将重复劳动交给人工智能能否真正释放创造力？研究表明答案并非肯定。[1]根据2024年《科学进展》的研究，虽然AI辅助写作能提高个人的创意度，但整体作品会变得更加相似。[1]2025年微软研究院对319名知识工作者的936个案例进行调查发现，人越相信AI，批判性思考的能力就越少。[1]麻省理工学院的54人脑电实验进一步证实了这一风险：连续使用大模型写作的参与者脑网络连接最弱。[1]
从效率角度看，依赖AI工具也未必带来预期的提升。[1]METR对16名开源开发者的246项任务试验显示，使用2025年初的AI工具后，完成速度反而下降了19%。[1]真正的创造力往往源自重复工作、失败和观察的积累，而不是灵光一现。[1]
针对何时应该将工作外包给AI，专家给出了明确建议。[1]有三类工作不应交给AI处理：问题的来源、第一轮判断和结果责任。[1]相反，当工作具备验收标准清楚、过程可追溯、出错可回滚等特点时，才适合交由AI处理。[1]
A critical examination of whether delegating repetitive tasks to artificial intelligence genuinely liberates human creativity reveals a more complex reality than commonly assumed.[1] Historical precedent demonstrates that genuine creative breakthroughs—from Newton's discoveries to penicillin's invention—emerged not from sudden inspiration but through accumulated repetition, failure, and careful observation.[1] Multiple recent studies challenge the narrative that AI assistance uniformly enhances creative output.
Research published in Science Advances in 2024 found that while AI-assisted writing increased individual creative output, the resulting works became increasingly similar to one another.[1] A 2025 Microsoft Research survey of 319 knowledge workers across 936 cases documented a troubling correlation: the more individuals trusted AI, the less critical thinking they exercised.[1] Brain imaging experiments at MIT involving 54 participants revealed that those using large language models continuously for writing tasks showed the weakest neural network connectivity.[1] A METR analysis of 246 tasks performed by 16 open-source developers found that those using AI tools from early 2025 actually completed work 19 percent more slowly.[1]
The research suggests that certain categories of work should not be delegated to AI.[1] These include identifying the source of problems, making initial assessments, and assuming responsibility for outcomes.[1] By contrast, tasks with clear acceptance criteria, traceable processes, and reversible errors represent suitable candidates for AI assistance.[1]
