From Score to Prompt Rethinking Musical Composition in the Age of AI English translation of #47 for Non Japanese Readers

Hello everyone.

This time, I would like to consider the relationship between the prompt—an essential element in creating AI-generated music with tools such as Suno—and the musical score, which has long been used in traditional music-making.

This topic is also closely related to what I have previously discussed as the “Theory of AI Performance.”
Therefore, as indicated in the title, I would like to treat this article as an addendum and develop the discussion from that perspective.

⸻

A. What Is a Musical Score?

At the time when music first emerged, musical scores most likely did not exist.

To be honest, I am not sufficiently knowledgeable about exactly when scores were invented or how they gradually evolved into the form we know today.

However, it is almost certain that musical notation was originally devised for the purpose of recording and preserving music.
And even today, that original function remains the primary reason why scores continue to be used.

In modern times, various new technologies have appeared for recording and preserving music.
Recording is probably the most representative example of such technology, and somewhat more technically speaking, MIDI, invented in the latter half of the twentieth century, is another.

Today, numerous electronic musical instruments and devices have been developed, many of which include built-in functions for saving musical data.
For this reason, it is now possible to record and preserve music even without relying on traditional paper scores.

Nevertheless, scores are still required in many situations.

There are several reasons for this:
• Paper scores are extremely portable.
• They allow a large amount of musical information to be viewed at a glance.
• While newer technologies developed in the twentieth century have many advantages, they also have certain limitations that scores can compensate for.

Perhaps most importantly, when we study music we are taught not only music theory but also how to read and write musical notation.

In other words, people involved in music are generally accustomed to scores.

It would not be an exaggeration to say that music and musical notation are often treated as almost inseparable.

Of course, this relationship varies somewhat depending on how one engages with music and what kind of music one performs.
However, even today it is safe to say that music and notation remain deeply connected.

Most of the new technologies developed in the twentieth century for recording and preserving music are based on recordings, that is, technologies that store the sound itself, or they incorporate recording as one of their functions.

The intuitive convenience of this method cannot be obtained from a score, which always requires the process of reading and interpreting written symbols.

Indeed, in commercial music production environments where speed and efficiency are crucial, this approach is widely used.

However, most of these modern methods require electricity.

In other words, in places without electricity—or when batteries run out—such “modern weapons” become completely useless.
In such situations, one may have no choice but to rely entirely on traditional musical notation.

Even in other circumstances, traditional scores are still widely used.

For example:
• when capturing the earliest stage of a musical idea in the form of a sketch
• or when asking instrumentalists to perform a passage during the composition process in order to hear how it sounds

Thus, even in an age when technology for handling music—especially digital technology—has advanced dramatically, musical notation continues to exist in close relationship with music itself.

⸻

B. What Happens in AI Music?

What about AI-generated music?

Here we find that scores are nowhere to be seen.

In AI music creation, as mentioned above, the score has finally disappeared from view.

Because of this, some people argue that music created with AI should not be considered composition.

At first glance, focusing only on what we can see, that argument might appear plausible.

But is it really true?

In fact, this question turns out to be quite profound.
It ultimately connects to a much deeper question:

What is music?

I have previously discussed this issue in another article and concluded that music is something created by the human mind.
I have no intention of changing that view.

However, this time I would like to consider the matter from a slightly different perspective.

Specifically, I would like to focus on the idea that many processes are involved before a piece of music is completed, something I myself have come to recognize more clearly in recent years.

Let us therefore break down the process of music creation into several stages:
1. The emergence of a musical idea and its recording or preservation
2. The creation of a piece based on that idea (form, melody, harmony, orchestration, expression markings, etc.)
3. Performance
4. Recording
5. Editing

Now let us see how different systems handle these stages.

A musical score is primarily involved in stages 1 through 3.

What about MIDI?

In the case of sequencing software equipped with a high-function DAW and plugins, it is possible to handle almost all stages from 1 through 5.
However, it is not always clear whether every stage can achieve a quality suitable for direct use.

