Generative artificial intelligence, human creativity, and art
Open accessAbstract
Recent artificial intelligence (AI) tools have demonstrated the ability to produce outputs traditionally considered creative. One such system is text-to-image generative AI (e.g. Midjourney, Stable Diffusion, DALL-E), which automates humans’ artistic execution to generate digital artworks. Utilizing a dataset of over 4 million artworks from more than 50,000 unique users, our research shows that over time, text-to-image AI significantly enhances human creative productivity by 25% and increases the value as measured by the likelihood of receiving a favorite per view by 50%. While peak artwork Content Novelty, defined as focal subject matter and relations, increases over time, average Content Novelty declines, suggesting an expanding but inefficient idea space. Additionally, there is a consistent reduction in both peak and average Visual Novelty, captured by pixel-level stylistic elements. Importantly, AI-assisted artists who can successfully explore more novel ideas, regardless of their prior originality, may produce artworks that their peers evaluate more favorably. Lastly, AI adoption decreased value capture (favorites earned) concentration among adopters. The results suggest that ideation and filtering are likely necessary skills in the text-to-image process, thus giving rise to “generative synesthesia”—the harmonious blending of human exploration and AI exploitation to discover new creative workflows.
AI snapshot
Written from the paper’s full text.
TL;DR
Using over 4 million artworks from more than 50,000 users on an art-sharing platform, this paper finds that adopting text-to-image generative AI increases creative productivity and peer-evaluated value while average content and visual novelty decline and peak content novelty rises. It argues that human ideation and filtering—“generative synesthesia”—are key to benefiting from AI-assisted art.
Key findings
- Adoption increased creative productivity: the abstract reports a 25% increase, while the Results report a 50% increase in the adoption month that doubles the following month, translating to about 7 additional artifacts in the adoption month and 15 in the following month for the average user.
- Artwork value, measured as favorites per view after 2 weeks, rose 50% by the sixth month, from a preadoption average of 2% to a steady 3% rate of earning a favorite per view; the initial upward trend was nonsignificant, with significant differences from nonadopters after 3 months.
- Average Content Novelty declined over time among adopters, but maximum Content Novelty increased and was marginally statistically significant within the first several months; both average and maximum Visual Novelty declined.
- Artists who successfully explored more novel ideas, regardless of prior originality, may produce artworks peers evaluate more favorably; Content Novelty was more predictive of value gains than Visual Novelty. AI adoption was associated with less concentrated favorites: treated Gini 0.794 vs never-treated 0.807 (D = −0.0128, p = 0.0673) and not-yet-treated 0.824 (D = −0.0298, p = 0.0026).
Methods
- Design: event-study difference-in-differences to estimate causal impact of AI adoption on productivity, value, and novelty; two-way fixed effects for heterogeneity by prior novelty quartiles; generalized random forests with SHAP for feature importance; Gini coefficients and permutation tests with 10,000 iterations for platform-level value concentration; generalized synthetic control method and falsification tests for robustness.
- Data: over 4 million artworks from more than 50,000 unique users; over 53,000 artists and 5,800 known AI adopters; novelty analysis uses adopters and a random sample of 10,000 control users.
- Novelty measurement: Content Novelty via BLIP-2 image captions plus BERT embeddings; Visual Novelty via DINOv2 embeddings; pairwise cosine distance to prior baseline artworks, with the baseline updated monthly.
Limitations
- Stated: the authors do not directly observe users’ creative process, so the BVSR/generative synesthesia discussion is speculative; selection into AI adoption is a concern, addressed with generalized synthetic control and falsification tests.
- Evident: AI adoption is identified by rule-based keyword/tag matching and AI art community labels, which may misclassify users or artworks; novelty and value are platform-specific proxies (embeddings and favorites per view after 2 weeks).
