By Abdul Malik
(Research Intern)
Abstract:
Generative artificial intelligence (AI) can now write poems, compose music, and produce images that ordinary people cannot reliably tell apart from human work. This raises a question that is no longer hypothetical as machines take on tasks long considered uniquely human, what happens to human creativity itself? This proposal frames the debate around a central tension visible in the recent empirical literature generative AI appears to raise the creativity of individuals while simultaneously reducing the diversity of what groups of people produce. The proposal sets out research questions, a synthesis of current evidence, and a mixed methods study design to investigate whether, and under what conditions, generative AI augments rather than erodes human creative capacity. The intended contribution is practical guidance for how creators, educators, and tool designers can use AI in ways that protect originality and human agency.
1. Introduction and Background:
Creativity the production of work that is both novel and useful has long been treated as a defining feature of human intelligence (Amabile, 2013; Boden, 2004). For most of history, the tools of creative work (the pen, the camera, the synthesizer) extended human ability without replacing the act of imagination itself. Generative AI breaks that pattern. Systems such as ChatGPT, DALL·E, Midjourney, and Stable Diffusion do not merely assist execution; they generate finished creative artifacts from a short prompt.
The shift happened quickly. As early as 2021, experiments showed that readers could not reliably distinguish AI generated poetry from human written poetry when a human curated the AI best output (Köbis & Mossink, 2021). By 2024, controlled studies reported that a large language model outperformed most human participants on standard psychological tests of divergent thinking the ability to generate many original ideas (Hubert et al., 2024). For a faculty once assumed to be ours alone, this is a profound development.
The public conversation tends to collapse into two opposing narratives. One is optimistic: AI is a democratizing creative partner that lifts everyone productivity and lowers the barrier to artistic expression. The other is pessimistic: AI will flood culture with derivative, homogenized content, hollow out creative professions, and weaken the very skills it claims to assist. The purpose of this research is to move past that binary by examining what the emerging evidence actually shows and where it disagrees.
2. Statement of the Problem and Research Gap:
A growing body of research has begun to measure AI’s effect on creative work, but the findings point in genuinely different directions, and that divergence is the problem this study addresses.
On one hand, studies of real creative output report large gains. An analysis of more than four million artworks found that text to image AI increased users creative productivity by roughly 25% and raised the perceived value of their work by about 50% over time (Zhou & Lee, 2024). On the other hand the same wave of research warns of a hidden cost. In a widely cited experiment, writers who received AI generated story ideas produced individually more creative stories but the stories produced across the group became more similar to one another, reducing collective diversity (Doshi & Hauser, 2024).
This is the central paradox that motivates the proposal: The generative AI may make each person more creative while making people collectively less varied. If true the consequences are not obvious. A more productive individual is a clear benefit a less diverse culture is a clear loss. The two effects operate at different levels the individual and the collective which is precisely why they are easy to conflate and hard to weigh against each other.
This is the central paradox that motivates the proposal:
First, most studies isolate a single domain (micro fiction, visual art, or laboratory divergent thinking tasks) and a single point in time so it is unclear whether the augmentation homogenization pattern generalizes across creative fields or persists with prolonged use.
Second, the mechanism is underexplored. We do not yet know whether the homogenization effect comes from the AI outputs themselves from how people choose to use them (passively accepting versus actively building on them) or from users loss of confidence in their own ideas.
Third, there is little work translating these findings into actionable guidance. If AI both helps and harms creativity the useful question is no longer “is AI good or bad for creativity?” but “under what conditions does AI augment human creativity rather than erode it?”
3. Research Questions and Objectives:
Central question: Under what conditions does generative AI augment human creativity and under what conditions does it erode it?
- How does collaborating with generative AI affect the novelty and usefulness of an individual creative output?
- How does the same collaboration affect the collective diversity of output across a population of creators?
- Does the mode of interaction using AI as a cocreator versus as a source to edit change these effects?
- How does AI use affect creators sense of ownership, authorship, and creative self confidence?
Objectives:
• To synthesize current empirical evidence on AI effect on individual and collective creativity.
• To design and conduct a controlled study testing how interaction mode shapes creative outcomes.
• To identify mechanisms behind any homogenization effect.
• To propose evidence based recommendations for creators, educators, and AI tool designers.
4. Significance of the Study
The findings matter for several groups. For creative professionals (writers, designers, musicians), the research informs whether and how to integrate AI without sacrificing the distinctiveness that defines their value. For educators, it speaks to a pressing concern whether students who lean on AI build or lose the underlying creative skills. For AI developers and policymakers, it offers design principles for instance, whether systems should be tuned to surface diverse rather than convergent suggestions. More broadly, the study contributes to a fundamental question in creativity research: whether creativity is a fixed human trait or a capacity that is reshaped by the tools we think with.
