The Recording Industry Association of America (RIAA)'s recent copyright infringement lawsuits against AI music generation companies like Suno and Udio are a misguided attempt to stifle technological innovation and maintain control over the future of music creation. While the RIAA's concerns about the impact of AI on the music industry are understandable, their legal arguments are fundamentally flawed and unlikely to succeed in court. This essay will explore the key reasons why the RIAA's lawsuits are doomed to fail.
The core of the RIAA's argument is that AI companies have committed copyright infringement by training their models on vast datasets of copyrighted songs without permission. However, this argument ignores the well-established legal doctrine of fair use, which allows for the use of copyrighted material without permission for transformative purposes such as commentary, criticism, or creating new works.
Training AI models on copyrighted music is a quintessential example of fair use. The AI systems are not simply copying or distributing the original songs, but using them as raw material to learn patterns, styles, and techniques which they then use to generate novel compositions. This is a highly transformative use that does not substitute for the original works in the market.
Courts have long recognized that copying for the purpose of analysis, learning, and creation of new works is protected under fair use. In the seminal case of Authors Guild v. Google, the court held that Google's scanning of millions of copyrighted books to create a search database was fair use because it was transformative and did not harm the market for the original works. Similarly, in Sony v. Universal City Studios, the Supreme Court ruled that private, non-commercial copying of TV shows for the purpose of time-shifting was fair use.
The RIAA's attempt to paint AI training as simple verbatim copying ignores the fundamental nature of machine learning. Just as a human musician might listen to thousands of songs to internalize musical concepts and inspire their own compositions, AI systems ingest large datasets to learn the building blocks of music creation. Treating this learning process as infringement would criminalize the very essence of AI development.
Perhaps the most glaring flaw in the RIAA's legal strategy is that it directly contradicts positions the organization has taken in previous copyright cases. In a 2017 amicus brief filed in the "Blurred Lines" copyright case, the RIAA itself argued that:
"...new songs incorporating new artistic expression influenced by unprotected, pre-existing thematic ideas must also be allowed. Most compositions share some elements with past compositions—sequences of three notes, motifs, standard rhythmic passages, arpeggios, chromatic scales, and the like. Likewise, all compositions share some elements of "selection and arrangement" defined in a broad sense. The universe of notes and scales is sharply limited. Nearly every time a composer chooses to include a sequence of a few notes, an arpeggio, or a chromatic scale in a composition, some other composer will have most likely "selected" the same elements at some level of generality."
In other words, the RIAA was arguing that copyright should not be extended to cover general musical ideas, styles, or short phrases because doing so would make it nearly impossible to create new works without infringing. They emphasized that allowing artists to be inspired by and build upon the unprotectable elements of prior works is essential to musical creativity.
Fast forward to the present lawsuits, and the RIAA is now claiming that AI systems infringe copyright simply by learning from these same unprotectable musical elements. They are arguing that the AI-generated songs are derivatives of their copyrighted works because they exhibit similarities in genre, style, instrumentation, and compositional techniques.
This is a blatant contradiction of the RIAA's prior position. If human artists must be free to learn from and incorporate generic musical elements, then surely AI systems should be afforded that same freedom. The RIAA cannot logically argue that "arpeggios, chromatic scales, and the like" are unprotectable for human composers but off-limits for AI.
The RIAA's own prior arguments fatally undermine their current claims. Any court considering these cases will likely see the glaring inconsistency and be highly skeptical of the RIAA's attempt to apply a double standard to AI music generation.
Beyond the RIAA's self-contradictory arguments, there is a strong affirmative case that the music generated by AI systems like Suno and Udio are original works that do not copy from any specific copyrighted compositions.
While these AI models are trained on datasets that include copyrighted songs, the process by which they generate new music is fundamentally different from mere copying. The models learn complex statistical patterns and relationships from the training data, which they use to probabilistically generate novel sequences of musical notes, chords, and structures.
The resulting compositions, while exhibiting stylistic similarities to the training data, are not direct copies of any particular songs. They are new permutations of musical elements filtered through the AI's learned representations. Courts have long held that works generated through such transformative recombinations and permutations of uncopyrightable elements are themselves original works entitled to their own copyright.
For example, in the case of Tresona Multimedia v. Burbank High School Vocal Music Assoc., the court ruled that a mashup combining elements of multiple songs was an original work because "the addition of new expression to an existing work is the very essence of authorship". Similarly, the AI music at issue here adds substantial new expression by generating novel arrangements and combinations of musical motifs derived from its training.
Moreover, the AI-generated songs respond directly to user prompts and parameters, imbuing them with creative input from the human users as well. The AI is in essence a tool for users to explore a vast probabilistic space of potential songs matching their desired style and mood. This human-machine collaborative creativity further distinguishes AI music from mere verbatim copying.