- Evident: causal estimates rely on difference-in-differences assumptions, though robustness checks are reported; the sample comes from one art-sharing platform, limiting generalizability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 2 cited sources checked · 0 supported · 0 flagged
- Findingown data
Users who were proficient in creating exceedingly novel visual features before adopting generative AI may garner the most Value gains from successfully introducing more novel ideas.
“Next, Fig. 2b suggests that users who were proficient in creating exceedingly novel visual features before adopting generative AI may garner the most Value gains from successfully introducing more novel ideas.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Findingown data
More novel ideas are linked to improved Value capture.
“In general, more novel ideas are linked to improved Value capture.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Findingown data
AI-assisted artists who successfully explore more novel ideas, regardless of their prior originality, may produce artworks that their peers evaluate more favorably.
“AI-assisted artists who can successfully explore more novel ideas, regardless of their prior originality, may produce artworks that their peers evaluate more favorably.”
- Also stated in: § Role of human creativity in AI-assisted value capture · hedged
“Figure 2a presents correlational evidence that users, regardless of their proficiency in generating novel ideas, might be able to realize significant gains in Value if they can successfully produce more novel content with generative AI.”
- Also stated in: § Discussion · hedged
“humans assisted by generative AI who can successfully explore more novel ideas may be able to push the creative frontier, produce meaningful content, and be evaluated favorably by their peers.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Also stated in: § Role of human creativity in AI-assisted value capture · hedged
- Findingown data
Less proficient users can also experience weak Value gains from introducing more novel ideas.
“While marginally significant, less proficient users can also experience weak Value gains.”
Quote in the paperNot a substantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Findingown data
Maximum Content Novelty increases over time among AI adopters, marginally statistically significantly within the first several months after adoption.
“peak artwork Content Novelty, defined as focal subject matter and relations, increases over time”
- Also stated in: § Content Novelty · plain
“Figure 1d, however, reveals that maximum Content Novelty is increasing and marginally statistically significantly within the first several months after adoption.”
- Also stated in: § Discussion · plain
“peak Content Novelty increases over time”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Also stated in: § Content Novelty · plain
- Findingown data
Adoption of text-to-image generative AI significantly increases artwork value, measured as favorites per view, by 50% by the sixth month.
“increases the value as measured by the likelihood of receiving a favorite per view by 50%”
- Also stated in: § Creative value · plain
“This translates to a 50% increase in artwork favorability by the sixth month, jumping from the preadoption average of 2% to a steady 3% rate of earning a favorite per view.”
- Also stated in: § Discussion · plain
“evaluated 50% more favorably by their peers over time”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Also stated in: § Creative value · plain
- Findingown data
Users who benefit from expressing more novel ideas may face penalties for producing more divergent visuals.
“However, those same users who benefit from expressing more novel ideas may also face penalties for producing more divergent visuals.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Findingown data
Adoption of generative AI leads to a less concentrated distribution of favorites earned among adopters.
“AI adoption decreased value capture (favorites earned) concentration among adopters.”
- Also stated in: § Platform-level value capture · plain
“favorites captured among AI adopters are more evenly distributed compared to both never-treated and not-yet-treated control units.”
- Also stated in: § Discussion · hedged
“generative AI appears to promote a more even distribution of platform-level favorites among adopters”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Also stated in: § Platform-level value capture · plain
- Findingown data
The differences in Gini coefficients are statistically significant between never-treated and not-yet-treated groups versus the treated group.
“The results from the permutation tests in Table 1, where column D shows the difference between the treated coefficient and the control group coefficients, show that the differences in coefficients are statistically significant between never-treated and not-yet-treated groups vs. the treated group.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Findingreasoning
Users capable of producing the most novel visual features may face penalties for pushing the boundaries of pixel-level aesthetics with generative AI.
“Conversely, users capable of producing the most novel visual features may face penalties for pushing the boundaries of pixel-level aesthetics with generative AI.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Findingown data
Adoption of text-to-image generative AI significantly increases artists' creative productivity.
