The Effects of Dual-Pathway AI Assistance on Divergent Thinking in Tasks of Varying Semantic Richness

Generative artificial intelligence (AI) shows promise for enhancing creative performance, but its capacity to augment human creative thinking remains debated. Grounded in the dual-pathway model of creativity, this study investigates how two AI assistance strategies, flexible (promoting cross-category exploration) and persistent (promoting within-category deepening), influence divergent thinking across tasks differing in semantic richness. In a within-subjects design (N = 35), participants completed the Alternative Uses Task for semantically rich and sparse objects under three assistance conditions: flexible AI, persistent AI, and no-AI baseline. Using a turn-taking brainstorming paradigm, the AI provided tailored cues after each participant’s idea. Results revealed a crucial task-strategy fit: persistent AI significantly increased mean originality only for semantically rich tasks, where deeper semantic exploration is needed. Temporal analyses further showed that, across both task types, both flexible and persistent AI elevated initial originality compared with the baseline, albeit with a slower subsequent increase. For semantically sparse tasks, these analyses provided tentative evidence of an originality benefit from flexible AI, but this pattern was not evident in overall mean originality. Cross-lagged models further showed an asymmetric influence in the persistent-AI condition: AI ideas predicted participants’ subsequent originality, but not vice versa. Mediation analysis suggested the persistent AI benefit was accounted for by stronger AI-to-human coupling. These findings suggest that AI’s creative benefits depend on aligning its assistance style with the cognitive demands of the task.