Before ChatGPT and Gemini, copyright clashes erupted over much simpler technologies. Copyright holders warned that search engines-tools that copied works at scale-would stifle creativity. Those arguments failed in court, which recognized that copying for understanding and indexing is a basic aspect of a free and open internet. Today, a similar debate is unfolding around artificial intelligence, centering on whether copyright should control how existing works are analyzed and built upon.

Fair Use and the Automation of analysis

Table of Contents

U.S. courts have consistently affirmed that analyzing, indexing, and learning from copyrighted material falls under fair use principles, a protection that doesn’t vanish simply because a machine is doing the work.

Key Takeaways:

  • AI training is a transformative use,akin to studying language,not replicating works.
  • Expanding copyright to control analysis would harm science, medicine, and innovation.
  • The Bartz v. Anthropic case provides a strong legal framework for fair use in AI training.
  • Copyright is not the solution to concerns about automation and worker displacement.
  • Fair use safeguards access to knowledge and prevents copyright from becoming a barrier to progress.

Throughout history, new technologies have built upon existing knowledge. The printing press, photography, and the internet all relied on the ability to copy, transform, and disseminate data. These technologies didn’t diminish creativity; they unlock new avenues for knowledge and expression.

Training AI models operates within this established tradition.An AI system learns by identifying patterns across numerous works, not by reproducing or replacing them, but by extracting statistical relationships to generate new outputs. This process embodies a transformative use.

Restricting AI training based on copyright concerns misconstrues the core issue. Expanding copyright to demand permission for analysis or learning would extend far beyond generative AI, potentially threatening established practices in machine learning and data mining crucial to advancements in science, medicine, and technology.

Researchers already rely on fair use to analyze massive datasets, like scientific literature. Imposing licensing requirements would be impractical, favoring only large corporations capable of negotiating broad agreements. Fair use safeguards access to knowledge and prevents copyright from becoming an insurmountable barrier to understanding.

A Path Forward: Bartz v. Anthropic and Beyond

The case of Bartz v. Anthropic offers a promising framework for evaluating these issues. The court resolute that utilizing copyrighted works to train an AI model constitutes a highly transformative use, akin to studying language rather than replicating existing books. Any potential harm to the market for the original works was deemed speculative.

The court in Bartz dismissed the notion that an AI model infringes simply as its output might, in some way, compete with existing works. While some disagree with aspects of the ruling, the court’s stance on AI training and fair use provides a sound approach. Courts should prioritize whether the training is transformative and non-substitutive, rather than succumbing to speculative fears about market disruption.

Addressing Concerns Without Expanding Copyright

Concerns about automation and its impact on workers are legitimate and deserve attention. However,copyright is not the appropriate tool to address these challenges. Managing economic transitions and protecting workers are vital governmental functions, but copyright law offers no solutions in this regard.Expanding copyright control over learning and analysis won’t halt worker automation-it never has-but it will distort copyright law and stifle free expression.

Imposing broad licensing mandates could further consolidate power among dominant tech companies. Only the largest firms can afford the extensive licensing deals required to cover millions of works, effectively excluding smaller developers, research teams, nonprofits, and open-source projects. Copyright expansion won’t curb Big Tech’s influence; it will amplify it.

The Enduring Importance of Fair Use

Learning from existing work is fundamental to free expression, and rightsholders should not be permitted to control it. Courts have rejected this concept before, and they should do so again. Search, indexing, and analysis didn’t destroy creativity-