CoChat
Sign Up
PreprintCited by 967Creativity in Education and NeuroscienceEthics and Social Impacts of AIArtificial Intelligence in Games

Generative artificial intelligence enhances creativity but reduces the diversity of novel content

Open access
Doshi, Anil, Hauser, Oliver·2023·Science Advances
DOI 10.2139/ssrn.4535536PMC PMC11244532
Open this paper in CoChat — free accountOpen PDF

Abstract

Creativity is core to being human. Generative artificial intelligence (GenAI) holds promise for humans to be more creative by offering new ideas, or less creative by anchoring on GenAI ideas. We study the causal impact of GenAI on the production of a creative output in an online experimental study where some writers are could obtain ideas for a story from a GenAI platform. Access to GenAI ideas causes an increase in the writer’s creativity with stories being evaluated as better written and more enjoyable, especially among less creative writers. However, GenAI-enabled stories are more similar to each other than stories by humans alone. Our results have implications for researchers, policy-makers and practitioners interested in bolstering creativity, but point to potential downstream consequences from over-reliance.

AI snapshot

Written from the paper’s full text.

TL;DR

In a preregistered online experiment, writers given generative AI story ideas produced stories rated as more creative, better written, and more enjoyable, especially less inherently creative writers. However, AI-assisted stories were more similar to one another and to the AI ideas, suggesting individual gains at the cost of collective novelty.

Key findings

  • Access to generative AI increased third-party evaluator ratings of novelty and usefulness: one AI idea raised novelty by 5.4% (b = 0.207, P = 0.021) and usefulness by 3.7% (b = 0.185, P = 0.039); five AI ideas raised novelty by 8.1% (b = 0.311, P < 0.001) and usefulness by 9.0% (b = 0.453, P < 0.001) versus Human-only.
  • Benefits were concentrated among less inherently creative writers: low-DAT writers with five AI ideas improved novelty by 10.7%, usefulness by 11.5%, how well written by up to 26.6%, enjoyment by up to 22.6%, and boringness decreased by up to 15.2%; high-DAT writers saw little effect.
  • AI-assisted stories were more similar to other stories in the same condition (one idea: b = 0.871, P < 0.001; five ideas: b = 0.718, P = 0.003) and more similar to the AI ideas (5.2% and 5.0% increases, respectively).
  • Writers’ self-assessments showed no statistically significant differences in novelty or usefulness across conditions; evaluators imposed an ownership penalty of at least 25% on writers who received AI ideas.

Methods

  • Preregistered two-phase online experiment: N = 293 writers on Prolific were randomized to Human-only, Human with one GenAI idea, or Human with five GenAI ideas; they wrote an eight-sentence story for a teenage/young adult audience, with GPT-4 providing three-sentence story ideas in AI conditions.
  • Writers completed the divergent association task (DAT) before writing; DAT mean was 77.24, SD = 6.48, computed for 291 of 293 writers.
  • N = 600 separate evaluators rated six randomly assigned stories each, yielding 3,519 evaluations; outcomes were analyzed with OLS regressions using robust standard errors, clustered at evaluator level for evaluator outcomes, plus cosine similarity of OpenAI embeddings.

Limitations

  • The paper states the creative task was constrained in length (eight sentences), medium (writing), and output type (short story), with no interactiveness with the LLM or variation in prompts, limiting generalizability.
  • The paper states the sample was relatively “typical” study participants, not professional writers or unusually creative individuals, and that the usefulness construct was adapted to this context.
  • Evident from the text: writers could opt into receiving AI ideas rather than being forced to use them, though the authors argue intention-to-treat analysis and high uptake (88.4%) support causal interpretation; no financial incentives for creativity were introduced.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.2139/ssrn.4535536

Cites

55 of 55 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

4 of 9 cited sources checked · 3 supported · 0 flagged · accuracy not yet measured

  • Interpretationcitation

    Having access to generative AI effectively equalizes the evaluations of stories, removing any disadvantage or advantage based on the writers' inherent creativity.

    “First, having access to generative AI effectively equalizes the evaluations of stories, removing any disadvantage or advantage based on the writers’ inherent creativity (25).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [25] H Wang, 2023, Scientific discovery in the age of artificial intelligence, Nature source

      The Abstract describes a task for measuring verbal creativity, and the Discussion > Strengths describes its automatic scoring; neither addresses AI-assisted story writing or story evaluations. Searches for generative AI, story writing, and equalized ratings found no passage supporting the claimed effect. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
  • Findingown data

    Access to generative AI ideas causes stories to be evaluated as more creative.

