[論文レビュー] Human-AI Synergy Supports Collective Creative Search

Chenyi Li, Raja Marjieh+5 more|arXiv (Cornell University)|Feb 10, 2026

Artificial Intelligence in Games被引用数 0

ひとこと要約

Hybrid human–AI groups outperform human-only and AI-only groups in a collective semantic search task, achieving faster convergence while preserving guess diversity. Humans and AI adapt to each other, with AI benefiting more from human input and humans gaining from AI assistance.

ABSTRACT

Generative AI is increasingly transforming creativity into a hybrid human-artificial process, but its impact on the quality and diversity of creative output remains unclear. We study collective creativity using a controlled word-guessing task that balances open-endedness with an objective measure of task performance. Participants attempt to infer a hidden target word, scored based on the semantic similarity of their guesses to the target, while also observing the best guess from previous players. We compare performance and outcome diversity across human-only, AI-only, and hybrid human-AI groups. Hybrid groups achieve the highest performance while preserving high diversity of guesses. Within hybrid groups, both humans and AI agents systematically adjust their strategies relative to single-agent conditions, suggesting higher-order interaction effects, whereby agents adapt to each other's presence. Although some performance benefits can be reproduced through collaboration between heterogeneous AI systems, human-AI collaboration remains superior, underscoring complementary roles in collective creativity.

研究の動機と目的

• Investigate how hybrid human–AI groups perform in a collective semantic search task compared to human-only and AI-only groups.
• Assess how collaboration affects performance, exploration, and diversity at both individual and collective levels.
• examine whether cognitive heterogeneity among AI systems contributes to gains beyond agent diversity.
• analyze second-order effects where humans and AI adapt their strategies in response to each other’s behavior.

提案手法

• Participants (humans and Gemini 2.5 AI agents) play a word-guessing game with a hidden target word and semantic similarity feedback.
• Guess scores are computed as a weighted product of the hidden-word score and the cosine similarity of embeddings.
• In each round, players see the best guess from previous rounds within the same game as a hint.
• Games are organized in a chain network with 10 rounds of 10 turns each (100 guesses per game).
• Four experimental conditions are compared: Human Social, Human Asocial, AI-only, and Human–AI Hybrid.

実験結果

リサーチクエスチョン

• RQ1 How do hybrid human–AI groups compare to homogeneous groups in terms of performance and exploration/diversity in collective semantic search?
• RQ2 Do human and AI strategies adapt to each other in hybrid settings, and what are the directional effects on performance and diversity?
• RQ3 Is the observed hybrid advantage driven by cognitive heterogeneity across AI systems, or by other factors such as prompting or information sharing?
• RQ4 How robust are hybrid gains to variations in AI type and communication channels?

主な発見

• Hybrid human–AI groups achieve the highest individual and collective performance across rounds and targets.
• AI-only groups show lower performance and limited diversity, often converging early to suboptimal regions.
• Humans in hybrid groups increase lexical diversity and unique guesses, while AI agents improve in performance and diversity when interacting with humans.
• Hybrid groups preserve higher diversity than AI-only groups and match human groups in collective diversity.
• Both humans and AI adapt their strategies in hybrid settings, with humans expanding exploration and AI exploiting efficiently; this co-adaptation underpins the hybrid advantage.
• Control experiments indicate benefits are not solely due to agent diversity but to cognitive differences and interactions between humans and AI.