20世纪初，管理顾问泰勒透过观察与记录，试图将工厂工人的传统经验与隐性知识系统化。如今，企业在将人工智慧导入白领阶级的工作时，也面临著类似的挑战，因为许多对AI模型至关重要的领域知识与判断力，仍锁在员工的脑海中。

由于许多复杂任务难以被明确定义与编码，通用型AI模型在处理需要专业判断的工作时表现不佳，甚至透过改良提示词也无法解决问题。因此，许多企业开始寻求员工的协助，利用企业内部的专业知识与数据来微调专属的AI模型，从而大幅提升模型的准确度与实用性。

然而，这种要求员工协助训练AI的过程引发了信任与利益的考验。员工可能会担心交出专业知识后被AI取代，或失去工作上的自主权；因此，在接下来的AI发展阶段中，那些能建立互信文化、并让员工相信能从AI发展中获益的企业，才更有机会在竞争中脱颖而出。

In the early 20th century, management consultant Frederick Winslow Taylor attempted to systematize the traditional and tacit knowledge of factory workers through observation and recording. Today, companies face a similar challenge when introducing artificial intelligence into white-collar workplaces, as much of the domain knowledge and judgment crucial for AI models remains locked inside employees' heads.

Because many complex tasks are difficult to explicitly define and codify, general-purpose AI models often underperform in jobs requiring professional judgment, a problem that even improved prompting cannot solve. Consequently, many businesses are enlisting their employees' help to fine-tune bespoke AI models using internal expertise and data, thereby significantly improving the accuracy and utility of the models.

However, this process of asking employees to help train AI tests workplace trust and interests. Workers might fear being replaced by AI after sharing their expertise, or losing their autonomy at work; therefore, in the next phase of AI development, companies that foster a culture of mutual trust and convince employees they will share in AI-driven gains will be more likely to succeed in the competition.