Week 7: Music Power and Technology-Feminist Readings of AI Music 1

Learning Outcomes

Upon completion of this material, students should be able to:

• Analyse the ways in which historical and contemporary technologies, with a specific focus on generative AI, influence the production, circulation, and regulation of music within established social and political structures.

• Apply feminist theoretical perspectives to critically evaluate the social, ethical, and environmental consequences of music generated by Artificial Intelligence.

Defining Feminism and Systematic Critique

Feminism is defined as a methodology for analysing and critiquing systems of power and their specific impacts on marginalised others. Its core functions include:

• Examination of Power and Inequality: Investigating the distribution of power within a system.

  • How does the group holding the power impact others?

    • Woman who of colour, disabled, elderly, etc.

• Inclusion and Exclusion: Asking critical questions regarding who is included in a system and who is excluded.

• Challenging Neutrality: Contesting what is traditionally treated as "normal" or "neutral."

• Revealing Hidden Labour: Highlighting the contributions and physical or intellectual labour that systems often obscure.

• Systemic Reproduction: Identifying how structures and systems work to reproduce existing inequalities.

Understanding AI-Generated Music

AI-generated music utilizes datasets and algorithms to produce music. This can result in entirely new compositions or variations of existing works. The technological framework includes:

• Techniques: Deployment of deep learning and neural networks.

• Architectures: Specific use of Generative Adversarial Networks (GANs), variational autoencoders, and transformers.

Sienna Rose

A singer that people are suspecting is AI generated. It is unclear are of right now if she is human or AI.

Music, AI, and Society

According to Hesmondhalgh & Meier (2018), "One striking feature of culture in modern capitalist societies is that the main ways in which people gain access to cultural experiences are subject to frequent, radical and disorienting shifts."

The societal importance of music is evidenced by the surveillance, control, and commodification of musical practices and institutions by colonialism, capitalism, and the patriarchy since the Renaissance. AI represents a "critical juncture" in this trajectory, characterized by:

• The culmination of intersecting systems.

• The complete integration of music into capitalist, patriarchal, and neo-colonialist structures.

The Historical Evolution of Copyright and Control

Control over musical circulation has evolved through distinct eras of regulatory and technological shifts:

• $12^{th}$ Century: The Era of Copyists: The Church and monarchs actively prevented the spread of material deemed heretical or undesirable.

  • You could only hear specific pieces of music at specific churches.

  • They would choose what could be notated distributed and what was withheld.

• $15^{th}$ Century: The Era of the Printing Press: While this broadened distribution, it simultaneously tightened regulation. Monarchs and religious institutions granted "royal privileges" to specific printers, creating monopolies and facilitating censorship.

  • Authorship had no legal rights/protection. It was all under the printing presses.

• $1710$: The Statute of Anne: This legislation protected authors and facilitated the growth of the early music-printing industry.

  • This was specifically for literary works- had not broadened to musical compositions and were not recognized as protected works.

• $1793$: French Copyright Law: This landmark law granted musicians the legal right to print, sell, and distribute their own works.

  • Affected musicians 6000 years after the concept of copyright had began.

• $1853$: Union of Authors, Composers, and Music Publishers: This allowed musicians to demand royalties for every performance, shifting power away from financiers and toward the creators.

• $20^{th}$ Century: The Era of the Record Company: Record companies shaped the economy and accessibility of music through the ownership of physical music production and copyright.

  • Record labels shaped the accessibility of music and where it went. They controlled what was produced and to whom and where it was shared.

  • Began the process of music becoming international.

• $1980$s – $2000$s: The Era of Tech Companies: Technology companies became the dominant force in music consumption. Music experience was mediated through CDs, cassettes, and eventually subscription-based streaming and mobile devices.

• $2023$ - Present: The Era of Generative AI: Commercial AI companies often circumvent existing copyright frameworks by training models on copyrighted music without obtaining prior permission or licenses.

• Copyright has always been about controlling profit and the distribution of music. The profits have never been given to the musicians in a just manner.

