Generative artificial intelligence is spreading rapidly and reducing workers' working hours, but these efficiency gains are not translating into productivity improvements for organizations as a whole, according to an analysis by the Bank of Korea. While AI allows individuals to complete tasks faster, work processes and organizational operations remain unchanged, limiting the effect on output.
The Bank of Korea (BOK) made the assessment in a BOK Issue Note titled "Does AI Adoption Boost Productivity?" released Monday. A survey of 5,512 employed people nationwide conducted by the research team in May and June last year found that 51.8% of Korean workers use generative AI in their jobs. That is about eight times faster than the pace of adoption during the spread of the internet.
AI proved effective in reducing actual working hours. The average working hours of workers who use AI fell 3.8%. Based on a 40-hour work week, that amounts to saving about 1.5 hours per week. The time-saving effect was largest among professionals (2.8%), followed by office workers (1.9%) and managers (1.5%). The effect was greater for cognitive tasks such as creating educational materials, statistical analysis and software development.
But output did not increase in proportion to the faster work. When the research team analyzed the relationship between the rate of working-hour savings and the rate of increase in workload, the correlation coefficient was 0.00. This means that while work was completed faster, the volume of work handled did not increase meaningfully. The research team explained that the saved time was likely not reallocated to higher-productivity tasks but scattered into waiting time or idle time.
In reality, the AI effect was concentrated in certain groups. For the self-employed, when AI reduced working hours by 1 percentage point, their workload increased 1 percentage point more than for wage earners. Young workers showed a productivity gain 0.6 percentage points higher than older workers, and professionals were 0.7 percentage points higher than office workers. High-intensity users who actively use AI also saw a greater output increase than low-intensity users.
By contrast, AI use did not lead to productivity gains for many wage earners. The research team analyzed that in environments where rewards for performance are unclear or work autonomy is low, there is insufficient incentive to put saved time into additional production. It particularly noted that corporate-level AI utilization stood at just 9.6%, far below the worker utilization rate of 51.8%. While individuals are using AI, companies have not changed their work processes and organizational structures to fit AI.
The BOK defined this as an "AI productivity disconnect" phenomenon. This means that while AI is improving the efficiency of individual tasks, it has not yet reached the stage of organizational-level productivity innovation. "AI's spread remains at the level of specific tasks, and a redesign of the overall work flow has not followed," the BOK said. "AI's productivity effects can begin to appear in earnest only when organizational operations and reward systems change together."
However, the research team did not view the current situation as a limitation of the technology. When general-purpose technologies such as electricity and the internet spread in the past, productivity effects did not appear initially but then expanded sharply later, a "J-curve" phenomenon. The BOK forecast that for AI as well, productivity improvement effects could begin in earnest once the process of combining it with corporate work structures advances.