5. Literature Review:
5.1 Rethinking what “creativity” means
A first challenge is conceptual. Creativity is conventionally defined as work that is novel and appropriate to a task (Amabile, 2013). Generative AI strains this definition because it can produce outputs that humans judge as original even though they are statistical recombination of existing human work. Köbis and Mossink (2021) gave this an empirical edge in an incentivized Turing test design, participants could not reliably identify AI written poems when a human selected the best AI output, although they showed a mild aversion to poetry once it was labelled as machine made. The study suggests that perceived creativity depends not only on the artifact but on the human’s role in shaping and framing it a theme that recurs throughout the literature.
5.2 The augmentation thesis: AI lifts individual creativity
A substantial line of evidence supports the view that AI enhances individual creative performance. Zhou and Lee (2024), studying a large real world platform, found that adopting text to image AI raised both the quantity and the perceived value of users artwork over time, consistent with AI lowering the technical barrier so that users can focus on ideas rather than execution. In controlled psychological testing, Hubert et al. (2024) found that GPT-4 scored higher than most of 151 human participants across three established divergent thinking tasks, producing responses rated as more original and elaborate. Earlier conceptual work anticipated this trajectory Wu et al. (2021), after reviewing more than 1,600 application cases, proposed a “Human AI Cocreation” model arguing that the productive path is collaboration that plays to each party strengths rather than competition between human and machine.
Taken together, this strand supports what some researchers call a lift in “small c” everyday creativity. It is important to note an interpretive caveat, however scoring well on a laboratory divergence task is not the same as producing culturally significant, “big C” creative work, and high originality scores for AI may partly reflect its tendency to produce abstract rather than concrete responses (Hubert et al., 2024).
5.3 The erosion thesis: homogenization, dependency, and authorship
The counter-evidence does not deny individual gains but locates the cost elsewhere. Doshi and Hauser (2024) found that AI generated story ideas made individual writers stories more creative especially for less naturally creative writers yet the body of stories produced across participants became more alike. AI raised the floor but narrowed the range. This collective effect aligns with concerns raised in the interdisciplinary agenda set out by Epstein et al. (2023), who argue that generative AI will fundamentally reshape creative processes and warn that aesthetic and cultural biases embedded in models can constrain the diversity of what is produced and valued.
A second concern within this thesis is human agency and skill. If creators come to rely on AI for ideation, the worry is that creative confidence and capacity may atrophy, and that creators may not experience genuine ownership over AI shaped work an authorship ambiguity flagged across the human computer interaction literature.
5.4 Cocreation versus editing: the role of interaction mode
The most useful recent insight is that how people use AI may matter more than whether they use it. Interactive writing systems such as Wordcraft demonstrated that large language models can function as genuine creative collaborators, suggesting continuations, reframing ideas, and helping writers explore directions rather than simply delivering finished text (Yuan et al., 2022). This points toward a hypothesis that organizes the proposed study: the augmentation effect may dominate when humans treat AI as a partner they build on, while the erosion effect may dominate when humans treat AI as a source they passively accept and lightly edit.
Synthesis: The literature does not present a contradiction so much as a layered picture. AI reliably helps the individual its effect on collective diversity and on creators own capacities is contingent and, on current evidence, often negative. The research gap is therefore best framed not as a verdict but as a set of conditions domain, interaction mode, duration of use, and effect on self efficacy that the proposed study is designed to test.
6. Proposed Methodology:
This study will employ a mixed-methods research design, as the research questions involve both quantitative and qualitative dimensions. Quantitative methods will assess creativity, novelty, and diversity, while qualitative methods will explore ownership, authorship, and lived creative practice.
The study will begin with a systematic literature synthesis of empirical studies published from 2021 onward. The reviewed studies will be categorized by domain, interaction mode, and reported effects to identify where findings on creative augmentation and erosion are most consistently observed.
A controlled experiment, extending the design of Doshi and Hauser (2024), will examine how different forms of AI interaction influence creative outcomes. Participants will complete a short creative task, such as writing a micro-story or generating product ideas, under three conditions:
(a) Unaided, with no AI access.
(b) AI as source, where participants edit complete AI-generated drafts.
(c) AI as partner, where participants collaborate interactively with AI through iterative suggestions.
The experiment will measure three outcomes:
1. Individual creativity will be assessed through trained judges’ ratings of novelty and usefulness, supported by creativity measures such as the Alternative Uses Task and Divergent Association Task (Hubert et al., 2024).
2. Collective diversity will be measured through semantic similarity between outputs using text-embedding distance to quantify potential homogenization.
3. Ownership and self-efficacy will be examined through validated self-report scales measuring perceived authorship and creative confidence.