In particular, stages 1 and 2 allow intuitive sound to be preserved, but the amount of additional information that can be recorded is somewhat limited.

As for performance, the quality of digital samples available today has improved dramatically, making it increasingly possible to produce works that can be used as they are.
However, this field requires considerable expertise and experience.

⸻

C. When We Use AI

What happens when we use AI systems such as Suno?

In stages 1 and 2, we interact with music in a way that had never existed before.

Interestingly, we can also choose the degree of our involvement.

In other words, we can adopt anything from
• a minimal level of involvement, leaving almost everything to the AI
to
• a method in which we control the AI as much as possible and guide the creative process ourselves.

However, even when we try to intervene as much as possible, we cannot directly specify which exact notes should be used at which point.

In other words, we cannot write notes on a staff as we would in traditional notation.

This is perhaps the most decisive difference from earlier methods of music creation.

Here we are confronted with a fundamental question:

What exactly do we mean by “composition”?

Has composition already become a fixed set of actions throughout history?

Our traditional image of composition is probably something like this:

A composer spreads out manuscript paper and writes notes on the staff, sometimes checking sounds on a piano or guitar while gradually fixing the music that exists in the mind.

This image has likely remained largely unchanged for several centuries.

Yet there may be an illusion hidden here.

We might be assuming that only visible actions constitute composition.

But if the essence of composition lies in the activity of the human mind, then composition must already begin before the moment when notes are written on a staff.

⸻

D. Examples

Let us consider several cases:
• Traditional music in various parts of the world
• Improvisation, such as jazz improvisation
• Certain forms of experimental avant-garde music
• Moments of sudden inspiration

In all of these examples, music already exists in some form or is in the process of emerging.

Many traditional musical cultures complete works without ever using notation.
Jazz improvisation is perhaps the clearest example.

In avant-garde music or moments of inspiration, the music may exist only partially—or even merely as a faint presence.
Yet music is already there, and composition has begun even without notation.

⸻

E. The Role of Prompts

Here we return to AI.

A prompt can participate in the creative process from this very stage, sometimes even recording information that notation cannot easily handle.

For example, a prompt can describe:
• the overall atmosphere of the music
• harmonic tendencies
• influences from particular historical periods or composers

These aspects may eventually appear in the notes of a score, but they are often difficult to express explicitly in notation.

For this reason, writing prompts is clearly an important part of composition, and I believe it is incorrect to claim that AI-based music creation is not composition.

⸻

F. Score vs Prompt

Let us now focus on the difference between scores and prompts.

When composing with notation, music must almost always pass through performance before reaching the listener.

This is nearly universal, which is why composition and performance are generally treated as separate activities today.

The main exception is when a composer who is also a performer plays their own works.

A prompt, however, has the unique ability to include instructions for performance itself.

In that sense, prompts contain part of the function of a score while also encompassing the performance derived from it.

A carefully crafted prompt can sometimes express the final musical form more explicitly than notation—including the style of performance and even the environment in which the music is performed.

When we consider this, comparisons such as “writing prompts is not composition” or “notation is superior because it specifies notes directly” begin to lose their meaning.

Both are powerful tools for creating music, and they are not mutually exclusive.

Prompts can specify harmonic tendencies, and although they do not resemble staff notation, they can still record contrapuntal motion.

If a specific melody must be used, it can be uploaded as audio and then incorporated through prompts.

Furthermore, with sufficient experience, it is sometimes possible to stimulate deeper structures within the AI through prompt writing and draw out unexpected beauty.

The musical meaning inside AI does not correspond to words in a simple one-to-one relationship.

Even small differences in spelling, word order, or sentence structure can produce different musical results.

In this sense, prompts can produce an almost infinite range of variations depending on how they are written, and they contain far greater expressive possibilities than we might initially imagine.

We have only just acquired this technology of a new era, and I cannot help feeling recently that exploring its possibilities is an incredibly stimulating journey.

⸻

With that, I will conclude today’s discussion.

Thank you very much for reading.