AI‑driven music generators are increasingly shaping what listeners hear, but new analysis shows they heavily favor Western genres while sidelining traditions from Africa, the Middle East and South Asia. researchers point to skewed datasets and a recommendation system that rewards popularity, creating a feedback loop that marginalises independent and non‑Western artists.

Turkish Makam flattened into standard Western pitch

According to the report, a study of AI‑generated compositions revealed that the Turkish Makam system – a melodic framework built on micro‑tonal intervals absent from the piano – was routinely reduced to conventional Western tuning.. This distortion erases the distinctive tonal colors that define Makam,effectively Westernising a centuries‑old musical language.

Western genres dominate AI music training sets

The same source notes that datasets feeding popular AI models contain a disproportionate share of rock, pop and electronic tracks, while recordings from Africa, the Middle East and South Asia represent a fraction of the total. As a result, AI‑generated outputs mirror the same imbalance, reinforcing the perception that “Western music” is the default.

Popularity bias fuels a self‑reinforcing recommendation loop

Industry infrastructure relies on play counts, saves and playlist histories to train recommendation algorithms. the report explains that this volume‑driven filtering amplifies already‑popular tracks, pushing lesser‑known, non‑Western songs further from mainstream visibility. The cycle not only limits discovery but also skews royalty calculations toward dominant genres.

Metadata gaps cripple non‑Western music visibility

Inconsistent tagging and incomplete metadata, as highlighted by the analysis, cause AI systems to misclassify or overlook tracks from underrepresented regions. These errors cascade into wrong recommendations,missed royalty payments and poorer training data, deepening the marginalisation of independent artists.

Will industry metadata standards improve?

The report leaves open whether music‑industry bodies will adopt uniform metadata protocols that capture regional nuances. Without coordinated action, the bias embedded in decades of data is likely to persist, keeping AI‑generated playlists homogenous.

As the article repeatedly stresses, “the bias is embedded in decades of music data, resulting in AI systems that prioritize Western styles and ignore other cultures.” Addressing these structural flaws now is crucial to prevent further marginalisation of global musical diversity.