How can automation be used without losing control? Kai Hoberg of Kuehne Logistics University (KLU), Christina Imdahl of Eindhoven University of Technology and William Schmidt of Emory University sought an answer to this question. Their work was published in the journal Production and Operations Management.
The researchers propose not automating everything indiscriminately, but doing so selectively. The idea is to trust the algorithm with decisions where human involvement changes almost nothing, and to leave people the cases where their experience helps improve the result. At the same time, a sufficient share of manual decisions must be retained so that the algorithm can adapt to changing conditions.
This approach makes it possible to combine the quick gains from automation with continuous improvement of artificial intelligence (AI).
To implement it, the researchers developed four steps. First, it is necessary to determine when human corrections lead to better decisions. Then the model learns to predict in which cases an employee's experience will be useful. After that, some decisions are automated and the rest are left to people. Finally, human corrections are fed back into the system so that the algorithm becomes more accurate over time.
"It is important not simply to give complex decisions to people, but to predict which decisions specifically allow a person to improve the system," Imdahl emphasizes. According to her, in the experiment conducted, about 84% of decisions were automated.
Hoberg advises businesses to introduce automation gradually - starting by learning to measure the benefit of human intervention. This will provide a basis for understanding which tasks can be handed over to the machine and where employees' experience is still needed.
The study analyzed a large dataset of order decisions collected over four years at a major commercial vehicle manufacturer with annual revenue of more than EUR 4 billion. The researchers looked for situations in which a human made a better decision than the algorithm. At the same time, they were interested not in the debate over "who is better - humans or AI," but in a model where both work as partners in a dynamic system. The goal is to free employees from routine work that brings almost no benefit while at the same time allowing AI to develop through feedback from people.