Abstract
The increasing use of various types of probabilistic AI technologies in decision-making roles limits the usefulness of traditional business analysis and design (B&A) methods for documenting non-AI decision-makers and other aspects of probabilistic AI systems‚ including human backup decision- making roles․ Traditional UML and other business analysis tools support only binary decision outcomes and offer few tools for documenting the shared decision-making responsibilities of AI and human decision makers․ This paper proposes an AI-Socio B&A method that integrates traditional business analysis methods with three new tools:(1) a Confidence-Based Delegation Matrix offering three levels of AI decision-making authority (automatic decision making when confidence exceeds 80%‚ assisted decision making when confidence is between 50% and 80%‚ and human backup decision making when confidence is below 50%); (2) an amended use case format that offers additional documentation of confidence levels and bias-prevention techniques along with other information; (3) a process for obtaining and using feedback from human decision makers to improve the AI system․ A case study of an AI-Powered Job Applicant Screening and Short-listing System‚ including documentation of an extended use case (UC-02)‚ actors matrix‚ two design diagrams‚ and comparison with traditional B&A and UML documentation on five design and safety metrics‚ illustrates use of the new AI-Socio B&A method․ Advantages of AI-Socio B&A include reduced automation bias and inclusion of anonymization as a requirement and other tools to share work between human and AI decision makers