Automation of Human Decision Making by Using Reinforcement-learning for Office Work with PC

AI technology is a key which improve diversified and complex business with saving personnel. However, the technologies considered so far have limitations in supporting human decision-making. In this paper, we aim to expand decision support by automating human decision making without expertise or analytical techniques by machine learning human actions for office operations. Reinforcement learning acquires optimal actions through interaction between the agent and the environment, and has a high affinity for learning operations on a computer. However, in actual workflow, it is essential to design rewards for reinforcement learning and to avoid interfering with workers and business systems. Therefore, we propose a new method for machine learning without the need to design rewards and without invading actual work environments, by using the operation logs collected from the personnel's computers and assigning rewards to human interactions with computers as correct answers.