Introduction
Artificial Intelligence has been rapidly developing over the past decade, surpassing even the most ambitious predictions about what technology could achieve in terms of thinking, reasoning, and processing information. Due to its computational power, AI has made significant progress in music generation. Anyone can now input a text prompt describing what they want. Everything is customizable: the genre, tempo, and even the musician's style you want.
But despite these developments, AI creativity remains limited. Although the music it generates may sound convincing, it is not a product of the computer's creativity but rather a pattern-finding algorithm. To understand why an algorithm cannot mimic human creativity, we must first define creativity in music.
Creativity
Creativity in music involves more than just technical correctness. Although someone may know all about different scales, how to harmonize, etc., they still will not be able to create a meaningful piece of music. This is because human musicians often create songs based on emotions, memory and intent. Music is more of a channel for them to express themselves rather than just sounds that fit well together, and its these elements that give music personal and cultural meaning. To illustrate, take a live performance where a musician subtly alters tempo, dynamics, or phrasing in response to the audience or their own emotional state. These expressive choices are not written into the music itself, but emerge from human intention and feeling in the moment.
AI cannot do this, however. Artificial intelligence generates music based on probability. It is trained on a vast dataset of music to find patterns and stylistic choices in different genres of music. Then, using the patterns it recognized, AI predicts what sound should come next. It does not think of why the note should be there, or what effect it would create on the song. Instead, it simply recognized that it happened a lot in similarly-styled songs, causing them to add it to theirs.
Emotion and Culture
No matter how human-like AI might seem, it is important to keep in mind that in the end, artifical intelligence is just a computer; it lacks lived experience like humans. Since it has not experienced emotion or culture, it does not know how to create it. The closest it can come is to anaylse exisiting examples of emotion and try to replicate it. This only works to an extent, however, because it cannot grasp the meaning behind emotional expression or the social and cultural forces that give it significance.
Cultural context plays a crucial role in how music is created and interpreted. For instance, protest songs are shaped by specific social and political moments, using music as a tool for resistance and collective expression. Although AI may reproduce similar sounds, it cannot comprehend the urgency or shared meaning behind such music. Because authenticity in protest music is rooted in intention, listeners respond not just to the sound itself, but to the message and purpose behind it. When that intent is absent, the music may appear convincing on the surface but lacks the significance and genuine emotion of human-made expression.
Storytelling
Another crucial aspect of musical creativity is storytelling. The narrative a song conveys shapes the emotions a listener experiences, meaning that music often relies on a clear sense of progression to feel authentic and creative. However, storytelling is a limitation embedded in the way AI generates music. As discussed in “AI in Production,” many AI models generate music in short segments rather than with a complete narrative in mind. This often results in repetitive structures instead of a natural emotional flow, limiting a piece’s ability to develop meaning over time.
Although more recent models, such as Transformers, are better at maintaining broader structural coherence, they still fall short of human-level storytelling: the Transformer only has an average listener rating of 4.3, noticably lower than the 4.8 given to human-made music. Human composers intentionally shape musical progression to match the emotions they wish to convey. They introduce tension and release, develop themes over time, and make deliberate changes in dynamics, harmony, and tempo to guide listeners through an emotional journey.
Variation
A final factor which limits the amount of creativity AI generated music can have is its variation. AI finds common patterns in music by analysing a large dataset of exisitng music. Many music AI models, particularly in academic research, have been heavily trained and evaluated using datasets such as MAESTRO, which primarily consists of classical piano music. This can bias models toward certain musical structures and limit their ability to generalise across genres, which is their ability to apply observed patterns into music beyond what was seen during training.
Conclusion
AI music generation represents a significant technological leap, as creative production has long been considered a defining boundary between human and machine. However, that boundary is not completely broken down yet. Because of the way AI generation works, it still has trouble mimicking human expression in its works: lack of intentional progression hinders storytelling and narrow training data reduces variation. However, the most fundamental limitation of AI is its lack of lived experience, as this hinders AI's understanding of emotion and cultural context.
Despite these limitations, AI still remains a valuable tool for musicians to use. It still has potential to assist with the technical aspects of music creation, leaving the intention behind each choice to humans. To learn more about how AI and humans and collaborate to create music, read "Human and AI Creativity."