KISDI Finds Generative AI Improves Productivity by Reducing Inefficiencies in the Division of Labor (detail)

※ URL(Korean): https://www.kisdi.re.kr/bbs/view.do?bbsSn=115009&key=m2101113055776&pageIndex=1&sc=&sw=

KISDI Basic Research (25-08)
Productivity Analysis of Generative AI

KISDI Finds Generative AI Improves Productivity by Reducing Inefficiencies in the Division of Labor
.................................................................................................................................................
▲ Experimental analysis comparing four groups based on the use of generative AI and the presence of sequential division of labor
▲ Productivity improves by approximately 10% when generative AI enables tasks to be completed without sequential division of labor
▲ Even when the division of labor is maintained, generative AI offsets inefficiencies arising from task handoffs
▲ Generative AI enhances productivity by either eliminating or mitigating the inefficiencies of the division of labor
▲ Replacing human labor with AI does not necessarily lead to greater productivity gains

The Korea Information Society Development Institute (KISDI, President Sangkyu Rhee) recently published its Basic Research Report (25-08), *Productivity Analysis of Generative AI*, which examines how the use of generative AI affects productivity in work environments characterized by sequential division of labor.

The use of generative AI has expanded rapidly across a wide range of occupations and business functions, and previous studies have consistently found that AI can improve productivity. However, much of the existing research has focused on individual tasks, making it difficult to capture the complexity of real-world work environments where multiple tasks are often performed through coordinated workflows.

To address this limitation, the study replicated a sequential division of labor in an experimental setting and analyzed how generative AI influences productivity under such conditions. The experiment compared four groups defined by the presence or absence of generative AI and sequential division of labor, enabling the researchers to examine the inefficiencies associated with sequential task allocation, productivity gains under an AI-driven labor substitution scenario, and productivity gains when AI complements rather than replaces human labor.

The experimental task simulated a marketing workflow for a hypothetical product. The first task required participants to summarize a product description, while the second involved writing a persuasive sales email based on the summary. The output of the first task served as the input for the second, replicating a sequential workflow commonly found in organizational settings.

To evaluate productivity when generative AI replaces one unit of labor and eliminates the need for sequential division of labor, the researchers compared a control group consisting of two individuals completing the two tasks separately without AI with an experimental group in which a single individual completed both tasks sequentially using generative AI. The results showed that the AI-assisted individual produced higher-quality outputs while completing the tasks more quickly than the two-person team. These findings suggest that generative AI can enable a single worker to perform tasks previously divided among multiple workers, thereby reducing coordination inefficiencies and improving productivity.

The study also compared a group in which two individuals each used generative AI to complete separate tasks with a group in which one individual used generative AI to complete both tasks. No statistically significant differences were found in either the quality of the final output or the time required to complete the tasks. This suggests that the productivity gains achieved by eliminating the inefficiencies of sequential division of labor are broadly comparable to those achieved when AI enables a single individual to perform work previously carried out through task specialization.

Daehong Min, Fellow at KISDI, noted that while generative AI may eliminate the inefficiencies associated with the division of labor when it replaces human workers, similar efficiency gains can also be achieved when AI is used to support workers without replacing them. He emphasized that AI-driven labor substitution does not necessarily result in higher productivity than AI-assisted collaboration, and that this distinction should be carefully considered by both policymakers and labor market participants when making future decisions.

The report is available for download from the KISDI website (http://www.kisdi.re.kr)).

KISDI News

KISDI Finds Generative AI Improves Productivity by Reducing Inefficiencies in the Division of Labor