Source-linked AI summary
What Makes Creation Human? Authorship, Reasons, and Meaningful Human Control in Generative AI
Yuxi Cao
TL;DR
Generative AI expands productive capacity without necessarily expanding creative agency or authorship, exposing a gap between creative outputs and human creative control. The paper extends Meaningful Human Control into dynamic-reflexive tracking, arguing that authorship depends on whether evolving human reasons are reflectively taken up and effectively shape the work’s trajectory. It concludes that meaningful authorship depends on continuous capacity to evaluate, reject, and redirect creative changes rather than on the amount of manual production.
Problem
Generative AI separates creative capacity at the level of works from creative agency and authorship at the level of human creators.
Method
The paper reconstructs Meaningful Human Control as dynamic-reflexive tracking of reasons that form, change, and influence creation through interaction with AI.
Results
Human authorship depends on whether human judgments and reasons genuinely shape the work, not on manual effort, automation level, or final selection authority.
Takeaways & Limitations
Technical execution can be outsourced while meaningful authorship remains possible when creators retain reflective evaluation, rejection, ownership, and redirection of changes.
Takeaways & Limitations
The reflective self-questioning tool is a diagnostic heuristic rather than a strict authorship test, and requires comparison with actual creative trajectories and revision histories.
Abstract
from arXiv · showhide
Generative artificial intelligence (GenAI) significantly expands creators' productive capacity, but this does not necessarily entail a corresponding increase in creative agency or authorship. This paper distinguishes creativity at the level of the work from creative agency at the level of the creator, and argues that human authorship cannot be determined solely by manual intervention, degree of automation, the origin of an initial idea, or final selection authority. Rather, authorship depends on whether human judgment and reasons genuinely shape the development of the work. To articulate this requirement, the paper introduces Meaningful Human Control (MHC) into generative creation and identifies a limitation of its classical tracking condition. Creative reasons are not always fully specified prior to interaction with AI; they may emerge, change, or be abandoned as the creative process unfolds. The paper therefore proposes dynamic-reflexive tracking (DRT), which requires that a creator's evolving reasons undergo reflective uptake, exert genuine influence on the subsequent trajectory of creation, and remain capable of rejecting and redirecting the system's default direction. DRT consists of four conditions: diachronic reason formation, reflective uptake, trajectory efficacy, and contestability and redirection, together with a minimal tracing requirement. The paper argues that human authorship under generative AI depends not on how many steps a person personally performs, but on whether that person's reasons continuously, reflectively, and effectively shape what the work becomes.
1 From Creative Outputs to Human Authorship
Generative AI separates the creativity of an artefact from the presence of a human creative subject behind it. Human authorship therefore requires examining human agency and contribution beyond the work’s apparent creative features.
- AI outputs can satisfy traditional criteria for artistic works, while their authorship and value are also shaped by markets, platforms, and cultural institutions.
- Creativity includes originality and value in the artefact, alongside an agential capacity to produce contextually appropriate novelty rather than fortuitous outcomes.
- Human-like creative features, including emotional expression, do not by themselves establish a human creative subject behind an AI-assisted work.
- AI’s apparent directionality reflects derived intentionality from training data, system design, and feedback rather than autonomous purposes or comprehension.
- 1 From Creative Outputs to Human Authorship: Generative AI makes a work’s creativity and a person’s authorship increasingly separable judgments.This separation affects attribution at individual, collective, and institutional levels.
2 Beyond Productive Capacity: Individual, Collective, and Industrial Dimensions of Creative Empowerment
Generative AI expands productive capacity, but productive, agential, process, and distributive empowerment can diverge. The paper therefore evaluates empowerment across individual, collective, and industrial conditions rather than treating it as a single gain.
- Individual Level: Generative AI can expand production and access to creative activities while also offloading ideation, monitoring, integration, and revision from creators.Efficiency gains may reduce cognitive engagement and subjective ownership.
- Individual Level: Accurate prediction of past preferences is not equivalent to responsiveness to a creator’s current creative reasons.Personalisation may reproduce familiar outputs without improving the creator’s capacity to understand or intervene.
- Collective Level: Model capacities built from collective human production do not automatically become collective creative resources for human communities.
- Collective Level: Creative experiences formed in private AI sessions may disappear rather than enter domain knowledge systems through preservation, exchange, selection, and recognition.
- Industrial Level: Platform defaults, recommendation logics, templates, and market feedback can make content increasingly dependent on standardised styles and quantifiable signals.
- Industrial Level: Industrial adoption may reduce creative production costs while compressing professional creative employment and bargaining power.
