Scholarlink

<p>Artificial intelligence (AI) is reshaping scientific knowledge production, yet how team-level AI capability translates into individual productivity remains insufficiently understood. Focusing on synthetic biology, we develop an AI dynamic capability (AIDC) framework—sensing, seizing, and reconfiguring—to unpack team-level AI effects on individual researchers. Using a mixed-methods design that&#xa0;combines, survey data&#xa0;with team interviews, we find domain-specific rather than universal productivity gains. Team AIDC is strongly associated with AI-related outputs, but only weakly linked to overall research output. These benefits are amplified in large teams and accrue disproportionately to frontline researchers, consistent with specialization and execution-layer engagement with models and data. Interviews attribute this domain specificity to downstream constraints,&#xa0;especially experimental validation,&#xa0;and identify reconfiguring as the pivotal mechanism through which teams redesign workflows and human–AI task boundaries to convert AI outputs into validated results. We conclude by discussing future research directions and policy implications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

One capability, many outcomes: unpacking AI dynamic capabilities in scientific teams

  • 期刊论文

  • 发表日期：

    2026年7月4日

查看全文
  • Liyun Hu,
  • Zhaopeng Li,
  • Ran Jin,
  • Bingdao Zheng,
  • Li Tang

摘要

Artificial intelligence (AI) is reshaping scientific knowledge production, yet how team-level AI capability translates into individual productivity remains insufficiently understood. Focusing on synthetic biology, we develop an AI dynamic capability (AIDC) framework—sensing, seizing, and reconfiguring—to unpack team-level AI effects on individual researchers. Using a mixed-methods design that combines, survey data with team interviews, we find domain-specific rather than universal productivity gains. Team AIDC is strongly associated with AI-related outputs, but only weakly linked to overall research output. These benefits are amplified in large teams and accrue disproportionately to frontline researchers, consistent with specialization and execution-layer engagement with models and data. Interviews attribute this domain specificity to downstream constraints, especially experimental validation, and identify reconfiguring as the pivotal mechanism through which teams redesign workflows and human–AI task boundaries to convert AI outputs into validated results. We conclude by discussing future research directions and policy implications.