AI behavioral intent implementation system and method in systems applying reinforcement learning

XAII-based Intent Implementation Technology for Pre-explaining the Behavioral Intent of Reinforcement Learning AI

This technology concerns explaining the behavioral intent of AI agents to users in systems applying reinforcement learning and reflecting those results in system operation. Specifically, it is a technology that enhances the explainability of reinforcement learning-based decision-making by outputting the AI's intent before it acts, through an XAII provision module.

Existing reinforcement learning systems made it difficult for users to understand the purpose or intent of an agent's chosen actions in advance, which could limit reliability, controllability, and field applicability. To address this, this technology proposes a method that places an XAII provision module and an AI mode provision module in the control unit of an AI terminal, providing behavioral intent information to the input/output unit, and reflecting user feedback or result values.

Accordingly, this technology enables pre-explanation of AI agent actions, user confirmation, and adjustment of intervention levels, making it applicable to safety-critical automation systems, robots, defense and operational support systems, and explainable AI services.

Key Features

• Implemented to explain the behavioral intent of a Reinforcement Learning AI agent to the user before it acts
• Supports the output of intent information and the reflection of user responses via an XAII provision module
• Controls action execution based on the AI intervention level by linking the AI mode and inference module
• Applicable to improving the reliability of explainable AI, robot control, and high-risk automation systems