“text-to-image AI significantly enhances human creative productivity by 25%”
- Also stated in: § Creative productivity · plain
“Figure 1a reveals that upon adoption, artists experience a 50% increase in productivity on average, which then doubles in the subsequent month.”
- Also stated in: § Discussion · hedged
“text-to-image generative AI can help individuals produce nearly double the volume of creative artifacts that are also evaluated 50% more favorably by their peers over time.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Also stated in: § Creative productivity · plain
- Findingown data
Maximum Visual Novelty decreases over time among AI adopters compared to nonadopters.
“consistent reduction in both peak and average Visual Novelty”
- Also stated in: § Visual Novelty · plain
“The same result holds for the maximum Visual Novelty seen in Fig. 1f.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Also stated in: § Visual Novelty · plain
- Findingown data
Average Content Novelty decreases over time among AI adopters compared to nonadopters.
“average Content Novelty declines”
- Also stated in: § Content Novelty · plain
“Figure 1c shows that average Content Novelty decreases over time among adopters, meaning that the focal objects and themes within new artworks produced by AI adopters are becoming progressively more alike over time when compared to control units.”
- Also stated in: § Discussion · plain
“average Content and Visual Novelty diminish”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Also stated in: § Content Novelty · plain
- Interpretationown data
Widespread adoption of generative AI in creative fields could lead to a long-run equilibrium where many artifacts converge to the same types of content or visual features.
“Our results hint that the widespread adoption of generative AI technologies in creative fields could lead to a long-run equilibrium where in aggregate, many artifacts converge to the same types of content or visual features.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Interpretationreasoning
Creative domains may be inundated with generic content as exploration of the creative space diminishes.
“Creative domains may be inundated with generic content as exploration of the creative space diminishes.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Interpretationreasoning
The gains from exploring creative ideas with AI outweigh the losses from visual divergence.
“Despite penalties for pushing visual boundaries, the gains from exploring creative ideas with AI outweigh the losses from visual divergence.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Interpretationreasoning
AI may be pushing artists toward visual homogeneity.
“In effect, AI may be pushing artists toward visual homogeneity.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Interpretationown data
Content is king in the text-to-image creative paradigm.
“In summary, these findings illustrate that content is king in the text-to-image creative paradigm.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Interpretationreasoning
Ideation and filtering are likely necessary skills in the text-to-image process.
“The results suggest that ideation and filtering are likely necessary skills in the text-to-image process, thus giving rise to “generative synesthesia”—the harmonious blending of human exploration and AI exploitation to discover new creative workflows.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Interpretationreasoning
Artists who successfully explore novel ideas and filter model outputs for coherence benefit the most from AI tools.
“The artists who successfully explore novel ideas and filter model outputs for coherence benefit the most from AI tools, underscoring the pivotal role of human ideation and artistic filtering in determining an artist’s success with generative AI tools.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Interpretationcitation
AI systems trained on outdated knowledge banks run the risk of perpetuating the generation of generic content at a mass scale in a self-reinforcing cycle.
“Without establishing new frontiers for creative exploration, AI systems trained on outdated knowledge banks run the risk of perpetuating the generation of generic content at a mass scale in a self-reinforcing cycle (17).”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [17] Burtch, 2023
Not checked in the free run.
- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Study design and conductown data
Unobservable characteristics that may correlate with users' productivity or interest shocks and selection into treatment are not driving the estimated effects.
“we also provide evidence that unobservable characteristics that may correlate with users’ productivity or “interest” shocks and selection into treatment are not driving the estimated effects by performing a series of falsification tests.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Interpretationown data
Content Novelty plays a more significant role in predicting posttreatment variations in Value gains compared to Visual Novelty.
“This suggests that Content Novelty plays a more significant role in predicting posttreatment variations in Value gains compared to Visual Novelty.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Interpretationown data
Generative AI may lead to a broader allocation of favorites earned, particularly among control units who eventually become adopters.
