nabu.science
MethodologyAboutContact

 

  • Snapshot
  • Key Claims
  • Strengths & Limitations
  • Commentary
  • From the Authors
  • Author Note / Field Feedback
  • Dimension breakdown
  • Readiness
Report #NBU-8E257C6B·Framework v5.10·Pipeline v0.4·Evaluated 23 Oct 2026·OECD: 5.1 Psychology and cognitive sciences·Subfield: Experimental and Cognitive Psychology·Basis: Full text·Type: Laboratory experiment

Creative or Uncreative Partner: Comparing Humans and AI in Collaborative Creative Tasks

Clin KY Lai et al.Clin KY Lai, Simone A Luchini, Nina Lauharatanahirun et al.·2026·Computers in Human Behavior·10.1016/j.chb.2026.109172 ↗

Primary claim

Our findings reveal that the apparent "AI advantage" in creative collaboration is illusory, driven primarily by increased AI verbosity rather than enhanced creativity.

Tentative · post-treatment covariate adjustment does not identify verbosity as the primary cause of the apparent advantage
In brief

A randomized comparison exposes verbosity-sensitive creativity scores, but dependence-blind trajectory analyses constrain its claims about the benefits of human collaboration.

ReliabilityCaveats4 of 4 checks completeRead the 3 entries ↓
Fragile — the claim's weakest requirement is that it works through the path described. Experimental. Every point below is traceable to a quoted passage, but this forecast has not yet been validated against known replication outcomes.
Abstract

: Creativity is fundamentally a collaborative process. As generative AI becomes increasingly integrated into creative work, understanding how it reshapes collaboration becomes critical. This pre-registered study directly compares human-human and human-AI collaboration dynamics across two creative tasks: the Alternate Uses Task (AUT) and creative short story writing. Participants were randomly assigned to pairs in either human-human (N = 68 pairs) or human-AI (GPT-4o; N = 72 pairs) conditions, with partners alternating turns as first responders to examine how initiation order shapes the creative process. Our findings reveal that the apparent "AI advantage" in creative collaboration is illusory, driven primarily by increased AI verbosity rather than enhanced creativity. Critically, collaboration with AI partners negatively impacted humans' own creative responses compared to human-human partnerships, with human-AI collaboration failing to enhance idea originality or diversity relative to human-human collaboration. Human partners demonstrated higher collaborative creativity that strengthened over time, indicating that current generative AI systems, while producing more verbose outputs, do not replicate the collective creativity characteristic of human-human collaboration. These results challenge assumptions about AI’s creative potential, with direct implications for AI system design and collaborative creative practice.

Key Claims

Primary finding · Tentative

Our findings reveal that the apparent "AI advantage" in creative collaboration is illusory, driven primarily by increased AI verbosity rather than enhanced creativity.

Fragile — the claim's weakest requirement is that it works through the path described. Experimental. Every point below is traceable to a quoted passage, but this forecast has not yet been validated against known replication outcomes.

post-treatment covariate adjustment does not identify verbosity as the primary cause of the apparent advantage

Abstract
Quote verified
Secondary · Tentative

collaboration with AI partners negatively impacted humans' own creative responses compared to human-human partnerships

the restricted human-only model adjusts for condition-affected word count and does not establish the broad creative-harm claim

Abstract
Quote verified
Secondary · Tentative

Human participants in Human-AI pairs demonstrated relatively stable semantic distance over time, while humans in Human-Human pairs showed increasing semantic distance across the sequence.

sequence regressions omit reported dyadic dependence correction and repeat identical output after AI responses are removed

Results, AUT Diversity
Quote verified
Secondary · Tentative

Human-AI collaboration does not enhance originality or diversity compared to human-human collaboration

task-specific automated measures and conditional or null comparisons do not establish a general absence of creative enhancement

Highlights
Quote verified
Secondary · Supported

AI as a creative partner greatly increased the length of human responses, even when analyzing only human contributions

mixed-effects analyses report consistent condition-related increases in response length across the two tasks

Discussion, Hypothesis 3: Elaboration
Quote verified

Derived from the full evaluation — not a separate score.

Strengths · Limitations

Strengths

#contribution

Direct comparison extends existing collaboration research

Human–human and human–AI dyads complete both idea-generation and storytelling tasks, with variation in the initiating role. This extends the fixed human-first comparison discussed in the Introduction.

↳ Introduction, Collaborator in Creativity Outputs; Methods

#methodological rigour

Randomized design with disclosed analysis changes

The study reports random assignment, preregistration, participant checks and mixed-effects analyses for originality and elaboration. Convergence failures and model simplifications are disclosed in the Results.

↳ Methods; Results, AUT Originality and Elaboration; Tables 2–3

#contribution

Length sensitivity receives explicit empirical attention

The initial all-output AUT originality advantage disappears after adding covariates including word count. Separating this result from the human-only comparison makes an important automated-scoring dependency visible.

↳ Results, AUT Originality; Table 2

Limitations

#methodological rigour

Trajectory inference ignores repeated dyadic observations

Forward-flow regressions analyze adjacent-response distances without a reported dyadic dependence correction. The AUT analyses including and excluding AI responses also repeat identical output, leaving the claimed human-only replication unresolved.

↳ Results, AUT Diversity; Results, Creative Short Story Task, Diversity

#methodological rigour

Originality interpretation depends on post-treatment adjustment

The human-only originality difference comes from a restricted model adjusting for word count and fluency. Because condition strongly affects response length, this adjustment does not establish either the total effect on human originality or verbosity as the primary explanation.

