Publicação
How do authorship labels shape listeners’ judgments of musical quality?
| Resumo: | This thesis examines how listeners perceive and evaluate AI-generated music across three preregistered studies. Study 1 investigates discriminability and shows that participants generally cannot distinguish AI-generated songs from human-composed ones (mean accuracy ≈ 44%), with performance varying by genre. Study 2 demonstrates a strong label-bias effect: songs labeled “Human-made” receive higher evaluations than identical songs labeled “AI-made,” regardless of true authorship. Study 3 tests whether removing vocals alters these evaluations and finds only modest effects, with authorship beliefs remaining dominant. Overall, results reveal limited discriminability, powerful framing effects, and genre-dependent patterns shaping responses to AI music. |
|---|---|
| Autores principais: | Rego, Vera Pardal Monteiro |
| Assunto: | Artificial Intelligence in music AI music perception Human-AI discriminability Label bias Authorship effects Vocal component Aesthetic experience with music Music evaluation Generative AI Music cognition Genre differences Algorithm aversio |
| Ano: | 2026 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso embargado |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This thesis examines how listeners perceive and evaluate AI-generated music across three preregistered studies. Study 1 investigates discriminability and shows that participants generally cannot distinguish AI-generated songs from human-composed ones (mean accuracy ≈ 44%), with performance varying by genre. Study 2 demonstrates a strong label-bias effect: songs labeled “Human-made” receive higher evaluations than identical songs labeled “AI-made,” regardless of true authorship. Study 3 tests whether removing vocals alters these evaluations and finds only modest effects, with authorship beliefs remaining dominant. Overall, results reveal limited discriminability, powerful framing effects, and genre-dependent patterns shaping responses to AI music. |
|---|