“This suggests that generative AI may lead to a broader allocation of favorites earned (value capture from peer feedback), particularly among control units who eventually become adopters.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- May combine two claims. 4
- The quote only partly states this claim. 2
- Classified as probably not a substantial claim. 1
May combine two claims.
Findingown dataContent Novelty significantly increases model performance within several months of adoption, whereas Visual Novelty remains marginally impactful until the last observation period.
“Figure 3 offers correlational evidence that Content Novelty significantly increases model performance within several months of adoption, whereas Visual Novelty remains marginally impactful until the last observation period.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
The quote only partly states this claim.
Findingown dataUsers in the lowest quartile of content creators may experience marginally significant gains in Value if they produce more novel content with generative AI.
“The lowest quartile of content creators may also experience marginally significant gains.”
Quote in the paperNot a substantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
The quote only partly states this claim.
Findingown dataAverage Visual Novelty decreases over time among AI adopters compared to nonadopters.
“consistent reduction in both peak and average Visual Novelty”
- Also stated in: § Discussion · plain
“average Content and Visual Novelty diminish”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
- Also stated in: § Discussion · plain
May combine two claims.
Interpretationown dataGenerative AI appears to promote a more even distribution of platform-level favorites among adopters, signaling a potential step toward an increasingly democratized, inclusive creative domain for artists empowered by AI tools.
“generative AI appears to promote a more even distribution of platform-level favorites among adopters, signaling a potential step toward an increasingly democratized, inclusive creative domain for artists empowered by AI tools.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
May combine two claims.
Interpretationown dataThe universe of creative possibilities is expanding but with some inefficiencies.
“This implies that the universe of creative possibilities is expanding but with some inefficiencies.”
Quote in the paperSubstantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
- Rests on an assumption: The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
Probably not a substantial claim: may not pass S2 (checkable).
InterpretationreasoningHumans' ideation proficiency and a refined artistic filter rather than pure mechanical skill may become the focal skills required in a future of human-AI cocreative process.
“our findings emphasize that humans’ ideation proficiency and a refined artistic filter rather than pure mechanical skill may become the focal skills required in a future of human–AI cocreative process as generative AI becomes more mainstream in creative endeavors.”
Quote in the paperNot a substantial claim- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
May combine two claims.
Study design and conductown dataThe results remain consistent when utilizing the generalized synthetic control method and in the presence of substantial measurement error.
“Our results remain consistent even when utilizing GSCM and in the presence of substantial measurement error.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [12] Xu, 2017, Generalized synthetic control method: causal inference with interactive fixed effects models, Polit Anal source
Not checked in the free run.
- Rests on an assumption: Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
- Rests on an assumption: The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
The paper relies on this without support.
Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.
“Platform-level policy commonly suggests that users disclose their use of AI assistance in the form of tags associated with their artworks.”
The paper relies on this without support.
The rule-based classification scheme accurately identifies AI-generated artworks and adoption timing.
“Thus, we employ a rule-based classification scheme.”
“We use simple keyword matching (AI-generated, Stable Diffusion, Midjourney, DALL-E, etc.) for each post to identify for which artworks a user employs AI tools.”
The paper relies on this without support.
The baseline set of artworks published before 2022 April 1 consists of non-AI-generated works.
“take all artifacts published before 2022 April 1, as the baseline set of artworks. We use this cutoff because nearly all adoption occurs after May 2022, so all artifacts in future periods are compared to non-AI-generated works in the baseline period”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Broader concerns exist about AI-generated media and the risk of fabricating emotion or persona.
“This resonates with broader concerns about AI-generated media and the risk of fabricating emotion or persona [18,81].”