Data analysis will use between-group comparisons to examine whether different interaction modes alter the relationship between individual creative gains and collective diversity. Inter-rater reliability will be reported for all human evaluations.
Semi-structured interviews with practicing writers, designers, and musicians will further explore how AI is incorporated into professional creative workflows and how practitioners maintain distinctiveness in their work. These qualitative insights will complement the experimental findings by capturing experiences that cannot be measured quantitatively.
Ethical considerations will include informed consent, anonymization of data, transparency regarding AI involvement, and attention to intellectual property concerns related to AI training data.
7. Expected Outcomes:
The study is expected to:
(1) Confirm or qualify the individual augmentation finding across more than one domain
(2) Establish whether the homogenization effect is driven primarily by interaction mode rather than by AI exposure as such
(3) Clarify whether cocreation, as opposed to editing, preserves both diversity and creators sense of ownership. A plausible and testable outcome is that AI as partner retains most of the individual benefit while reducing the collective diversity cost.
8. Conclusion and Recommendations:
The evidence reviewed here resists a simple headline. Generative AI is neither purely a liberator of human creativity nor purely a threat to it. The most defensible reading of current research is that AI dependably enhances individual creative output while posing a real risk to the diversity of collective culture and, potentially, to creators own confidence and skill (Doshi & Hauser, 2024; Epstein et al., 2023; Zhou & Lee, 2024). The future of human creativity, then, is unlikely to be decided by the technology alone it will be shaped by how we choose to use it.
On that basis, the proposal advances four preliminary recommendations, each to be tested or refined through the study:
- Use AI as a collaborator, not a vending machine: Treat AI output as raw material to interrogate and build on, not as a finished answer to lightly edit the mode that appears most protective of originality and ownership.
- Design for divergence: Tool designers should consider features that deliberately surface varied, less predictable suggestions to counter homogenization, rather than optimizing only for a single “best” output.
- Protect creative self efficacy in education: Curricula should have students generate ideas before consulting AI, so that the technology supplements rather than substitutes for their own creative reasoning.
- Keep the human in the loop, visibly: Since perceived creativity and value depend heavily on the human’s role (Köbis & Mossink, 2021), maintaining clear human authorship and judgment is both an ethical and a practical safeguard.Human creativity is not being made obsolete it is being relocated away from execution and toward direction, judgment, and the framing of problems worth solving. The central task of this research is to understand that relocation well enough to guide it.
References:
1) Amabile, T. M. (2013). Componential theory of creativity. In E. H. Kessler (Ed.), Encyclopedia of Management Theory (pp. 134–139). Sage Publications. https://www.hbs.edu/ris/Publication%20Files/12-096.pdf
2) Boden, M. A. (2004). The creative mind: Myths and mechanisms (2nd ed.). Routledge. https://books.google.com/books/about/The_Creative_Mind.html?id=kFxymAEACAAJ
3) Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. https://doi.org/10.1126/sciadv.adn5290
4) Epstein, Z., Hertzmann, A., the Investigators of Human Creativity, Akten, M., Farid, H., Fjeld, J., Frank, M. R., Groh, M., Herman, L., Leach, N., Mahari, R., Pentland, A., Russakovsky, O., Schroeder, H., & Smith, A. (2023). Art and the science of generative AI. Science, 380(6650), 1110–1111. https://doi.org/10.1126/science.adh4451
5) Hubert, K. F., Awa, K. N., & Zabelina, D. L. (2024). The current state of artificial intelligence generative language models is more creative than humans on divergent thinking tasks. Scientific Reports, 14, 3440. https://doi.org/10.1038/s41598-024-53303-w
6) Köbis, N. C., & Mossink, L. D. (2021). Artificial intelligence versus Maya Angelou: Experimental evidence that people cannot differentiate AI generated from human-written poetry. Computers in Human Behavior, 114, 106553. https://doi.org/10.1016/j.chb.2020.106553
7) Wu, Z., Ji, D., Yu, K., Zeng, X., Wu, D., & Shidujaman, M. (2021). AI creativity and the human–AI cocreation model. In M. Kurosu (Ed.), Human Computer Interaction. Theory, Methods and Tools (HCII 2021), Lecture Notes in Computer Science (Vol. 12762, pp. 171–190). Springer. https://doi.org/10.1007/978-3-030-78462-1_13
8) Yuan, A., Coenen, A., Reif, E., & Ippolito, D. (2022). Wordcraft: Story writing with large language models. In Proceedings of the 27th International Conference on Intelligent User Interfaces (IUI ’22) (pp. 841–852). ACM. https://doi.org/10.1145/3490099.3511105
9) Zhou, E., & Lee, D. (2024). Generative artificial intelligence, human creativity, and art. PNAS Nexus, 3(3), pgae052. https://doi.org/10.1093/pnasnexus/pgae052