    “We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
  • Findingown data

    The effect of access to generative AI ideas on creativity is larger among less creative writers.

    “We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers.”
    • Also stated in: another passage · plain
      “However, the gains from writing more creative stories benefit some more than others: Less creative writers experience greater uplifts for their stories, seeing increases of 10 to 11% for creativity and of 22 to 26% for how enjoyable and well written the story is.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
  • Findingown data

    Access to generative AI ideas causes stories to be evaluated as more enjoyable.

    “We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers.”
    • Also stated in: another passage · plain
      “we find that stories written by writers with access to generative AI ideas are more enjoyable (Human with one GenAI idea condition: b = 0.216, P = 0.028; Human with five GenAI ideas condition: b = 0.375, P < 0.001, see table S6)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
  • Findingown data

    Generative AI-enabled stories are more similar to each other than stories by humans alone.

    “However, generative AI–enabled stories are more similar to each other than stories by humans alone.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
  • Classified as probably not a substantial claim. 35
  • May combine two claims. 23
  • The quote only partly states this claim. 1

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Creativity is typically assessed across two dimensions: novelty and usefulness.

    “Creativity is typically assessed across two dimensions: novelty and usefulness (11, 12).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [11] W Kenower, 2020, The Cold Open: Facing the Blank Page. Writer's Diget source
    • cites [12] S Harvey, 2022, Toward a meta-theory of creativity forms: How novelty and usefulness shape creativity source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    That generative AI particularly benefited less able writers is paralleled in recent studies focusing on other domains in which generative AI has been shown to help less productive workers.

    “That generative AI particularly benefited less able writers is paralleled in recent studies focusing on other domains in which generative AI has been shown to help less productive workers (4, 5).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [4] S Noy, 2023, Experimental evidence on the productivity effects of generative artificial intelligence, Science source
    • cites [5] E Brynjolfsson, 2023, Generative AI at work, National Bureau of Economic Research source

  • May combine two claims.

    Study design and conductcitation

    The study sample consists of relatively "typical" study participants often used in academic studies, not professional writers or unusually creative individuals.

    “We do so among a sample of relatively “typical” study participants often used in academic studies (which comes with limitations on population representativeness) (41)—that is, we do not study professional writers or unusually creative individuals.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [41] #* (-1"",#(&%)'(#"'# source

  • May combine two claims.

    Findingown data

    Generative AI assistance increases both the novelty and usefulness of stories.

    “We find that generative AI assistance increases both the novelty and usefulness of stories (results are discussed in section S4).”
    • Also stated in: another passage · plain
      “We find that having access to generative AI causally increases the average novelty and usefulness—two frequently studied dimensions of creativity—relative to human writers on their own.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Most evaluators indicated that disclosure of the use of AI or the underlying text from AI should be part of publications that used such tools.

    “Most evaluators also indicated that disclosure of the use of AI or the underlying text from AI should be part of publications that used such tools.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Writers' self-assessments of their own stories show no statistically significant differences in the story characteristics across conditions.

    “Again, writers’ self-assessments of their own stories show no statistically significant differences in the story characteristics across conditions (see table S7).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Stories in the Human with five GenAI ideas condition are not evaluated as more funny than those in the Human-only condition.

    “Stories in the Human with five GenAI ideas condition are, however, not evaluated as more funny than those in the Human-only condition; if anything, the coefficient is negative but not statistically significant (b = −0.106, P = 0.115).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.

  • May combine two claims.

    Findingown data

    Across all conditions, high-DAT writers' stories are evaluated relatively highly, in terms of both novelty and usefulness, and providing them with access to generative AI does not affect their high evaluations.

    “Across all conditions, high-DAT writers’ stories are evaluated relatively highly, in terms of both novelty and usefulness, and providing them with access to generative AI does not affect their high evaluations.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.

  • May combine two claims.

    Findingown data

    Most evaluators found the use of AI in writing stories to be ethical and still a "creative act."

    “Overall, however, most evaluators found the use of AI in writing stories to be ethical and still a “creative act.””
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Most evaluators indicated that the content creators on which the models were based should be compensated.