↳ Results, AUT Originality; Tables 2–3

#positioning

Mechanistic conclusions exceed directly measured outcomes

Offloading, effort and agency were not directly measured. The Conclusion and practical implications nevertheless use these mechanisms to explain the results and motivate broad deployment advice.

↳ Discussion, practical implications and Limitations and Future Directions; Conclusion

Does this match your read? Add your assessment ↓

Nabu Commentary

The clearest supported result is longer output in the AI condition, including longer human contributions. Table 2 also shows that the initial all-output originality advantage is sensitive to word-count adjustment, but that adjustment does not identify the underlying mechanism or establish general creative harm. The forward-flow evidence carries less weight because repeated response distances are analyzed without a reported dyadic dependence correction, and the AUT output is repeated after AI responses are removed. The paper's explicit acknowledgment that mechanisms were not measured further limits its conclusions about cognitive offloading and organizational AI policy.

Check another paper

Free · About 5 minutes · Research preview

Author Note / Field Feedback

Nabu’s assessment, alongside the field’s view.

Are you an author of this paper?

Dimension breakdown

Study Quality

Limited2.8

Confidence high
Contribution
3.4

The direct human–human versus human–AI comparison adds initiation-role variation and human-only analyses to the cited comparative literature. The advance is bounded by task-specific automated outcomes and claims that exceed the adjusted analyses.

Introduction, Collaborator in Creativity Outputs; Methods; Results, AUT Originality
Methodological Rigour
2.5

Random assignment, preregistration and mixed-effects analyses support several outcome comparisons. The trajectory analyses omit reported dyadic dependence correction, while the originality interpretation depends on adjustment for condition-affected word count.

Methods; Table 2; Results, AUT Diversity; Results, Creative Short Story Task, Diversity
Reporting
2.9

The hypothesis-by-hypothesis structure and reported model sequence are followable. Headline claims convert conditional automated-score results into broader conclusions about creative harm and unmeasured engagement.

Abstract; Results, AUT Originality; Discussion; Conclusion
Positioning
2.8

The Introduction engages both AI-benefit and diversity-concern literature, including the closest comparator. The explicit concession that mechanisms were not examined conflicts with subsequent offloading interpretations.

Introduction; Discussion, Limitations and Future Directions; Conclusion

Lower confidence on Methodological Rigour — domain match limited.

Reliability

Caveats4 of 4 checks


Results checkNoted

3 reported values do not reconcile, but the key claims do not rest on them.

  • Human partners demonstrated higher collaborative creativity that strengthened over time: "main effect of running idea sequence (B = -0.002, SE = 0.001, t = -2.20, p = .028)" (Results, AUT Diversity, forward flow (with AI responses)) vs "main effect of running sequence (B = -0.002, SE = 0.001, t = -2.20, p = .028)" (Results, AUT Diversity, forward flow (AI responses removed)). The analyses including and excluding AI responses repeat the same sequence coefficient, standard error, t statistic and p value despite different response sets. → Effect on result unclear
Reporting and transparencyNoted

Consent is described without an ethics-approval declaration, and exclusions lack a numerical disposition account. Table labels and the elaboration-model note contain reporting errors.

Conflict of interest: declared · Ethics approval: not stated · Preregistration: declared

Reference IntegrityNoted

66 references in manuscript 64 of 66 checkable references found in an index

References not verified automatically are resolved manually within 24 hours.

Post-Publication RecordClean

No retraction or expression of concern found in Crossref (includes Retraction Watch).

Last checked: 23 Oct 2026

Sources: Crossref (incl. Retraction Watch) ✓

Readiness

Where this paper’s evidence sits on the path from initial observation to real-world use.

Mechanistic

The online dyadic experiment provides an early controlled demonstration rather than an operational intervention. Sustained use and implementation outcomes are not established.

Methods, Participants and Materials and Procedure; Discussion, Limitations and Future Directions
Report an issue →

AI-generated, human-governed. Something look off? Contact us to request a review.

Keep reading

Related evaluations

Other papers Nabu has evaluated blind, starting with the same field.

  1. Untangling the mechanisms of sleep restriction therapy

    Stanyer, Skeldon, Kyle · 2025 · Clinical medicine

    Primary claim

    it seems likely that adhering to SRT recommendations—that is, reducing sleep opportunity and following prescribed bed and rise-times—drives rapid improvement in nocturnal symptom severity.

    Tentative
    Study quality
    Sound3.3
    Reliability
    No concerns
  2. Enhancing sleep slow waves with auditory stimulation in people with Parkinson's disease: a randomized proof-of-concept trial

    Schreiner, Horlacher, Fattinger, et al. · 2026 · Clinical medicine

    Primary claim

    PTAS increased low-frequency SWA by almost 30% during ON windows, with corresponding increases in slow-wave amplitudes, in 21 analyzed participants.

    Tentative
    Study quality
    Sound3.6
    Reliability
    Caveats
  3. Mind the gap: obtaining reliable sleep estimates and the diagnostic value of sleep discrepancy in individuals with Alzheimer's disease and Lewy body disease

    Gabb, Blackman, Morrison, et al. · 2026 · Clinical medicine

    Primary claim

    Subjective sleep and subjective-objective sleep discrepancy discriminated between cohorts with 80% accuracy.

    Tentative
    Study quality
    Sound3.3
    Reliability
    Caveats

How the rubric works →