Fig 1.
Study overview: Human-AI collaborative brainstorming system.
Conceptual diagram illustrating the turn-based brainstorming system where users and AI agents collaborate on a shared grid interface. The system implements two distinct AI strategies: Vertical AI focuses on deepening existing ideas through incremental development, while Horizontal AI emphasizes exploring brand-new ideas to diversify the creative space. This study investigates which strategy is more effective for building trust and encouraging idea adoption in human-AI creative collaboration.
Fig 2.
System interface and idea selection process.
Left panel shows the main brainstorming interface with the 15x15 grid. Right panel displays the idea selection interface where participants evaluate generated ideas for adoption.
Fig 3.
AI strategy behavioral patterns.
Left panel shows the Vertical (deepening) strategy creating idea clusters in specific columns. Right panel displays the Horizontal (broadening) strategy spreading ideas across different columns to maximize diversity.
Table 1.
Participant distribution across experimental conditions.
Number of participants in each AI strategy condition after excluding those with technical issues.
Fig 4.
Diagram showing the complete experimental procedure from participant briefing through practice session, main brainstorming session, idea evaluation, to post-session questionnaire.
Fig 5.
MDMT trust scale results showing significantly higher trust ratings for the Vertical strategy across multiple dimensions including competence trust and moral trust.
Table 2.
MDMT trust scale results by AI strategy condition.
Table 3.
Godspeed questionnaire results by AI strategy condition.
Fig 6.
Plots showing the distribution of the number of AI ideas selected across experimental conditions. The Vertical strategy demonstrates significantly higher selection counts with reduced variability.
Table 4.
Comprehensive idea selection analysis by strategy condition.
Number of ideas selected across different categories, showing AI strategies primarily affect AI-generated idea adoption rather than overall collaborative productivity.
Fig 7.
Heatmaps showing spatial distribution of ideas and selection rates across the 15x15 grid. Demonstrates the positional bias with higher selection rates in the upper-left regions.
Fig 8.
t-SNE embedding analysis of brainstorming ideas.
Two-panel t-SNE visualization showing: (A) Distribution by AI strategy with distinct clustering patterns between horizontal and vertical conditions, and (B) Distribution by creator type revealing overlapping but distinct regions for human and AI ideas.
Table 5.
Factors significantly affecting idea selection.
Mixed-effects logistic regression results with standardized variables and Creator control variable showing main effects on idea selection (N = 5,848 ideas from 148 participants).
Table 6.
Comprehensive position-content interaction analysis.
Mixed-effects logistic regression results testing for interactions between positional and content effects (N = 5,848 ideas from 148 participants).
Table 7.
Complete strategy pairwise comparisons.
Mixed-effects logistic regression results showing all pairwise comparisons between AI strategies with Creator control variable (N = 5,848 ideas from 148 participants).