AI and Music Theory Discovery

Posted on by brucearnoldfoundation
Music & Technology

AI and Music Theory Discovery

How AI assistance accelerated the analysis and production of the SuperScales book — what the collaboration between human musical expertise and AI capability actually looked like.

Bruce ArnoldMusic & Technology

The SuperScales book, like the Sound Cells series, was produced with AI assistance. This is worth being precise about: AI did not discover the MMA universality property, did not compose the exercises, and did not make the musical judgments that shape the pedagogical approach. What AI did was accelerate specific tasks that are tractable for AI — mathematical verification, text generation, debugging, and organization — freeing human time and attention for the tasks that require musical expertise.

Understanding what AI actually contributes to music theory production — and what it cannot contribute — is itself a STEAM lesson about the nature of human-AI collaboration in creative and analytical work.

The Division of Labor

What AI Contributed
Mathematical verification of scale-chord pairings. Drafting book text for each chapter. Debugging LilyPond notation errors. Organizing large datasets (84 relationships, 864 drone specifications). Generating alternative explanations of the same concept. Proofreading for consistency across 84 chord-scale relationships. Writing the STEAM articles you are reading now.
What Remained Human
The discovery itself — asking the key center question that revealed MMA’s universality. Musical judgment about which exercises were pedagogically useful. Compositional decisions about the 2,521 études. Selection of which AI-drafted text captured the musical reality accurately. All decisions about the ear training curriculum’s structure and sequence. The 35 years of musical experience that made the discovery possible.

Specific AI Tasks in the SuperScales Project

AI Contribution — Task by Task
Mathematical checkingVerifying that the 84 chord-scale relationships were correctly classified — chord tones, tensions, and avoid notes correctly identified for each of the 12 MMA transpositions against each chord type. Checked hundreds of pairings for consistency with the Python algorithm’s results.
Text generationDrafting the chapter text for each of the 12 transpositions — explaining what chord each MMA transposition covers, which mode applies, and how to practice the relationship. Arnold reviewed and corrected each draft for musical accuracy.
LilyPond debuggingIdentifying the source of notation errors in the LilyPond source files — stem direction conflicts, accidental placement errors, spacing issues in multi-voice scores. AI pattern-matched the error messages and identified the relevant LilyPond syntax corrections.
Consistency checkingEnsuring that the 84 chord-scale relationships were described consistently across the book — same terminology for the same chord types, same mode names throughout, same interval naming conventions. Large-scale consistency is a strength of AI review.
Alternative explanationsWhen a concept was unclear in the first draft, generating multiple alternative explanations at different levels of technical detail — from the brief summary to the full mathematical derivation. Arnold selected the level appropriate for each section of the book.
STEAM articlesThe 20 SuperScales STEAM articles and the 20 hexatonic STEAM articles published on steam-education.org — drafted by AI based on Arnold’s system, reviewed and corrected for musical accuracy. This is the current article series.

What AI Cannot Do in Music Theory

The Limits of AI Contribution to Music Theory
✕Musical judgment: AI cannot determine whether an exercise sounds good, whether a progression is musically compelling, or whether a pedagogical sequence is optimally ordered. These require ears and experience.
✕Discovery: The MMA universality property was not found by AI processing scale data. It was found by a musician who had spent decades developing key center hearing and asked the right question from that vantage point. AI can verify a discovery; it cannot make one.
✕Pedagogical design: The sequence of exercises in the SuperScales book — which mode first, which key first, how many repetitions, what practice duration — is a pedagogical design decision that requires knowledge of how musicians learn. This is not something AI can determine from first principles.
✕Musical authority: When Arnold says that the Altered scale sounds right over a G7alt chord in key of C, this is a statement backed by decades of playing, teaching, and listening. When AI says the same thing, it is repeating a pattern from training data. The authority is different in kind.

A Model for Human-AI Collaboration in Creative Work

The SuperScales project offers a model for productive human-AI collaboration in creative and analytical work: the human contributes expertise, judgment, and discovery; the AI contributes speed, scale, and consistency on well-defined tasks. Neither replaces the other. The human cannot verify 792 scale-chord pairings in an afternoon. The AI cannot spend 35 years developing the ear training system that made the discovery possible.

The resulting work — the SuperScales book, the 864 drones, the 2,521 études, the STEAM articles — is neither purely human nor purely AI. It is a collaboration in which each contributor did what it does best. The musical knowledge is Arnold’s. The production scale is enabled by AI. The synthesis is the work.

Human-AI collaboration in analytical work

Division by tractability

The productive division of labor between human and AI is determined by tractability: tasks that are well-defined, large-scale, and pattern-based are AI-tractable; tasks that require judgment, creativity, or experiential knowledge are human-tractable. Neither type is inherently more valuable.

Verification vs. discovery

AI excels at verification — checking that a claimed result is correct across all cases. Discovery — finding the claim worth verifying — requires the kind of curiosity, pattern recognition, and knowledge that comes from sustained engagement with a domain. These are different cognitive activities.

Scale through AI

Tasks that were previously tractable only for institutions (large-scale content production, exhaustive mathematical verification, consistent terminology across hundreds of pages) become tractable for individuals with AI assistance. The production scale of the SuperScales book would have required a team without AI.

Authority and authenticity

The musical authority of the SuperScales system rests on Arnold’s expertise, not on AI’s contribution. AI-assisted production does not dilute this authority — it extends the reach of expertise into scales of production that were previously unavailable to a single practitioner.

The technology section of this series is now complete. The final two articles move into engineering: the acoustic engineering of the MMA drone system, and the complete journey from discovery to publication.

The role of AI in music theory and composition is an active research area. For background see Briot, Hadjeres, and Pachet, Deep Learning Techniques for Music Generation (2020). The SuperScales book, produced with AI assistance, is at muse-eek.com.

Leave a Reply

Your email address will not be published. Required fields are marked *