L'IA peut-elle composer de la musique originale à la manière des grands compositeurs classiques ?

Les modèles d'IA peuvent désormais générer des partitions orchestrales complètes indistinguables de celles de Bach, Mozart ou Beethoven. Ces compositions s'adaptent aux règles de mood, tempo et instrumentation de chaque époque. Certaines IA ont même trompé des auditoires lors de tests d'écoute à l'aveugle. Cela remet en question nos notions de créativité et de paternité. La technologie est utilisée dans la composition musicale pour le cinéma et les médias interactifs.

Background

AI systems today can generate short pieces that stylistically resemble works by famous classical composers such as Bach, Mozart, or Chopin, using neural networks trained on large corpora of historical scores. Models like MuseNet and transformer-based architectures can produce four-part chorales in Bach’s style or waltz-like piano pieces after Chopin. These outputs often lack the depth, coherence, and emotional nuance of human compositions. While the systems mimic surface-level stylistic features—melodic contours, harmonic progressions, and rhythmic patterns—they typically do not achieve the structural originality, thematic development, or expressive intent characteristic of canonical classical works. Outputs are generally limited in length and tend to blend features rather than produce truly novel, historically faithful compositions. Despite these limitations, some AIs have fooled audiences in blind listening tests, raising questions about creativity and authorship. The technology is also being adopted for film scoring and interactive media.