From Exploring the Collaborative Co-Creation Process with AI: A Case Study in Novice Music Production (2025) · cites this paper as [81]
The Introduction describes concerns about AI-generated art, including its originality and its potential to replace human creativity. The Discussion also mentions disinformation as a potential risk. Neither section addresses fabricating emotion or persona, so the source supports the broader concern but not the specific risk named here. (LLM, conf=0.86)
Experimental — accuracy not yet measuredGenAI can support various human tasks such as writing, software development, composing lyrics, and academic research, often at a performance similar to that of humans.
“GenAI can thus support various human tasks such as writing, software development, composing lyrics, and academic research, often at a performance similar to that of humans [6, 11, 26, 28, 50, 72].”
From Generative AI and Creativity: A Systematic Literature Review and Meta-Analysis (2025) · cites this paper as [72]
The Discussion (page unavailable) says GenAI assistance can increase productivity in coding and written assignments, and the Introduction says generative algorithms can produce music and stories. The copy does not specifically address composing lyrics or supporting academic research, and it does not establish human-comparable performance across the listed tasks. (LLM, conf=0.87)
Experimental — accuracy not yet measuredAI systems can function effectively as brainstorming aids, offering a seemingly endless array of possibilities.
“As multiple studies have noted, AI systems can function effectively as brainstorming aids [45,67,82], offering a seemingly endless array of possibilities.”
From Exploring the Collaborative Co-Creation Process with AI: A Case Study in Novice Music Production (2025) · cites this paper as [82]
Several studies find benefits from GenAI in creative tasks, but these findings are typically limited to specific creative tasks.
“For example, several studies find benefits from GenAI in creative tasks, but these findings are typically limited to specific creative tasks (e.g., story writing, ideating business models) [16, 39, 64, 72].”
From Generative AI and Creativity: A Systematic Literature Review and Meta-Analysis (2025) · cites this paper as [72]
Cites
Efficient Estimation of Word Representations in Vector Space
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013) · arXiv · doi:10.48550/arxiv.1301.3781 · Cited by 26,568 · see citing works
A Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707 · see citing works
Difference-in-Differences with multiple time periods
Callaway, B., & Sant’Anna, P. H. C. (2021) · Journal of Econometrics · doi:10.1016/j.jeconom.2020.12.001 · Cited by 9,126 · see citing works
Blind variation and selective retentions in creative thought as in other knowledge processes
Campbell, D. T. (1960) · Psychological Review · doi:10.1037/h0040373 · Cited by 2,329 · see citing works
Experimental evidence on the productivity effects of generative artificial intelligence
Noy, S., & Zhang, W. (2023) · Science · doi:10.1126/science.adh2586 · Cited by 2,150 · see citing works
DINOv2: Learning Robust Visual Features without Supervision
Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al. (2023) · arXiv · doi:10.48550/arxiv.2304.07193 · Cited by 1,078 · see citing works
Generalized Synthetic Control Method: Causal Inference with Interactive Fixed Effects Models
Xu, Y. (2017) · Political Analysis · doi:10.1017/pan.2016.2 · Cited by 938 · see citing works
Li, J., Li, D., Savarese, S., & Hoi, S. (2023) · arXiv · doi:10.48550/arxiv.2301.12597 · Cited by 908 · see citing works
High-Resolution Image Synthesis with Latent Diffusion Models
Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2021) · arXiv · doi:10.48550/arxiv.2112.10752 · Cited by 737 · see citing works
Elgammal, A., Liu, B., Elhoseiny, M., & Mazzone, M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1706.07068 · Cited by 353 · see citing works
The Impact of AI on Developer Productivity: Evidence from GitHub Copilot
Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023) · arXiv · doi:10.48550/arxiv.2302.06590 · Cited by 287 · see citing works
Discernment and creativity: How well can people identify their most creative ideas?