    “most evaluators indicated that the content creators, on which the models were based, should be compensated.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The increase in similarity from having access to one or five generative AI ideas represents 10.7% and 8.9% of the total range, respectively.

    “To put these values in context, consider that in the Human-only condition, the similarity scores span a range of 8.10 points; therefore, the increase in similarity from having access to one or five generative AI ideas represents 10.7% and 8.9% of the total range, respectively.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.

  • May combine two claims.

    Findingown data

    The results show that generative AI professionalizes the writing but reduces the variance in creative outputs.

    “with our results showing that generative AI professionalizes the writing but reduces the variance in creative outputs”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Participants were randomly assigned to one of three conditions.

    “Participants were randomly assigned to one of three conditions: Human-only, Human with one GenAI idea condition, and Human with five GenAI ideas (see table S1 for balance across conditions).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    These results point to an increase in individual creativity at the risk of losing collective novelty.

    “These results point to an increase in individual creativity at the risk of losing collective novelty.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
    • Rests on an assumption: Collective novelty is valuable.

  • May combine two claims.

    Interpretationown data

    With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced.

    “This dynamic resembles a social dilemma: With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.
    • Rests on an assumption: Collective novelty is valuable.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    Little is known about generative AI's potential impact on human creativity.

    “little is known about generative AI’s potential impact on a fundamental human behavior: the ability of humans to be creative.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The emergence of generative AI technologies is challenging several long-standing assumptions about the uniqueness and superiority of human-generated content.

    “However, the emergence of generative artificial intelligence (AI) technologies—such as large language models (LLMs) as used in our study—is challenging several long-standing assumptions about the uniqueness and superiority of human-generated content (2).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] Z Epstein, Investigators of Human Creativity (2023a). Art and the science of generative AI, Science source

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Study design and conductown data

    The conditions were balanced across conditions.

    “see table S1 for balance across conditions”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The results point to an increase in individual creativity, but there is risk of losing collective novelty.

    “While these results point to an increase in individual creativity, there is risk of losing collective novelty.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    The number of evaluations per story varied because of random assignment of stories to evaluators, with counts ranging from 9 to 14 reviews per story.

    “The number of evaluations per story varied because of random assignment of stories to evaluators: One story received 9 reviews, 9 stories received 10 reviews, 61 stories received 11 reviews, 141 stories received 12 reviews, 77 stories received 13 reviews, and 4 stories received 14 reviews.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Despite the enhancement effect that generative AI had on individual creativity, there may be a cautionary note if generative AI were adopted more widely for creative tasks.

    “our results suggest that despite the enhancement effect that generative AI had on individual creativity, there may be a cautionary note if generative AI were adopted more widely for creative tasks.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The order in which stories were presented for review was randomized across evaluators.

    “The order in which the stories were presented for review was randomized across evaluators.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    The study was designed such that endogenous decisions by the writer are minimized, but not fully eliminated.

    “We designed our study such that endogenous decisions by the writer are minimized, but not fully eliminated.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study was preregistered at AsPredicted.org (ID 136723).

    “The study was preregistered at AsPredicted.org (ID 136723); a copy of the preregistration is included in section S9.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    If writers were given full control over the AI engine or real-time interactions, the effectiveness and magnitude of the impact of generative AI on creativity may increase.

    “both of which may increase the effectiveness and magnitude of the impact of generative AI on creativity.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    When given the option to call upon generative AI more than once in the Human with five GenAI ideas condition, participants did so on average 2.55 times, with 24.5% requesting the maximum of five generative AI ideas.

    “When given the option to call upon generative AI more than once in the Human with five GenAI ideas condition, participants did so on average 2.55 times, with 24.5% requesting the maximum of five generative AI ideas.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    If individual writers find out that their generative AI-inspired writing is evaluated as more creative, they have an incentive to use generative AI more in the future, but by doing so, the collective novelty of stories may be reduced further.

    “If individual writers find out that their generative AI-inspired writing is evaluated as more creative, they have an incentive to use generative AI more in the future, but by doing so, the collective novelty of stories may be reduced further.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Informed consent was obtained for both the writer study and the evaluator study.

    “Informed consent was obtained for both the writer study and the evaluator study.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study did not study or vary the myriad of motivating factors that encourage creativity in the real world.

