AI, Creativity and Authorship: Understanding the Human Role in Generative AI
Generative artificial intelligence has introduced a new kind of creative workflow. Instead of producing every visual element manually, artists and creators can describe an idea, generate multiple possibilities, evaluate the results, make changes, and continue the process until a desired outcome develops.
But this technological shift raises an important philosophical question:
If an AI system generates an image, what exactly is the human contribution to the resulting creative work?
This question is explored in the research article “When Generation Becomes Creation: Distinguishing Human Creativity from AI Derivation” by Choi B., published in Philosophy International Journal, Volume 9, Issue 3. The article introduces generative agency as a way of understanding when human involvement in AI-assisted generation becomes artistically significant.
Why AI Challenges Traditional Ideas of Creativity
Traditional discussions of creativity often focus on concepts such as originality, intention, transformation, control, and authorship.
Generative AI complicates these ideas because a human can initiate a process without specifying every detail that appears in the final work.
For example, an instruction may describe a particular subject, atmosphere, viewpoint, or style. The AI system may then determine many additional aspects of the image, including composition, lighting, texture, gesture, spatial relationships, and other details.
This means that simply causing an output to exist does not necessarily explain the source of all its aesthetically significant features.
The distinction between causing an output and creating a work becomes central.
The Prompt Is Not the Entire Creative Process
One common way of evaluating AI-assisted creativity is to focus on the prompt.
However, the paper argues that the relationship between prompt and output is more complicated than the number of words used in the instruction.
A detailed prompt can still leave important expressive decisions to the AI.
At the same time, a short instruction can have substantial significance if it establishes a rule, concept, or artistic direction that meaningfully organizes the work.
Therefore, prompt length is not a reliable measure of creative agency.
The paper instead emphasizes the complete trajectory:
Prompt → Generation → Judgment → Revision → Commitment → Integration
This process shows how an AI-generated possibility can become part of a developing artistic decision-making structure.
AI Generation Involves Contingency
Generative AI does not simply execute a fully specified human blueprint.
The system operates within constraints created by prompts, model architecture, training, and sampling conditions, while particular formal configurations remain partly undetermined.
The paper describes this relationship as constraint plus contingency.
This means that an AI system may produce something that the user did not specifically anticipate.
An unexpected feature might be:
- accepted,
- rejected,
- modified,
- tested,
- repeated,
- suppressed,
- or used to redirect the creative process.
The important question is what the human does with the unexpected result.
Unexpected Results Can Become Creative Material
Artists have long worked with uncertainty.
Chance, improvisation, material resistance, accidents, and procedural rules can all play roles in creative practice.
Generative AI introduces a different form of uncertainty because the system can provide highly articulated possibilities rather than merely producing an accidental mark or material event.
An AI system may produce an unexpected composition that appears almost finished.
At that point, simply recognizing that the image looks attractive is different from incorporating that result into an ongoing artistic problem.
The distinction becomes important when the unexpected result changes what happens next.
What Is Generative Agency?
The paper defines generative agency as the capacity to exercise aesthetically consequential judgment within an indeterminate generative process.
The objective is not to deny the contribution of AI.
Instead, it is to determine when human decisions become part of the explanation for why the resulting work has the organization and significance that it does.
This approach recognizes that AI-assisted creative production can be distributed.
The model, dataset, interface, prompt, user, and generated contingencies may all participate.
However, distributed production does not necessarily mean that every participant has the same creative responsibility.
The Five-Part Structure of Generative Agency
The paper develops this idea through five connected dimensions.
1. Aesthetic Intentionality
The human begins with an artistic orientation.
This does not require knowing exactly what the final work will look like.
An artist can begin with an atmosphere, conceptual problem, relationship, or tension and allow the process to develop from there.
Intentionality therefore provides direction without requiring complete prediction.
2. Consequential Intervention
Human intervention becomes particularly significant when it changes something aesthetically important.
A person may generate dozens of outputs, but the number of generations alone does not establish creative agency.
A single intervention can be more significant if it reorganizes the work's:
- spatial relationships,
- color structure,
- expressive qualities,
- conceptual relations,
- or overall composition.