Silvia, P. J. (2008) · Psychology of Aesthetics Creativity and the Arts · doi:10.1037/1931-3896.2.3.139 · Cited by 250 · see citing works
AI model GPT-3 (dis)informs us better than humans
Spitale, G., Biller-Andorno, N., & Germani, F. (2023) · Science Advances · doi:10.1126/sciadv.adh1850 · Cited by 241 · see citing works
Mapping Dimensions of Creativity in the Life-Space
Ivčević, Z., & Mayer, J. D. (2009) · Creativity Research Journal · doi:10.1080/10400410902855259 · Cited by 129 · see citing works
Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
Reimers, N., & Gurevych, I. (2019) · arXiv · doi:10.48550/arxiv.1908.10084 · Cited by 86 · see citing works
Nelson Goodman's Languages of Art
Wollheim, R. (1970) · The Journal of Philosophy · doi:10.2307/2024577 · Cited by 14 · see citing works
McGregor, S., Agres, K. R., Rataj, K., et al. (2019) · Frontiers in Psychology · doi:10.3389/fpsyg.2019.00765 · Cited by 13 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Dell'Acqua, F., McFowland III, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4573321 · Cited by 794 · see citing works
A Comprehensive Review on Generative AI for Education
Mittal, U., Sai, S., Chamola, V., & Sangwan, D. (2024) · IEEE Access · doi:10.1109/access.2024.3468368 · Cited by 383 · see citing works
Terwiesch, C., Meincke, L., Girotra, K., Mollick, E. R., Nave, G., & Ulrich, K. T. (2023) · doi:10.2139/ssrn.4526071 · Cited by 174 · see citing works
Dell’Acqua, F., McFowland, E., Mollick, E., Lifshitz, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026) · Organization Science · doi:10.1287/orsc.2025.21838 · Cited by 1,189 · see citing works
“ChatGPT seems too good to be true”: College students’ use and perceptions of generative AI
Baek, C., Tate, T., & Warschauer, M. (2024) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2024.100294 · Cited by 150 · see citing works
The Double-Edged Roles of Generative AI in the Creative Process: Experiments on Design Work
Hou, J. (., Wang, L., Wang, G., Wang, H. J., & Yang, S. (2025) · Information Systems Research · doi:10.1287/isre.2024.0937 · Cited by 34 · see citing works
Tang, M., Hofreiter, S., Werner, C. H., Zielińska, A., & Karwowski, M. (2025) · The Journal of Creative Behavior · doi:10.1002/jocb.1519 · Cited by 32 · see citing works
Bian, C., Wang, X., Huang, Y., Zhou, S., & Lu, W. (2025) · Humanities and Social Sciences Communications · doi:10.1057/s41599-025-05860-2 · Cited by 30 · see citing works
Exploring the Collaborative Co-Creation Process with AI: A Case Study in Novice Music Production
Fu, Y., Newman, M., Going, L., Feng, Q., & Lee, J. H. (2025) · Proceedings of the 2025 ACM Designing Interactive Systems Conference · doi:10.1145/3715336.3735829 · Cited by 30 · see citing works
“Such synergy has found applications across various domains, including writing [59], fashion design [25], architecture [72], marketing and advertising [39], visual arts [84], and more.”
The Impact of Generative AI Images on Consumer Attitudes in Advertising
Zhang, L., & Hur, C. (2025) · Administrative Sciences · doi:10.3390/admsci15100395 · Cited by 28 · see citing works
Sensemaking with AI: how trust influences human–AI collaboration in health and creative industries
Daly, S. J., Hearn, G., & Papageorgiou, K. (2025) · Social Sciences & Humanities Open · doi:10.1016/j.ssaho.2025.101346 · Cited by 27 · see citing works
Zhou, Y., Liu, Q., Huang, J., & Li, G. (2026) · Technology in Society · doi:10.1016/j.techsoc.2025.103087 · Cited by 24 · see citing works
The Impact of Generative AI on Creativity in Software Development: A Research Agenda
Jackson, V., Vasilescu, B., Russo, D., Ralph, P., Prikladnicki, R., Izadi, M., D’Angelo, S., Inman, S., Andrade, A., & van der Hoek, A. (2025) · ACM Transactions on Software Engineering and Methodology · doi:10.1145/3708523 · Cited by 23 · see citing works
Integrating large language models into project-based learning based on self-determination theory
Zhang, L., & Zhang, W. (2025) · Interactive Learning Environments · doi:10.1080/10494820.2024.2446535 · Cited by 18 · see citing works
Generative AI and Creativity: A Systematic Literature Review and Meta-Analysis
Holzner, N., Maier, S., & Feuerriegel, S. (2025) · arXiv · doi:10.48550/arxiv.2505.17241 · Cited by 17 · see citing works
“GenAI can thus support various human tasks such as writing, software development, composing lyrics, and academic research, often at a performance similar to that of humans [6, 11, 26, 28, 50, 72].”