    “we did not study or vary the myriad of motivating factors that encourage creativity in the real world.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    All stories were evaluated by multiple evaluators on novelty, usefulness, and several emotional characteristics.

    “All stories were evaluated by multiple evaluators on novelty, usefulness, and several emotional characteristics, which comprise key outcome variables related to our main research question (see section S1).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationown data

    These results indicate support for the use of generative AI in creative outputs, with important potential limits on ownership or credit and requirements for disclosure.

    “These results indicate support for the use of generative AI in creative outputs, with important potential limits on ownership or credit and requirements for disclosure.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    These constraints limit the generalizability and conclusions that can be drawn from the study.

    “These constraints limit the generalizability and conclusions we can draw from this study.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Participants were randomized into one of three experimental conditions: Human-only, Human with one GenAI idea, and Human with five GenAI ideas.

    “Participants were randomized into one of three experimental conditions: Human-only, Human with one GenAI idea, and Human with five GenAI ideas.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    In the Human with one GenAI idea condition, 82 of 100 writers opted to generate one idea.

    “Of the 100 writers in the Human with one GenAI idea condition, 82 opted to generate one, while 93 of 98 writers in the Human with five GenAI ideas condition did so.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    There were a total of 3519 evaluations of 293 stories made by 600 evaluators.

    “There were a total of 3519 evaluations of 293 stories made by 600 evaluators.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Cronbach's α for the novelty and usefulness indices was 0.92 and 0.89, respectively.

    “Cronbach’s α for the novelty and usefulness indices was 0.92 and 0.89, respectively.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The experimental design offers a fairly stringent test to measure the causal impact of generative AI on creativity.

    “Our choice of the experimental design offers a fairly stringent test to measure the causal impact of generative AI on creativity (34).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [34] If a human creator (author) uses AI in part of the writing of a story, the AI-generated content should be accessible alongside the final story source

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Evaluators read six randomly selected stories without being informed about writers being randomly assigned to access generative AI.

    “Evaluators read six randomly selected stories without being informed about writers being randomly assigned to access generative AI in some conditions (or not).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    Writers were not allowed to customize the call to the generative AI engine, nor were repeated interactions between writers and generative AI allowed.

    “We do not allow writers to customize the call to the generative AI engine, nor do we allow for repeated interactions between writers and generative AI”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The creative task is constrained in length (eight sentences), medium (writing), and type of output (short story), and there is no interactiveness with the LLM or variation in prompts.

    “our study has limitations in that the creative task is constrained in its length (i.e., eight sentences), medium (i.e., writing), and type of output (i.e., short story), and there is no interactiveness with the LLM or variation in prompts.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Writers were randomly assigned to one of the generative AI conditions to mitigate selection bias.

    “In our study, we randomly assigned writers to one of the generative AI conditions to mitigate selection bias.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study was preregistered.

    “This paper aims to provide an initial answer to these questions through a preregistered, two-phase experimental online study on written creative output”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    If the publishing (and self-publishing) industry were to embrace more generative AI-inspired stories, the produced stories would become less unique in aggregate and more similar to each other.

    “if the publishing (and self-publishing) industry were to embrace more generative AI-inspired stories, our findings suggest that the produced stories would become less unique in aggregate and more similar to each other.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The study was approved by the ethics boards at the UCL School of Management (ID UCLSOM-2023-002) and the University of Exeter (ID 1642263).

    “The study was approved by the ethics boards at the UCL School of Management (ID UCLSOM-2023-002) and the University of Exeter (ID 1642263).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The overall effect is a more novel and even more useful story that is well written and enjoyable.

    “The overall effect is a more novel and even more useful story that is well written and enjoyable.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Writers in the Human-only condition did not use generative AI.
    • Rests on an assumption: GPT-4 is representative of generative AI more broadly.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Cronbach's α for the ownership index was 0.92.

    “Cronbach’s α for the ownership index was 0.92.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    Constraining the task may constrain the extent to which participants are able to express their creativity and may not generalize to other less-constrained creativity tasks.

    “Constraining the task in such a way may constrain the extent to which participants are able to express their creativity and may not generalize to other less-constrained creativity tasks.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The DAT score was successfully computed for 291 of 293 writers.

    “Two writers failed to properly submit seven valid words; thus, the DAT score was successfully computed for 291 of 293 writers.”
    Quote in the paperNot a substantial claim
48 of 48

  • The paper relies on this without support.

