The AI Copyright Grab: Why Expanding Copyright Hurts Us All

Stifling Innovation and Research

Imagine needing permission to read—whether with your eyes or with the help of software. That's the potential consequence of expanding copyright for AI training. AI tools offer countless benefits, but forcing developers to license training data threatens innovation and inclusivity in AI models. It also hinders crucial uses of AI in creative expression and scientific breakthroughs.

Restricting fair use for AI training data could cripple socially valuable research in machine learning, text, and data mining. These fields have thrived under fair use principles, leading to vital advancements. Licensing the massive datasets required for high-quality research is often prohibitively expensive and practically impossible.

Fair use safeguards crucial scientific progress, and empirical studies confirm this: research using text and data mining thrives in countries that protect it from copyright restrictions. The opposite is also true - restrictive copyright laws stifle beneficial research.

The stakes are astronomical. Machine learning is essential for understanding our world, from distant nebulae to the very proteins within us. Imagine attempting to license the data from every telescope in the world! Machine Learning makes analyzing these immense datasets possible.

Crushing Competition and Empowering Monopolies

Requiring licensing for AI training would create an uneven playing field, favoring companies with vast data troves and deep pockets. This would lead to the predictable downsides of limited competition: higher costs, inferior service, and increased security risks. Furthermore, it would restrict the diversity of data used to train AI, impacting the expressive potential for users.

The Federal Trade Commission has warned that controlling AI training data allows a few powerful companies to "dampen or distort competition" and exert "outsized influence over a significant swath of economic activity."

The case of Thomson Reuters v. Ross Intelligence illustrates this danger. Ross Intelligence, aiming to challenge the legal research duopoly, sought to license training data from Westlaw. Westlaw refused and sued, ultimately forcing the promising startup into bankruptcy.

Getty Images, after suing Stability AI over training data, then launched its own AI image generator trained on its vast image library. This illustrates how large companies can use copyright to stifle competition and then introduce their own, controlled alternatives. This benefits tech monopolists, solidifying their market dominance through insurmountable barriers to entry for smaller players.

Silencing Expression and Creativity

Generative AI tools are powerful engines of creativity, making content creation—especially images and videos—accessible to a wider audience. By simplifying the process and reducing the need for specialized skills or expensive equipment, generative AI unlocks new forms of artistic expression.

Certain art forms, particularly those rooted in the African American community like hip-hop and collage, thrive on remixing existing works to create something new. Restricting training data would stifle this tradition and worsen the harm copyright law has already inflicted on these art forms.

Generative AI democratizes speech and content creation. Just as the internet empowered individuals to share their voices, generative AI enables ordinary users to tell stories and express opinions through easily generated text, graphics, and videos.

Undermining Fair Use

Fair use—the right to use copyrighted material in specific circumstances without permission—is a crucial counterbalance to restrictive copyright claims. However, publishers are attempting to impose a new licensing regime for AI training, despite lacking any recognized legal basis for this control. This undermines fair use for everyone.

Fair use is essential for academic research, education, competition, and critical commentary. By weakening fair use, the AI copyright grab exacerbates all the dangers mentioned earlier.

A Better Path Forward

There are legitimate concerns surrounding AI, including job displacement, potential for abuse (like non-consensual imagery), privacy violations, misinformation, and environmental impact. However, expanding copyright won't solve these issues.

Instead, we need solutions that address the root causes: strengthening labor rights, protecting privacy, promoting media literacy, and enforcing antitrust regulations. Targeting specific problems with specific policies is more effective than a blanket expansion of copyright.

Strengthening environmental protections, enacting comprehensive privacy laws, protecting workers' rights, and fostering media literacy are far more effective than relying on copyright expansion. These strategies build a resilient ecosystem against potential harms from new technologies, not just AI.