PF-ENV: Extending Problem Frames for AI-Enabled System Requirements Environment Modeling

AI models learn patterns and abilities from big data rather than relying on code logic. AI models exhibit a certain tolerance to varying inputs in open environments, enabling AI components to address issues in dynamic environments. However, the complexity of objects and the ever-changing conditions in open environments often result in AI-enabled system failing to meet user expectations regarding performance. Consequently, during the early development stages of AI-enabled system, it is essential to thoroughly consider both their actual environment and the environment required for effective functioning. In response, this paper introduces the PF-ENV, a model of the actual environment, and the required environment of AI-enabled system through extended Problem Frames and model-driven engineering (MDE). The paper demonstrates its application through a case involving the AI-enabled systems in Autopilot. Finally, the practical implementation of PF-ENV demonstrates its effectiveness in modeling the actual environment and environmental requirements.