TLDR: The Human-AI RE Synergy Model (HARE-SM) is a new framework designed to integrate AI, specifically LLMs and NLP, into Requirements Engineering (RE) while maintaining human oversight. It addresses traditional RE challenges and AI-related issues like bias and explainability by emphasizing human-in-the-loop validation, transparency, bias mitigation, and trust calibration. A prototype, the Acceptance Criteria Assistant, demonstrates how AI can assist in generating and refining requirements, with all interactions logged for continuous improvement and ethical evaluation. The model aims to enhance RE efficiency and reliability through collaborative human-AI workflows.
Requirements Engineering (RE) is a critical phase in software development, acting as the bridge between what stakeholders need and the final software solution. Traditionally, this process has been manual, labor-intensive, and prone to errors, leading to issues like miscommunication, information overload, and inconsistencies. However, the landscape of RE is rapidly changing with the advent of Artificial Intelligence (AI), particularly Large Language Models (LLMs), Natural Language Processing (NLP), and generative AI.
These AI technologies offer transformative solutions to automate and enhance RE tasks, from eliciting requirements to analyzing and validating them. For instance, LLMs can extract structured requirements from unstructured data, while machine learning models can help classify, prioritize, and validate these requirements. Despite these advancements, integrating AI into RE is not without its challenges. Concerns such as algorithmic bias, a lack of explainability, and ethical considerations related to automation often hinder widespread adoption.
To address these critical issues, researchers have introduced the Human-AI RE Synergy Model (HARE-SM). This innovative conceptual framework is designed to integrate AI-driven analysis with essential human oversight, aiming to improve the entire requirements lifecycle. HARE-SM emphasizes the ethical use of AI through core principles like transparency, explainability, and bias mitigation, ensuring that AI acts as a responsible and effective assistant rather than a complete replacement for human expertise.
The HARE-SM is built upon four key principles:
Human-in-the-Loop Validation
This principle ensures that while AI assists with suggestions and automation, humans retain the final decision-making authority, preventing over-reliance on AI and mitigating automation bias.
Explainability & Transparency
AI-generated requirements are made interpretable and traceable, allowing stakeholders to understand the reasoning behind AI suggestions and fostering trust.
Bias Mitigation Strategies
The model incorporates mechanisms to detect and correct skewed outputs, addressing potential biases that can arise from LLM techniques or source data.
Also Read:
- Agentsway: A New Software Development Approach for AI Agent Teams
- PRISM: Unifying AI and Model-Driven Engineering for Verifiable Artifact Generation
Stakeholder Trust Calibration
HARE-SM includes feedback loops and adjustment mechanisms, allowing the AI’s behavior to be refined over time based on user input and specific stakeholder needs.
The model outlines a structured approach across several phases of RE:
- Requirements Elicitation: AI assists in gathering initial requirements using NLP, identifying ambiguities for human review.
- Requirements Analysis: LLMs and NLP process requirements to find redundancies and conflicts, with human experts validating these insights against project goals.
- Requirements Validation: AI aids in consistency checks and compliance analysis, while human experts ensure alignment with stakeholder expectations and ethical standards.
- Continuous Learning: The model learns from human feedback and project outcomes, continuously improving its performance and adaptability.
To operationalize HARE-SM, a functional prototype called the Acceptance Criteria Assistant has been developed. This tool allows users to input a user story, configure various AI models (like Flan-T5, Gemini, LLaMA-3), and compare their generated acceptance criteria side-by-side. The interface emphasizes modularity, transparency, and human oversight, enabling engineers to select, edit, or regenerate outputs. Crucially, all interactions, including human edits and preferences, are logged in real-time. This data is vital for analyzing common issues like bias and ambiguity, and for informing future design iterations and empirical validation of the HARE-SM workflow.
The research behind HARE-SM follows a multi-phase methodology, with preliminary studies and prototype development already completed. Future work involves fine-tuning AI models on curated datasets, experimenting with explainability techniques, and conducting rigorous empirical evaluations through stakeholder workshops and controlled case studies. The ultimate goal is to synthesize these findings into practical ethical guidelines for AI-augmented requirements engineering, ensuring fairness, transparency, and accountability.
By combining cutting-edge AI with responsible design, the Human-AI RE Synergy Model paves the way for a future where AI enhances human expertise in requirements engineering, leading to more efficient, accurate, and ethically sound software development processes. You can learn more about this framework in the full research paper: Towards Human-AI Synergy in Requirements Engineering: A Framework and Preliminary Study.
