On certain nights in Pittsburgh the clubs still fill with people who come to hear musicians work out ideas in real time. A standard is stated, then stretched, answered, and returned in altered form. The music depends on influence that is open, reciprocal, and visible to anyone in the room. That practice sits at a useful distance from the way large AI systems currently absorb and redeploy existing work.
How Jazz Handles the Past
Standards as Shared Material
Jazz musicians have long treated a body of songs as common starting points. The melody and chord structure are known. The performance is valued for what the players do with that known material—phrasing, harmony, rhythm, and the conversation between instruments. Influence is expected. Direct imitation without transformation is recognized as such and generally valued less.
Attribution in the Room and on the Record
Even when the source material is decades old, the immediate creative act is credited to the people present. Liner notes, announcements from the bandstand, and the oral culture of the music keep track of who introduced a variation or a new composition. The system is imperfect, yet it maintains a working distinction between drawing on a tradition and claiming the tradition’s products as one’s own unacknowledged output.
How Large Models Handle the Past
Training as Large-Scale Ingestion
Most current generative systems are trained on extensive collections of existing text, images, music, and code. The process converts that material into statistical patterns that allow the model to produce new combinations. The original works are not quoted in the ordinary sense; they are dissolved into parameters. The scale of the ingestion is difficult to reconcile with older, smaller notions of fair use or transformative borrowing.
Output Without Provenance
When the system produces a passage of text or a passage of music, it does not reliably identify which earlier works most strongly shaped the result. The user receives fluent material that may stand in close relation to existing copyrighted work, yet the relation is not declared. The absence of provenance makes ordinary mechanisms of credit and compensation hard to apply.
The Legal and Practical Distinction
Influence as Transformation With Acknowledgment
In artistic and scholarly traditions, influence typically involves a human creator who has absorbed earlier work, altered it through labor and judgment, and offered the result under their own name. Copyright law has long attempted, with varying success, to protect the original expression while leaving room for later creators to build on ideas and styles. The human intermediary remains visible and accountable.
Extraction as Unacknowledged Scale
Extraction, in the sense relevant to current AI practice, describes the conversion of large volumes of protected work into model capability without continuous human transformation at the point of use and without a clear mechanism for returning value to the people who made the earlier work. The difference is not merely one of degree. It is a difference in whether the new system preserves a recognizable chain of creation and responsibility.
Why the Jazz Analogy Matters Here
Proximity to a Living Practice
Pittsburgh still supports working musicians who rely on both the openness of the tradition and the possibility of earning a living from their performances and compositions. When generative systems trained on recorded music begin to supply low-cost alternatives for background scores, practice tools, or even simulated improvisation, the economic pressure lands on people whose craft depends on scarcity of skilled human attention. The local scene makes the stakes concrete.
A Standard for Reciprocity
Jazz influence works because the people who draw on the tradition also sustain it—by playing, teaching, recording, and showing up. The loop is incomplete if a system draws on the recorded history of the music at industrial scale and returns no comparable support to the living practitioners. Copyright is one imperfect instrument for closing that loop. Social and contractual norms are others. None of them function well when the absorption is invisible.
Practical Lines Worth Drawing
Transparency of Training Sources
Systems that disclose the major categories and, where feasible, the specific collections used in training make it possible to have a concrete argument about consent, compensation, and opt-out. Systems that treat the training corpus as an undifferentiated and proprietary resource make the argument nearly impossible.
Separation of Generative Utility From Market Substitution
A tool that helps a musician explore harmonic options or a writer test sentence rhythms can be additive. A tool that supplies finished commercial material in the style of living artists, at a price no individual creator can match, functions as a substitute. Policy and practice can distinguish the two rather than treating every generative use as equivalent.
Compensation Mechanisms That Scale
If extraction at the current scale is to continue, some workable method of returning value to the people whose work supplied the training signal will be required. Voluntary licenses, collective licensing structures, or new statutory frameworks are all under discussion. The absence of any such method leaves the practice closer to extraction than to influence.
Jazz demonstrates that a culture can treat earlier work as raw material for new creation while still recognizing the people who produced both the earlier work and the new. Large AI systems currently demonstrate a different pattern: large-scale absorption followed by outputs that carry no reliable trace of their sources. The difference is not aesthetic preference. It is a difference in whether the chain of human labor remains visible and whether the economic and ethical obligations that travel with that chain can still be enforced.
The facts end here. The inference ends here. The judgment is yours.
No letters yet — pray write the first.