- Industrial Level: Generative AI’s empowerment is multidimensional: productive capacity may increase while judgmental autonomy, process control, or value capture declines.
3 From Factual Participation to Normative Attribution: Authorship as a Position of Accountability
Authorship in generative creation is a normative position of accountability, not a direct record of who performed the most operations. The relevant question is whether human judgments shape the work at a higher-order level.
- Authorship cannot be determined by prompt volume, automation level, manual operations, or final selection alone.The key distinction is between enabling a work’s existence and exercising creative judgments that shape it.
- Creative contribution increasingly centers on evaluation, selection, rejection, integration, and redirection as AI makes options easier to generate.
- As external assistance increases, people attribute less authorship, creatorship, and responsibility to the human creator.
- Human authorship is more attributable when creators determine evaluative frameworks, retention standards, and how outputs are integrated.
- Human authorship anchors responsibility because creators must defend, revise, retract, and answer for the work even when production is distributed across humans, models, data, and platforms.
4 From Human Involvement to Meaningful Human Control
Meaningful Human Control evaluates whether human reasons substantively shape generative creation rather than merely registering operational intervention. In creative contexts, this requires preserving the formation, revision, and effective redirection of reasons over time.
- Operational intervention, such as prompting, regenerating, or vetoing, does not establish that a system has normatively tracked a creator’s reasons.
- MHC is context-dependent, so its requirements must be specified according to the normative purpose of sustaining creative agency and attribution.
- For generative creation, MHC aims to sustain creative agency and authorship by assessing whether human reasons substantively shape the process.
- A system may causally follow inputs while its defaults and platform aesthetics narrow the creator’s preferences rather than track expressive reasons.
- Preference-locking loops repeat past preferences, whereas meaningful creative development requires new possibilities, surprise or conflict, re-evaluation, and new directions.
- Creative control requires structural shaping: creators must be able to propose, explain, maintain, revise, and overturn reasons that influence the work’s direction.
- Generative AI can expand non-professionals’ agency when it supplies technical execution for judgments they already possess; the concern is not assistance itself but weakened creative control.
5 Reconstructing the Tracking Condition
Classical tracking is insufficient for generative creation because creative reasons may emerge, change, or be abandoned during interaction. The paper therefore proposes dynamic-reflexive tracking, which evaluates whether evolving reasons are reflectively taken up and actually shape the work’s trajectory.
- 5 Reconstructing the Tracking Condition: Dynamic-reflexive tracking replaces fixed-intention tracking with attention to how creative reasons form, change, and influence development over time.The initial idea or first prompt does not by itself establish authorship.
- 5 Reconstructing the Tracking Condition: The creative process is cyclical: AI outputs become objects of evaluation that can prompt revised reasons and influence subsequent generation.The final work is not treated as the simple realization of a fixed initial intention.
- 5 Reconstructing the Tracking Condition: DRT requires four conditions: diachronic reason formation, reflective uptake, trajectory efficacy, and contestability and redirection.Together, these conditions require reasons to develop through interaction, receive reflective uptake, alter subsequent choices, and enable new directions.
- 5 Reconstructing the Tracking Condition: Creative control concerns whether reasons continuously participate in and actually shape the work’s development, not merely whether users perform inputs, clicks, or selections.The account retains a minimal tracing requirement for an identifiable human who understands the system’s role and can bear answerability for key choices.
6 Operationalisation: A Set of Self-Interrogative Diagnostic Tools
The paper operationalises dynamic-reflexive tracking through reflective self-interrogation rather than a strict authorship test. This diagnostic framework focuses on whether creators understand, evaluate, reject, and redirect AI outputs while recognizing that individual control does not settle broader institutional questions.
- 6 Operationalisation: A Set of Self-Interrogative Diagnostic Tools: The proposed self-interrogation tool examines whether creators’ reasons continue to shape the work, rather than measuring authorship through clicks or AI-use proportions.The tool is intended as an evaluative language for continuity of reasons in creative control.
- 6 Operationalisation: A Set of Self-Interrogative Diagnostic Tools: Creators need not understand model internals, but they must interpret outputs, evaluate them, reject them, and redirect the system.The paper notes that interface and explanatory tools for these users remain comparatively underdeveloped.
- 6 Operationalisation: A Set of Self-Interrogative Diagnostic Tools: The framework borrows due-diligence questioning to probe why a course was chosen, what evidence would change it, and when it would be abandoned or redirected.This structure is presented as resembling the reflective and counterfactual questions central to DRT.