The issue is the consequence of the decision, not simply the amount of activity.
3. Iterative Responsiveness
Creative development can occur when one generated result changes how the human approaches the next stage.
The process becomes:
Generation → Judgment → Redirection → Generation
rather than simply:
Generation → Generation → Generation.
According to the paper, responsiveness occurs when earlier results acquire significance for later decisions.
This creates a developing history within the work.
4. Selective Commitment
Selection alone is not necessarily equivalent to authorship.
A person can select one image from many alternatives because it looks attractive.
But the role of selection changes when the chosen feature becomes a constraint on future decisions.
For example, a creator may retain a particular visual relationship and then modify the rest of the work around it.
The choice has moved from preference to commitment.
5. Integrative Responsibility
The final stage concerns how generated possibilities become part of an organized artistic achievement.
The human does not need to claim that every element originated from their own mind.
Instead, responsibility concerns how generated possibilities were incorporated into the final work.
This creates a connection between attribution and answerability.
Causal, Directive and Constitutive Contributions
One of the useful distinctions in the paper is between three levels of human involvement.
Causal
The person contributes to the existence of the output.
For example, entering a prompt.
Directive
The person constrains the possibilities.
This may include selecting subjects, styles, references, parameters, or rejecting outputs.
Constitutive
Human judgments help explain why significant features of the final work have their particular form and role.
This third level is particularly important to the concept of generative agency.
Why Selection Alone May Not Be Enough
Imagine a person asks an AI system to generate 1,000 images and selects one image because it looks good.
That selection demonstrates a preference.
But does it explain the internal artistic organization of the selected image?
The paper argues that this question matters.
Selection becomes more formative when the selected feature influences later decisions.
For example:
- a particular composition is retained;
- later outputs are judged against it;
- other elements are modified around it;
- conflicting alternatives are rejected;
- the selected relationship becomes central to the developing work.
Here, selection becomes commitment.
AI Creativity and Copyright Should Be Kept Separate
The paper also emphasizes the distinction between creative attribution and copyright law.
Aesthetic attribution asks questions about artistic achievement and human contribution.
Copyright law addresses different questions concerning protectable human expression, ownership, infringement, and related legal doctrines.
Consequently, demonstrating creative agency does not automatically establish copyright protection.
Similarly, a legally permissible use does not necessarily establish that a human created every aspect of a work.
This distinction is particularly important as AI-assisted creative practices continue to develop.
Three Questions for Evaluating AI-Assisted Creative Work
The paper proposes three diagnostic questions.
Question 1
Would removing the relevant human decisions leave the significant aesthetic identity of the work largely unchanged?
Question 2
Do the human decisions explain why important relationships were preserved, rejected, transformed, or integrated?
Question 3
Did generated possibilities become commitments that influenced subsequent decisions, or were they simply accepted because they looked attractive?
These questions move the discussion beyond simplistic measurements such as:
- number of prompts,
- number of generations,
- amount of editing,
- amount of AI involvement,
- or amount of human labor.
What Does This Mean for the Future of AI Creativity?
Generative AI does not necessarily eliminate human creativity.
Instead, it changes where creative judgment may occur.
In traditional artistic production, a person might directly execute many formal decisions.
In generative AI workflows, some formal possibilities may originate from the system.
Human creativity can therefore become increasingly visible through:
- defining artistic problems,
- evaluating unexpected possibilities,
- making consequential interventions,
- developing new criteria,
- selecting meaningful relationships,
- transforming outputs,
- and integrating possibilities into a coherent work.
The paper's argument is not that AI contribution disappears.
Rather, the focus is on how generated possibilities become organized through human judgment.
Conclusion
The question “Who created an AI-generated image?” may be too simple to capture the complexity of human–AI creative production.
A more useful question is:
How did human judgment shape what the generated possibilities became?
The concept of generative agency provides a framework for examining this question.
It connects:
Intentionality → Intervention → Response → Commitment → Integration
When these dimensions converge sufficiently, the resulting work can be understood through a trajectory of human artistic decisions rather than merely as a selected product of an AI system.
Generative AI therefore does not make the distinction between generation and creation irrelevant.
Instead, it makes that distinction more precise.
