As AI systems grow more sophisticated and transform human expression, they raise questions about copyright law. Who owns the output of these systems and how can they be protected?
GAI software may be infringing copyright protections if it uses unauthorized third-party materials to generate its output. However, this does not mean that these outputs are not eligible for copyright protection.
Machine Learning
Generative AI has the potential to transform how artwork is produced, software is developed and written, and text is created. But this disruptive technology also raises new legal questions, particularly about intellectual property protections of these works. As a result, copyright law may need to be updated to address these new realities.
The first area of consideration is whether AI-generated work qualifies for copyright protection. Generally, copyright protects works that are original and expresses some form of human creativity. A work produced by GAI is considered to satisfy this requirement if it can be shown that the AI process involved substantial creative control over the work’s expression, and the finished product reflects a high level of creativity that could not have been achieved without the use of the GAI.
However, this determination will be based on a case-by-case assessment of the individual situation. It is unclear whether this test will be sufficient to protect AI-generated works against infringement.
Another important issue arises from the use of copyrighted materials in AI training data. This issue has resulted in lawsuits around the world alleging infringement of copyrighted works by owners of AI systems. However, it is difficult to assess the scope of this problem due to the secrecy surrounding AI training methods and proprietary nature of AI systems more generally.
The Copyright Office has recently issued guidance on this issue, and will be holding public listening sessions and publishing a notice of inquiry to seek input on other legal issues related to AI. Until these legal issues are resolved, businesses using AI should consider reviewing their transaction terms to ensure that they have adequate protections in place and demanding from generative AI platforms proof of proper licensure of the training data used to create outputs. They should also consider seeking broad indemnification for any infringement of IP rights by AI platforms that fail to comply with these requirements.
Additionally, businesses should consider requiring attribution through their terms of use, and implementing a “notice of use” system that requires all users to acknowledge the use of an AI-generated work. This would be consistent with the USCO’s new guidelines on copyright registration and Australia’s non-binding AI Ethics Framework, which calls for transparency and responsible disclosure in the use of AI technologies.
Neural Networks
Neural networks are one of many different tools and approaches used by machine learning algorithms to process complex data inputs into a form that computers can understand. The concept behind neural networks is based on the way neurons function in the human brain to translate information from our senses and perceive our surroundings. While these algorithms are important for advancing AI, their use raises some interesting issues under copyright law.
For example, the collection and training of AI algorithms often involves large amounts of data, including copyrighted works such as pictures, music, videos and other media. These copyrighted works are used to identify patterns and extrapolate conclusions from the data, which then provides the output of the algorithm (such as an answer to a question, a new painting or song). However, some courts have found that using these copies of copyrighted materials to train neural networks infringes copyright in the original work.
Similarly, the proprietary nature of these algorithms and their tendency to be protected as trade secrets makes it difficult for anyone other than their creators to understand how they function. This lack of transparency interferes with the ability to discern whether an algorithm has been biased or discriminated against, or has committed copyright infringement.
Another issue is the difficulty of determining who owns an AI-generated work or invention. In response, the World Intellectual Property Organization (WIPO) has been conducting a series of sessions with Member States and stakeholders to identify the nature of any IP rights in these creations.
As the world begins to adopt and rely on AI, there is a growing call for greater transparency and accountability regarding its functions. However, this call has been complicated by the fact that these algorithms are typically developed and owned by private companies, who have an incentive to protect their investments. Interrupting the traditional protections that copyright law affords authors—such as initial vesting of ownership and transferability of copyright—could discourage humans from embracing and contributing to AI development, depriving society of the tremendous benefits it offers.
Deep Learning
Generative AI is a revolutionary technology that has the potential to revolutionize how artwork is created, software is developed, and text is written. However, this disruptive innovation raises a number of new legal questions related to the intellectual property protections of these new works—in particular, copyright.
Infringement or non infringement?
The most pressing issue relates to the use of copyrighted works in the training of generative AI systems. Most GAI programs rely on large amounts of publicly available data that the program copies, stores, and learns from in order to create its outputs. The question is whether this activity constitutes a “reproduction” or other exclusive right of the copyright owner without permission.
If the underlying work has copyright protections, the AI system’s use of that underlying work in its output may constitute a copyright infringement under Section 120(1) of the U.S. Copyright Act. However, if the underlying work is not protected by copyright, there may be an argument that the resulting output qualifies for fair use under Section 107 of the Copyright Act.
As a result of these issues, companies that use generative AI should be sure to evaluate their transaction terms to ensure they are not infringing on copyrights by failing to secure the proper licenses for the input data used in the creation of their AI products and services. Businesses should also consider seeking broad indemnification for intellectual property infringement arising out of the use of generative AI tools, including demanding terms of service from generative AI platforms that confirm proper licensure of training data or self-reporting by those tools to flag their usage of unauthorized works as appropriate.
As the legal landscape continues to evolve, it is important for businesses that use generative AI to keep abreast of developments in the law. The Copyright Office’s recent guidance on the registration of AI-generated works clarifies that applicants have a duty to disclose and disclaim the use of AI in their applications for copyright registration and must update any pending or previously filed applications to reflect this disclosure and disclaimer. The Copyright Office also has announced plans to publish a notice of inquiry and hold public listening sessions to seek input on further legal and policy topics involving AI technology.
Artificial Intelligence
Generative AI, which uses data lakes and question snippets to recover patterns and relationships, is becoming increasingly common in creative industries. However, the legal implications of using such systems are unclear, particularly in relation to copyright infringement claims.
Until recently, there were no federal copyright laws specifically addressing the use of generative AI. This created a confusing situation for both businesses and creators who are using this technology. In some cases, if human input is present in the final work, it may be possible to register the work under the same categories as other works that are created solely by machine. In other cases, however, it is more difficult to make that claim.
This confusion may be resolved by future court decisions or new copyright law amendments. For now, businesses and creatives who are using generative AI should consider adding disclosures to vendor or customer agreements if the resulting works will be registered for copyright. These disclosures should include language that specifies that any AI-generated works are subject to the same intellectual property rights as other works produced by humans and that the user will abide by the same standard for determining authorship of the work.
It is also a good idea to review policies regarding employee use of generative AI tools, to ensure that employees are aware of any copyright-related issues that could arise. This will help to reduce the risk of unauthorized uses and potential infringement claims.
Additionally, a review of company policies regarding training and data sources for the AI systems used by the business should be undertaken. This will be important because of the potential for biased training data to introduce vulnerabilities into the AI systems, which could affect their performance or create security concerns.
It is worth considering whether copyright law should be amended to clarify that only a human author or artist can claim copyright protection for works generated solely by an AI system. In light of the increasing number of cases where generative AI has been accused of copying preexisting works, and the trend toward loosening originality standards that reward skill, labor and effort, such an amendment would seem to be a reasonable option.