Bridging Theory and Practice: Ethical AI User Story Generator (EAI-USG) - A Tool for Translating Ethical AI Requirements into Ethical User Stories

Context: Rapid advances in AI have intensified high-profile ethical failures (e.g., bias, privacy abuse, deepfakes), while existing guidelines remain abstract and hard to operationalize in software practice. 

Goal: Present the Ethical AI User Story Generator (EAI-USG), a tool that translates high-level ethical AI requirements into Ethical User Stories (EUS) to embed ethics early in requirements engineering.

Method: We conducted a Systematic Literature Review to map gaps, designed the artifact under a Design Science Research approach, and implemented LLM-based generation enhanced with Retrieval-Augmented Generation (RAG) and fine-tuning (QLoRA). We evaluated candidate models (Falcon, BERT/RoBERTa, Mistral) and validated EAI-USG with 30 practitioners via a mixed-method study (Likert + open-ended feedback).

Results: Mistral-7B offered the best balance of quality and cost. Practitioners rated the generated EUS as clear, coherent, useful, and efficient, and most would recommend the tool. A noted limitation was uneven coverage of some principles (e.g., sustainability) linked to dataset gaps.

Conclusion: EAI-USG demonstrates a practical path to operationalize AI ethics by converting abstract principles into actionable user stories. Future work will broaden principle coverage, incorporate objective metrics (e.g., coverage mapping, time-on-task), improve usability (GUI), and assess adoption in real development teams.