The rapid evolution of generative artificial intelligence (AI) has ushered in a new era of musical creation, automating composition, arrangement, and sound synthesis.
However, AI creativity is the result of recombining existing data rather than generating fundamentally new cultural meanings. This calls into question the established triad of “composer–work–audience,” blurring the boundaries of authorship and precipitating a crisis in the understanding of creative agency.
The problem becomes particularly acute in the context of using AI to work with elements of cultural heritage. Algorithms trained on corpora of traditional works (for example, Mongolian long songs) can synthesize hybrid compositions while ignoring their historical context. This practice strips cultural markers of their meaning, turning them into commodities and exacerbating imbalances in the global musical ecosystem, where platform dominance marginalizes local practices.
This study is an empirical investigation that systematizes and analyzes the views of the expert community on the risks of deploying generative AI. The goal of the paper is to analyze the ethical aspects of applying generative AI to music, with a focus on the risks of cultural appropriation of Chinese heritage and on developing regulatory mechanisms capable of balancing innovation with the protection of creative expression.
1. Generative Artificial Intelligence in Music
1.1 Current State and Ethical Stakes of Creating Music with Generative Artificial Intelligence
All musical generative systems belong to the category of “narrow AI.” Unlike the hypothetical “strong AI” (Artificial General Intelligence, AGI), which would possess human-like cognitive flexibility and consciousness, narrow AI is capable of solving only specific, narrowly defined tasks, such as generating a sequence of notes in a given style.
Contemporary deep learning models—including recurrent neural networks (RNNs), LSTM (Long Short-Term Memory) architectures, transformers, and generative adversarial networks (GANs)—are trained on massive datasets to produce new content. It is this training and generation process that raises ethical questions: what data are the models trained on, who are the authors of those data, and do the algorithms reproduce biases embedded in the training samples?
The choice of architecture (RNN or transformer) and of training data are not neutral technological decisions but acts that help shape the future cultural landscape.
1.2 Collaborative Nature of AI Use and the Spectrum of Creativity
Artificial intelligence functions as an assistive system, generating alternative harmonic progressions and rhythmic patterns, thereby enhancing learning efficiency and the creative engagement of novice users.
However, AI in music is not an “autonomous creator” but an instrument, co-author, or “creative agent” that always operates in tandem with a human who frames the problem, sets the context, and evaluates the outcome.
This interaction raises the question of the nature of creativity itself. “Soft” creativity is the generation of new ideas within an existing cultural paradigm (for example, writing a pop song according to the conventions of the genre). “Hard” creativity involves changing the paradigm itself and creating a fundamentally new artistic language.
Is AI capable of “hard” creativity? So far, no—it recombines what is already known. However, by calling into question the exclusivity of human creativity, the mere existence of AI forces us to reconsider the boundaries of that spectrum.
1.3 Ethical Dilemma of Generative Music Creation Using AI
Traditional musical creativity was based on stable relations within the triad “composer–work–audience,” where the composer served as the source of creative intent, the work was regarded as the outcome of their intentions and skill, and the audience as the recipient of the aesthetic message. Generative artificial intelligence blurs these boundaries, provoking a crisis in the definition of authorship. This leads to a diminished central role for humans in creative activity and sparks ethical and philosophical debate about the legitimacy of ascribing subjectivity to machines—including authorial, moral, and phenomenal subjectivity.
By authorial subjectivity we mean the capacity to be the source of creative intent; by moral subjectivity, the capacity to bear responsibility for the consequences of one’s creative acts; and by phenomenal subjectivity, the possession of an inner experiential life, which in the context of narrow AI is more a metaphor than a real property. The automation of functions that previously relied on human experience leads to skill erosion and the devaluation of traditional abilities.
A more nuanced approach is required when discussing the impact of AI on the standardization of music. Concerns that AI will lead to homogenized musical content are valid, but it is important to recognize that popular music underwent extensive standardization long before AI emerged.
1.4 Risk of Imbalance in Musical and Cultural Ecology
This study employs the concepts of “musical” and “cultural” ecology. In this context, “cultural ecology” refers to the set of practices and institutions aimed at preserving cultural diversity and preventing processes of cultural homogenization that threaten the sustainability of local and transnational musical communities.
The concept of “musical ecology” is treated here not as a harmonious system but as a field of competitive struggle: the dominance of global platforms, comparable to the expansion of invasive species, contributes to the weakening and marginalization of local musical practices and jeopardizes cultural diversity.
AI methods isolate components of ethnic music from large-scale datasets, often without accounting for their cultural and sociocultural contexts. This practice produces representational and structural distortions, fosters homogenization and an uneven distribution of attention within the musical ecosystem, and can exacerbate imbalances in the status and perception of musical diversity.
Contextual deprivation occurs when AI algorithms generate so-called “pseudo-ethnic” music, detaching cultural markers from their original sociocultural practices. As a result, the melodic-rhythmic and semantic elements of Mongolian urtyn duu (long songs), which are closely tied to a pastoral-nomadic way of life, or the distinctive structural, vocal, and rhythmic techniques of Peking opera, associated with Confucian norms, are reduced to raw material for electronic dance music (EDM), losing their authenticity, functional-semantic content, and communicative significance.
This phenomenon can be described as technological colonization. Therefore, the ethical implications of AI in music directly concern the preservation of cultural diversity and fair representation in the global digital sphere.
2. Methodology
The primary data collection method consisted of semi-structured surveys.
Participant selection was conducted using purposive sampling, targeting Chinese experts with relevant experience. This qualitative methodology allows the research to focus on informed respondents capable of providing detailed empirical data and a deep understanding of the phenomenon under study.
The focus on Chinese respondents was intended to ensure contextual relevance for analyzing risks of cultural appropriation related to Chinese heritage.
The sample was balanced to represent key stakeholders across music, AI, law, and culture.
A key criterion for determining sample size in qualitative studies (such as grounded theory or thematic analysis) is data saturation. The sample size (N = 23) was deemed sufficient to achieve data saturation.
3. Results
The literature review confirmed that modern generative AI models (transformers, GANs, LSTMs) demonstrate strong technical capabilities for music generation, but their “creativity” largely consists of data recombination rather than the production of new cultural meanings. Using elements of traditional music intensifies ethical dilemmas around authorship, data transparency, and benefit sharing.
A survey of 23 experts from China confirmed the theoretical risks: 95% of respondents are concerned about cultural appropriation (e.g., Beijing opera, Mongolian long songs); 90% identified the problem of unlicensed datasets and demand transparency. Experts acknowledge the legitimacy of fair use but note that it does not resolve the ethical problem of using heritage without community consent. Views on AI’s creative agency vary: 65% allow for agency in hybrid models, while 30% insist creativity is exclusively human.
Experts endorse transparency of data and model interpretability (90%); using blockchain to trace data provenance and distribute remuneration (80%); a risk-assessment matrix (75%); and prioritizing human-created content in algorithms (92%). The concept of a “cultural alignment threshold” (70% support) raised concerns among 30% of respondents about risks of censorship and excessive rigidity.
Based on empirical data, a “three-level ethical architecture” (data/algorithms, communication, governance) is proposed as a normative proposition, aligned with and grounded in empirical findings. Recommendations are framed as policy options and design principles addressed to regulators (transparency standards), developers (ethical design), musicians (participation in evaluation), and the international community (harmonization of approaches).
IMPORTANT: note that this article contains scientific references that we have omitted to lighten the text. Please consult the original article if you wish to quote excerpts.
Contact us at mmiam@hec.ca to request a free copy of the full article published in the International Journal of Arts Management, Volume 29, Number 2, Winter 2027.
You can also browse our abridged research articles here.