近年來生成式 AI 技術迅速成長，Suno、Udio 等平台只需短短一分鐘，就能生成高水準的音樂。然而，這些 AI 模型究竟是如何「創作」音樂的？它們真的理解音樂中的和聲、旋律、節奏、配器與混音等核心概念嗎？ 本場演講將探討生成式 AI 的「音樂大腦」， AI 到底學習並掌握了哪些音樂知識，以及我們該利用這些理解，實現更精準、可控的音樂生成，並開發出更全面、具備音樂知識的AI模型。

Driven by the rapid growth of generative AI, platforms like Suno and Udio can now produce high quality music in just one minute. However, how do these models actually "compose" music? Do they truly understand core concepts such as harmony, melody, rhythm, instrumentation, and mixing? This talk explores the "musical brain" of generative AI, examining what musical knowledge these models acquire and master. Furthermore, we will discuss how to leverage these insights to achieve more precise, controllable music generation, and develop AI models equipped with comprehensive musical knowledge.

創造力究竟是一個客觀事實，還是一種被塑造的動態觀點？你以為的「有創意」，是真的有創意，還是只是你的背景偏見？ 本研究基於社會建構論，透過 Reddit 輿情分析、半結構化訪談與雙盲聽力實驗，提出 AI 音樂創意的評估架構。研究發現，「風格創新」是驅動創意判斷的主因，且專業音樂人與一般聽眾存在顯著的認知落差（Expertise Gap）。本研究證實，AI 時代的音樂創意並非機器的固定屬性，而是由人類感知持續重塑的動態過程。

Is creativity an objective fact, or a dynamically constructed perspective? Is what you consider "creative" truly creative, or merely a reflection of your own background bias? Grounded in social constructivism, this study develops an evaluation framework for AI musical creativity using Reddit discourse analysis, semi-structured interviews, and a blind listening test. The findings reveal that "Style Innovation" serves as the primary driver of perceived creativity, while uncovering a significant "Expertise Gap" between professional musicians and casual listeners. Ultimately, this research demonstrates that musical creativity in the AI era is not a fixed algorithmic attribute, but a dynamic process continuously reshaped by human perception.