Abstract:

Generative AI agents, leveraging their natural language capabilities for human interaction and emergent intelligence that helps humans transcend cognitive fixedness, are increasingly participating autonomously in interactions and collaborations within enterprise new product development (NPD) teams, akin to human members. This creates a novel "multiple-humans-one-machine" collaborative context. Generative AI teammates have become significant new members of NPD teams as well. However, their practical effectiveness and impact on team creativity remain contentious. Existing research regrettably exhibits a threefold disconnection. Firstly, NPD team related research remains confined to contexts involving exclusively human members, overlooking the novel team paradigm introduced by the integration of AI agents. Secondly, AI agent related research predominantly stagnates at the individual level of analysis, lacking a comprehensive team-level perspective. Thirdly, Human-AI collaboration related research is largely limited to dyadic human-AI interaction, failing to extend into the complexities of multi-human-AI teams. Consequently, following the theoretical logic of the Input-Process-Output (IPO) model of team effectiveness, this study investigates the influence of generative AI teammates on team performance in new product idea generation from the team perspective. Based on the three distinct phases of new product idea generation: divergence, convergence, and formation, this study includes three sub-studies.

Sub-study 1 focuses on the divergence phase. It explores the cognitive fixation mechanism underlying the inhibitory effect of generative AI teammates from the perspective of team task processes and identifies related mitigation strategies. The study posits that the superior information processing and logical articulation capabilities of generative AI teammates not only foster team consensus but also discourage human members from voicing unique ideas and intuitions, thereby exacerbating team cognitive fixation and inhibiting the diversity of ideas generated by human members. Compared with interactive groups, this inhibitory effect of generative AI teammates will be effectively mitigated in nominal groups.

Sub-study 2 examines the convergence phase. It investigates the social identification mechanism underlying the reinforcing effect of generative AI teammates from a team affective process perspective and proposes related enhancement strategies. As human members perceive generative AI teammates as lacking value judgments, subjective preferences, and emotional capabilities, the study posits that human members perceived team social identification - including emotional exchange - among themselves is heightened after generative AI teammates joined their team. This consequently enhances the convergence of idea adoption among human members. The reinforcing effect of generative AI teammates in the convergence phase can be further amplified by high team diversity beliefs.

Sub-study 3 addresses the formation phase. It systematically explores the double-edged sword effect of generative AI teammates and constructing corresponding coping strategies. The study contends that generative AI teammates, by inhibiting the diversity of ideas generated in the divergence phase and enhancing the convergence of idea adoption in the convergence phase, ultimately increase the speed of team new product idea generation while decreasing the quality of team new product idea generation in the formation phase. Furthermore, when human members receive generative AI skill training, the positive effect of generative AI teammates on the team new product idea generation speed will be further strengthened, while the negative effect of generative AI teammates on the team new product idea generation quality will be effectively alleviated.

This study specifically focuses on the role of generative AI teammates as new members within NPD teams. It reveals the mechanisms through which generative AI teammates influence team interaction and collaboration among human members at the team level, and constructs optimized collaborative strategies for generative AI teammates operating within the multiple-humans-one-machine context. Consequently, this research not only enriches the theoretical understanding of team effectiveness models and expands human-AI collaboration strategies from "one-human-one-machine" to "multiple-humans-one-machine" contexts, but also provides significant practical implications for enterprises seeking to effectively leverage generative AI teammates in NPD teams and offer significant decision-making references for the Chinese government in implementing its “AI+” initiative.

Key words: generative artificial intelligence, new product development team, new product idea generation, human-AI team collaboration strategy