    Writers in the Human-only condition did not use generative AI.

    “In our Human-only baseline condition, writers were assigned the task with no mention of or access to generative AI.”

  • The paper relies on this without support.

    GPT-4 is representative of generative AI more broadly.

    “we gave writers the option to call upon a generative AI platform (OpenAI’s GPT-4 LLM) to provide a three-sentence starting idea to inspire their own story writing.”

  • The paper relies on this without support.

    Collective novelty is valuable.

    “These results point to an increase in individual creativity at the risk of losing collective novelty.”

  • The paper relies on this without support.

    Allowing writers to opt into receiving generative AI ideas ensures that writers are invested in and receptive to what generative AI produces.

    “We do this to ensure that writers are invested in, and receptive to, what generative AI produces.”

  • The paper relies on this without support.

    Evaluators in the evaluator study represent “regular” readers.

    “Participants were not selected on the basis of prior experience in the publishing industry, but represent “regular” readers.”
5 of 5

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?

  • Prior studies compared essays written entirely by students and those generated entirely by ChatGPT, which may not accurately reflect how most students interact with AI chatbots in reality.

    “First, prior studies compared essays written entirely by students and those generated entirely by ChatGPT 7-8,14,15 . However, this may not accurately reflect how most students interact with AI chatbots in reality.”

    From Style, sentiment, and quality of undergraduate writing in the AI era: A cross-sectional and longitudinal analysis of 4,820 authentic empirical reports (2025) · cites this paper as [8]

    The Introduction and Materials and Methods (page not supplied) describe a comparison between human-written stories produced without AI access and human-written stories whose writers could request GPT-4 starting ideas. The study did not compare student essays with essays generated entirely by ChatGPT. The uncited sentence about how students interact with chatbots is the citing authors’ own qualification, not a claim attributed to this source. (LLM, conf=0.90)

    Checked by a reasoning model — may be wrong.

    Experimental — accuracy not yet measured

  • The most widely accepted framework evaluates innovation across two primary dimensions: novelty and usefulness.

    “The most widely accepted framework evaluates innovation across two primary dimensions: novelty and usefulness (Amabile, 1982; Doshi and Hauser, 2024; Harvey and Berry, 2022).”

    From AI-assisted design synthesis and human creativity in engineering education (2026) · cites this paper as Doshi and Hauser, 2024

    The Introduction says creativity is typically assessed across novelty and usefulness, and Materials and Methods > Outcome variables uses those two indices. These passages support the two dimensions, but describe a framework for creativity rather than establishing that it is the most widely accepted framework for evaluating innovation. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • Generated ideas may be less diverse than those of humans due to the training data of GenAI.

    “Hence, simply by means of the training data of GenAI, the generated ideas may also be less diverse than those of humans [16].”

    From Generative AI and Creativity: A Systematic Literature Review and Meta-Analysis (2025) · cites this paper as [16]

    The Results section, “Similarity of stories” (page unavailable), finds that stories written with access to GenAI ideas were more similar to one another than human-only stories. It also reports that writers were anchored to the GenAI ideas they received. The paper does not establish that GenAI-generated ideas themselves are less diverse than human ideas, or attribute the reduced diversity to GenAI’s training data. (LLM, conf=0.87)

    Experimental — accuracy not yet measured

  • AI can generate unexpected ideas and thus evoke a stronger sense of novelty.

    “Equipped with a vast knowledge base and divergent thinking, AI can generate unexpected ideas and thus evoke a stronger sense of novelty (Doshi and Hauser, 2024; Tigre Moura et al., 2023)”

    From From humans to AI: understanding why AI is perceived as the preferred co-creation partner (2025) · cites this paper as Doshi and Hauser, 2024

    In Results (Baseline versus generative AI conditions; page unavailable), stories written with access to AI-generated ideas received higher novelty ratings than stories written without that access. The paper does not establish that the AI ideas themselves were unexpected or that they evoked a stronger subjective sense of novelty; it measured evaluators’ ratings of the resulting stories. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • LLMs are trained on large corpora of existing scientific text, including published papers and public records of funded projects, and may tend to produce high-probability, discipline-typical language.

    “Because LLMs are trained on large corpora of existing scientific text, including published papers and public records of funded projects, they may tend to produce high-probability, discipline-typical language.”