Digital Neocolonialism and Environmental Ethics

Digital neocolonialism is defined by the reassertion of the exploitation of labour, data, and the environment for the benefit of AI companies and their home countries. It mirrors historical colonialism by creating dependency systems through external governance and the exploitation of resources in the Global South.

Environmental and Labour Impacts

The construction of AI infrastructure involves:

• Deforestation: Decimating forests to acquire materials.

• Water Contamination: Poisoning critical water sources.

• Labour Exploitation: Utilizing the labour of children and climate refugees.

• Economic Dependency: Leaving local populations impoverished while they remain dependent on these systems.

• Energy Use: Reactivating dormant nuclear power plants to meet the energy demands of AI.

• Displacement: Many communities are being displaced because of the environmental crisis, and the only way for them to make profits to live off of are through using the technology that is causing their displacement.

The power of AI development is affecting minority groups and their living spaces and labour exploitation.

AI Music in Academia: Music Information Retrieval (MIR)

AI music generation is rooted in the academic field of Music Information Retrieval (MIR). MIR focuses on making music "computationally legible" through:

• Dataset curation and annotation.

• System evaluation.

• Music models.

Data Selection and Dataset Curation

Research in MIR often prioritizes licensed or public domain audio due to copyright restrictions, making datasets easier to share. This process influences how music is defined computationally.

Case Study: RWC (Real World Computing) Music Dataset (2000)

• Content: $5$ sub-collections (Classical, Jazz, Pop, Genre, and Royalty-Free).

• Scope: $328$ tracks, totalling approximately $23$ hours of audio.

• Methodology: All music was recorded specifically for research to avoid copyright issues, but this resulted in limitations based on budget, time, and available production resources.

Case Study: GTZAN Dataset (2002)

• Content: $1000$ audio clips representing $10$ genres (including Pop, Jazz, Hip Hop, Metal, and Country), with $100$ clips per genre.

• Issues: The dataset was built by technical researchers rather than ethnomusicologists. It contains significant errors: $7.2\,\%$ duplicate recordings and $10.6\,\%$ mislabelled tracks. Despite these flaws, it remains widely used in MIR research.

Gender Statistics in Music and Artificial Intelligence

Lewis et al. (2018) state: "It is clear to us that the country to which AI currently belongs excludes the multiplicity of epistemologies and ontologies that exist in the world."

Women in the Music Industry

• Producers: $3.2\,\%$

• Songwriters: $13.4\,\%$

• Artists: $22.3\,\%$

• Music Academia: Women hold $20\,\%$ of full-time positions.

• Commercial Studio Production: Men hold $95\,\%$ of roles.

• Record Company Leadership: Men hold $70\,\%$ of senior management and $85\,\%$ of top-level roles.

Women in the AI Workforce and Academia

• Global AI Roles: $22\,\%$ are held by women.

• Tech Industry Research: Women make up $15\,\%$ of AI research staff at Facebook and $10\,\%$ at Google.

• AI Start-ups: Only $2\,\%$ of venture capital investment goes to female-founded AI start-ups.

• AI Academic Faculty: Women hold only $16.1\,\%$ of tenure-track positions; over $80\,\%$ of AI professors are male.

• Research Contributions: Women account for $12\,\%$ of contributions at major AI conferences, $6\,\%$ of single-author papers, and $25.4\,\%$ of co-authored papers.

Questions & Discussion

Exercise 1: Who is making AI music? Groups were asked to research the following:

• AI Companies (Suno, Udio): Who owns the company? Where is the music sourced? How is generated music used? What are the copyright implications?

• Musicians (Grimes, Holly Herndon): Why did they build AI tools? Where was the music sourced? How is generated music used? What are the copyright implications?

Exercise 2: Case Study Analysis Evaluation of "Sienna Rose" — questioning whether she is a real artist or an AI creation.

Exercise 3: Dataset Impact Discussion on the impacts of dataset curation and labelling on music. Points of consideration included the absence of women in recorded music limiting their presence in AI training datasets, which erases contributions and amplifies gender inequality.