Herath, S., Bashardoust, A., Bole, Y., & Shrestha, Y. R. (2026) · European Journal of Information Systems · doi:10.1080/0960085x.2026.2616042 · Cited by 16 · see citing works
Perception of AI creativity: dimensional exploration and scale development
Yang, Y., & Xu, H. (2025) · The Journal of Creative Behavior · doi:10.1002/jocb.70028 · Cited by 15 · see citing works
Imagining the Far East: Exploring Perceived Biases in AI-Generated Images of East Asian Women
Lan, X., An, J., Guo, Y., Chiyou, T., Cai, X., & Zhang, J. (2025) · Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706599.3719678 · Cited by 8 · see citing works
The illusion of reality: How AI-generated images (AIGIs) are fooling social media users
Miskolczi, M. (2026) · Computers in Human Behavior · doi:10.1016/j.chb.2025.108876 · Cited by 5 · see citing works
Fu, Y., Bin, H., Zhou, T., Wang, M., Chen, Y., Lai, Z. G. D. C., Wobbrock, J. O., & Hiniker, A. (2024) · arXiv · doi:10.48550/arxiv.2411.00168 · Cited by 4 · see citing works
Fu, Y., Zhou, T., Han, B., Wang, M., Chen, Y., Gomes Da Costa Lai, Z., Pang, R. Y., Reinecke, K., Wobbrock, J. O., & Hiniker, A. (2026) · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791329 · Cited by 2 · see citing works
Exploring the Collaborative Co-Creation Process with AI: A Case Study in Novice Music Production
Fu, Y., Newman, M., Going, L., Feng, Q., & Lee, J. H. (2025) · arXiv · Cited by 30 · see citing works
“Such synergy has found applications across various domains, including writing [59], fashion design [25], architecture [72], marketing and advertising [39], visual arts [84], and more.”
- Authors
- Zhou, E., & Lee, D.
- Affiliations
- Boston University (US)
- Corresponding author
- Zhou, Eric
- Lee, Dokyun
- Venue
- PNAS Nexus
- Publisher
- Oxford University Press (OUP)
- Published
- 2024-02-29
- Volume
- 3
- Issue
- 3
- Pages
- pgae052-pgae052
- ISSN
- 2752-6542 (ISSN-L)
- PMC
- PMC10914360
- OpenAlex
- W4392447866
- Cited by
- 459
- References
- 17
- Citations by year
- 2026: 194 · 2025: 196 · 2024: 66 · 2023: 3
- FWCI
- 1803.69
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- PNAS Nexus (publishedVersion, cc-by-nc-nd)
- PubMed Central (publishedVersion, cc-by-nc-nd)
- PubMed (publishedVersion)
- Funders
- University of Wisconsin-Milwaukee
- Topics
- Social Sciences › Arts and Humanities › Visual Arts and Performing Arts › Art, Technology, and Culture
- Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Digital Media and Visual Art
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
- Keywords
- text-to-image generative AI, human creativity, creative productivity, value capture, creative idea generation, digital art, Midjourney, Stable Diffusion, DALL-E
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