    From The rise of large language models and the direction and impact of US federal research funding (2026) · cites this paper as [36]

    The Abstract and Discussion report that AI-assisted short stories were more similar to one another and that AI reduced variation in creative outputs, lending limited support to a tendency toward less varied writing. The paper does not establish that LLMs were trained on published scientific papers or funded-project records, or that they produce high-probability, discipline-typical language. (LLM, conf=0.76)

    Experimental — accuracy not yet measured

Cites

Flagged first, then by citations

Research works printed but not linked: 18

  1. W Kenower, 2020, The Cold Open: Facing the Blank Page. Writer's Diget
  2. K Girotra, 2023, Ideas are Dimes a Dozen: Large Language Models for Idea Generation in Innovation
  3. If a human creator (author) uses AI in part of the writing of a story, the AI-generated content should be accessible alongside the final story
  4. $&4&*+2
  5. (*('622+&
  6. &(-$""(*& 5'6'$-"#+
  7. (&'("''''"(# #+"$
  8. #* (-1"",#(&%)'(#"'#
  9. #&('!')&'0'+ ' ##((&"*) #!$#
  10. (("-'''(" 49-&9#, 8
  11. +, #")((#,!"(!"%)'(#"3-$#
  12. ), #* (-")') "''-#"(#"0"&)"('&&''#
  13. ), &#)'("''('('#&#('0++ " )&
  14. #"'&"#('(('( -'""(&#! #(&26+ '( &$#&(, 70+$ "(##!"((+# #"(#"'"(##"#"(#"(#"#&#)&!"" -''(!"#)(#!*& '"(#'(+##
  15. -#&" -''$)&$#''2+ '#&#$&'$#
  16. A>::(&4$&(-* )(#&'6#+* )('','(#&'&+
  17. #", #&#&"('(#&-'0(27#" ' &#!;(#?0#&+++ '(
  18. + '##"'& ; #)(#!'"&), Writer study screenshots Condition: Human only Condition: Human with 1 AI idea (before story has been generated) Condition: Human with 1 AI idea (after story has been generated) Condition: Human with 5 AI ideas (before story has been generated) Condition: Human with 5 AI ideas (after story has been generated) Only shown in condition: Human only Only shown in condition: Human only and answered "yes" on previous page Only shown in condition: Human with 1 AI idea Only shown in condition: Human with 5 AI ideas (before rating any story) Only shown in condition: Human with 5 AI ideas (after rating a story) Section 6. Evaluator study screenshot Example of a story from the Human Only condition: the writer said they did not use an outside AI tool Example of a story from the Human Only condition: the writer said they did use an outside AI tool Example of a story from the Human with 1 or 5 AI idea(s) condition but the author did not access the GenAI idea Example of a story from the Human

Cited by

Sorted by citations

Related

Authors
Doshi, A., & Hauser, O.
Affiliations
  • University College London (GB)
  • University of Exeter (GB)
Corresponding author
  • Doshi, Anil
  • Hauser, Oliver
ORCID
Venue
Science Advances
Publisher
Elsevier BV
Published
2023
Volume
10
Issue
28
Pages
eadn5290-eadn5290
ISSN
2375-2548 (ISSN-L)
DOI
10.2139/ssrn.4535536
PMC
PMC11244532
OpenAlex
W4400578758
Semantic Scholar
5409f3fa0d9e1a7ffea029557a5871a25b3b1b7e
Cited by
887
References
42
Citations by year
2026: 483 · 2025: 321 · 2024: 74 · 2023: 9
FWCI
359.70
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Science Advances (publishedVersion, cc-by)
  • SSRN Electronic Journal (acceptedVersion)
  • PubMed (publishedVersion)
  • UCL Discovery (University College London) (submittedVersion, cc-by)
  • PubMed Central (submittedVersion, cc-by)
  • Figshare (submittedVersion, cc-by)
  • Open Research Exeter (University of Exeter) (submittedVersion, cc-by)
Funders
  • Harvard Business School
  • University of Exeter
  • University of Oxford
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Creativity in Education and Neuroscience
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence in Games
Keywords
generative AI, large language models, creativity, short stories, story ideas, individual creativity, novelty, social dilemmas
MeSH terms
Artificial Intelligence, Creativity, Humans, Language, Writing

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 2026-09-28.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-28.

Cited sources are checked by gpt-6-sol.

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