Bank of Korea report: 'Does AI adoption raise productivity?'

Correlation between time saved and output growth stands at zero

'AI has reached the efficiency stage but not yet the productivity stage'

An "AI productivity disconnect" — where reduced working hours fail to translate into higher output — has been found to exist among domestic workers. [Created with Copilot]
An "AI productivity disconnect" — where reduced working hours fail to translate into higher output — has been found to exist among domestic workers. [Created with Copilot]

By Kim Byeo-ri, The Herald Business

Workers in South Korea have cut their working hours by 3.8 percent through generative AI, but the time savings are not translating into actual production gains, the Bank of Korea said. Boosting productivity will require changes to organizational structures and workflow, the central bank added.

The Bank of Korea released the findings Sunday in a report titled "Does AI Adoption Raise Productivity? An Analysis of Effects in the First Three Years."

The report analyzed the impact of AI use on working hours and output among 5,512 employed workers between the ages of 15 and 64 nationwide, surveyed between May and June last year.

Workers who used generative AI reduced their average working hours by 3.8 percent — equivalent to roughly 1.5 hours per week based on a 40-hour workweek. If those time savings were fully converted into additional output, the report estimated the potential productivity gain would amount to around 1 percent.

However, the report said the reduction in working hours is not leading to actual production increases. The correlation coefficient between the rate of time savings from generative AI use and the rate of increase in work throughput was zero — a phenomenon the report described as an "AI productivity disconnect," in which shorter hours do not translate into higher output.

The report attributed this to saved time not being reallocated to higher-value tasks, and to insufficient changes in work procedures and organizational structures.

Productivity gains were observed, however, among groups with stronger performance incentives and greater work autonomy — including the self-employed, professionals and heavy AI users. This suggests that AI's impact may be determined less by the technology itself than by task structure and incentive systems, the report said.

The report went on to say that AI has entered the "efficiency" stage but has yet to transition to the "productivity" stage — a pattern it described as a typical J-curve phenomenon seen in the early adoption of general-purpose technologies. Potential benefits such as time savings accumulate in the period following a technology's introduction, but actual productivity gains are difficult to achieve without accompanying organizational and systemic change.

In practice, corporate AI adoption stood at just 9.6 percent as of 2024, far below the individual worker adoption rate of 51.8 percent — evidence that the spread of AI has yet to drive systematic changes in workflow at the firm level.

"To realize AI's productivity effects, it is important to redesign work procedures and organizational structures, reallocate roles, and build performance-based incentive systems," the report said. It also stressed the need for ongoing review of how AI is reshaping skill-formation pathways for youth.


kimstar@heraldcorp.com