AI Mediation, Delegation, and Emergence

AI music systems divide action, judgment, authorship and control among people, models, interfaces and institutions. Whether a machine is creative in the abstract tells us little about who does what in a musical process.

Delegation and mediation

Session 12 describes contemporary system-based composition through delegation, direction and selection. Musicians work through systems that generate material, which moves some compositional agency away from direct execution (p. 5).

The course calls this a shift from execution to mediation. Skill can move from direct manual control toward guiding a system, specifying signs or conditions, selecting among outcomes and deciding when an output works (Session 12, p. 6).

The Assignment 4 brief develops the same point through possible roles such as editor, selector, semantic designer, co-navigator and system interactor. It asks which skills these systems foreground and which they displace (pp. 1–3).

Tool as actant

Session 12 defines an actant as a human or nonhuman part of a network that influences outcomes. A model or tool becomes an actant when it conditions possibilities, affords actions or co-determines decisions (p. 8).

Calling a tool an actant does not make it a person or settle authorship. It identifies causal participation, which analysis must trace to the people, data, interfaces, rules and institutions that give the tool its capacity to act.

Apparatus and dispositif

Session 12 treats model, interface, platform and user behavior as an apparatus that makes some musical actions thinkable, promptable and valuable (p. 52). Session 13 broadens this through the dispositif: technical systems, discourse, institutions, regulation, economics, cultural practice, power and user behavior work together to orient action (pp. 5–17).

The runtime engine is only one part of a generative-music service. Function labels, subscription access, data permissions, health claims, interface defaults, licensing and company control also shape musical practice.

Intersemiotic mediation

Intersemiotic translation moves between systems of meaning, such as text and sound. Prompt-based music systems turn language, genre tags and descriptive labels into compositional controls (Session 12, pp. 65–76).

Genre knowledge, naming, comparison, curation and symbolic judgment then become part of musicianship. The interface and model determine which words and musical relations the system can interpret.

[!note] Application inference A context-responsive music app may perform another form of intersemiotic mediation by translating time, weather, movement, heart rate, a chosen function or another nonmusical signal into musical parameters or state changes. Claims about musicianship depend on whether the artist designs those mappings or supplies only sound assets.

Emergence and bounded possibility

The course defines emergence as a property or pattern produced by interactions among simpler parts that cannot be predicted from any one part alone (Session 12, pp. 78–79).

This supports a distinction between a fixed work and a bounded field of outcomes. Materials, constraints, parameters and rules can remain stable while each sounding version differs.

Emergence still depends on design. A system can produce unpredictable local detail inside a tightly authored range. Analysis must separate fixed material, variable parameters, rules, inputs, random or probabilistic processes and platform control.

Listening as musicianship

Session 12 treats repeated listening as part of system-based creation. In its prompt-cartography examples, users compare outputs, discover tendencies and thresholds, trace drift, select promising regions and decide which variants deserve extension or revision (pp. 77–98).

This makes behavioral listening a compositional skill: the musician judges one result through the system’s behavior across repeated runs and edge cases.

From prompted production to continuous listening

Session 13 connects AI mediation to platform circulation. It discusses AI-generated mood playlists and suggests that streaming may move from static catalogs toward continuous sound generated or transformed around preference, context and listening habits (pp. 31–32, 52–56). Music then becomes a function: a dynamic flow responsive to a listener rather than a finite product (p. 55).

The course recap describes musicians as mediators and semiotic laborers who design conditions for music to emerge. They navigate symbolic systems, curate outputs and guide generative tools through interfaces, language and metadata (pp. 61–64).

[!note] Application inference Existing generative functional-music apps may connect platformed mood listening with possible on-demand AI streams. They already replace some fixed sequencing with runtime behavior, though this does not prove that they use open-ended text-to-audio generation or machine learning. Each mechanism needs its own evidence.

Critical risks

• Distributed agency can hide labor that still needs attribution.
• Interfaces make some genres, states and gestures easier to request than others.
• Endless generation may turn musical works into temporary service outputs.
• A platform governs adaptation by deciding which states count, which inputs matter and which responses are desirable.
• AI language can overstate novelty because procedural, generative, adaptive and automated music precede current text-to-audio systems.

Course sources

• Session 12, pp. 5–8, 49–55, 65–79, 94–98.
• Session 13, pp. 5–17, 31–32, 52–64.
• Assignment 4, pp. 1–7.