too much of their algorithmic processes. The challenge lies in ensuring that enough information is provided for accountability without undermining the competitive interests of companies.

Data mining versus copyrights 

Data mining is essential for training the GenAI systems, but it presents several legal challenges when copyrighted content is involved. AI models require access to large volumes of input data, often drawn from text, images, and other media protected by copyright. The Chilean bill’s provision allows the use of this data for non-commercial purposes, but it is not always clear where the boundary lies between permissible use and copyright infringement.

One major concern is that AI systems could unintentionally generate content that closely resembles the works used in their training. While the AI may not directly replicate a copyrighted work (known as “digital replicas”), the similarity between the generated content and the original could lead to copyright infringements. This issue has surfaced in other jurisdictions such as the United States, where AI-generated outputs have raised questions about fair use and derivative works.

For example, in the European Union, the Directive on Copyright in the Digital Single Market (CDSM) allows text and data mining (TDM) for research purposes but gives copyright holders the ability to exclude their works from this exception. This provides a mechanism for protecting the rights of creators while allowing AI development to continue. Chile’s proposed bill, however, is less clear on the extent to which copyright holders can control the use of their works in AI training, potentially leading to conflicts and legal disputes.

Commercial exploitation and misuse

While Chile’s bill seeks to prevent unauthorized commercial exploitation of copyrighted materials, the distinction between non-commercial and commercial use can be difficult and challenging. Developers may argue that using copyrighted works for training AI does not directly result in a commercial product, but the resulting AI model could have a clear commercial value.

Even if the AI-generated content is not an exact copy of the original copyrighted work, it could still diminish the value of the original by producing similar outputs. For example, if an AI model generates artwork based on copyrighted images, it could reduce the demand for the original works, raising questions about the economic harm to the copyright holder.

Collaboration among the stakeholders

The success of Chile’s AI bill will depend largely on collaboration between developers, policymakers, and copyright holders. A key objective should be finding a balance where AI innovation can flourish without undermining the rights of creators. This will require clear guidelines on the use of copyrighted works for AI training, along with mechanisms for compensation that ensure that creators are fairly compensated for their contributions to AI systems.

A few conclusions

Chile’s proposed AI bill presents a unique opportunity to modernize intellectual property laws in the context of AI development. However, the legislation introduces potential conflicts, particularly in relation to copyright protection and data mining for AI training. 

The balance between innovation and protecting the rights of creators will be critical to the bill’s success.

Developers will have to navigate transparency requirements without compromising their trade secrets, and copyright holders will need assurances that their works will not be used unfairly and infringed. 

As GenAI continues reshaping the future, finding the right balance between these competing interests will be essential to ensuring sustainable technological progress.