- 6 Operationalisation: A Set of Self-Interrogative Diagnostic Tools: The framework is not a legal copyright judgment, a complete theory of AI moral agency, or a claim that all authorship requires the same explicit reflection.Its limitations include jurisdiction-sensitive legal authorship, unresolved metaphysical questions, and variation across media and collaborative structures.
- 6 Operationalisation: A Set of Self-Interrogative Diagnostic Tools: DRT is a normative reconstruction rather than an empirical measurement scale, and the individual-level account does not determine who sets the creative ecosystem’s goals and values.The paper identifies future operational indicators including version history, rejection trajectories, counterfactual redirection, and user understanding.
7 Main Objections
The objections clarify that creative agency cannot be inferred from explicit verbal reasons, independent reason origins, automation levels, or post hoc endorsement alone. Instead, agency depends on reflective incorporation, effective influence on the work, and continued capacity to contest and redirect system influence.
- 7 Main Objections: Reflective uptake can include tacit judgments such as sensing that “something is wrong,” provided those judgments consistently guide rejection, revision, or redirection.Reason-giving supports reflective control but is not its only form.
- 7 Main Objections: External influences do not by themselves diminish agency; the relevant issue is whether creators can reflectively absorb, question, revise, or reject them.AI can introduce new directions or change earlier judgments without necessarily reducing agency.
- 7 Main Objections: Automation level is not an inverse measure of creative agency because technical execution can be delegated while creative judgment remains human-controlled.Many prompt revisions and selections may still amount only to reaction if they do not involve reflection, evaluation, and redirection.
- 7 Main Objections: Creative credit can remain distributed across multiple agents, but attributing a share to a person requires explaining how that person’s judgments and reasons shaped the work.DRT does not require every work to have one independent author.
- 7 Main Objections: Post hoc rationalisation is insufficient because endorsing a successful AI result afterward does not show that the claimed reason shaped the creative process.A reason must have played a role during creation, not merely be supplied after the outcome.
- 7 Main Objections: Personalisation is not inherently agency-reducing, but preference-locking loops can displace present reason formation with repeated reproduction of past preferences.The relevant safeguard is continued room for surprise, contestability, and redirection.
8 Theoretical and Methodological Limitations
The paper limits creative MHC to a functional control relation in which system behavior responds effectively to human reasons, without requiring AI to possess human-like understanding. It also restricts the framework’s scope to agential empowerment and process autonomy and treats its diagnostic and individual-level claims as non-exhaustive.
- 8 Theoretical and Methodological Limitations: Creative tracking requires effective reason-responsiveness from the system, not human-like understanding or intrinsic intentionality.Tracking concerns the functional control relation between human reasons and system behavior.
- 8 Theoretical and Methodological Limitations: Creative MHC primarily addresses agential empowerment and process autonomy, not productive empowerment or distributive justice.The paper warns against expanding MHC into a general theory of every social and institutional problem raised by generative AI.
- 8 Theoretical and Methodological Limitations: The reflective self-questioning tool is a diagnostic heuristic rather than a strict authorship test.Verbal fluency does not prove genuine control, while tacit judgments may still support stable and effective creative agency.
- 8 Theoretical and Methodological Limitations: DRT is currently a normative reconstruction and does not directly measure degrees of creative agency in real-world practice.The paper identifies this as a methodological limitation of the framework.
- 8 Theoretical and Methodological Limitations: The individual-level account asks whose reasons shape the work but does not explain who determines the goals and values of the wider creative ecosystem.Model developers, data contributors, platform designers, and cultural institutions fall outside its comprehensive treatment.
9 Conclusion
The paper separates creative properties of AI-generated works from human creative agency and argues that authorship depends on whether human reasons genuinely shape the work’s development. It proposes dynamic-reflexive tracking as a normative standard for assessing this control, while acknowledging that it is not a complete theory of authorship.
- 9 Conclusion: Creative work-level properties and human creative subjectivity are distinct judgments, since novelty, aesthetic effect, and emotional expression do not establish human authorship.
- 9 Conclusion: Dynamic-reflexive tracking extends meaningful human control by tracking reasons as they form, undergo evaluation, change through interaction, and continue shaping the work.
- 9 Conclusion: DRT requires diachronic reason formation, reflective uptake, trajectory efficacy, and the ability to contest system defaults and redirect creation.Reason changes must make an actual difference to the work’s trajectory rather than merely rationalize it retrospectively.
- 9 Conclusion: DRT offers a normative standard for creative agency and human authorship, but it does not by itself provide a complete account of authorship.
- 9 Conclusion: Human authorship depends on whether a creator’s reasons genuinely alter what the work becomes, not on how many production stages the person performs.Technical execution may be outsourced while human judgment remains decisive over the work’s development.