Responsible AI Question Bank: A Comprehensive Tool for AI Risk Assessment
The rapid growth of Artificial Intelligence (AI) has underscored the urgent need for responsible AI practices. Despite increasing interest, a comprehensive AI risk assessment toolkit remains lacking. This study introduces the Responsible AI (RAI) Question Bank, a comprehensive framework and tool designed to support diverse AI initiatives. By integrating AI ethics principles such as fairness, transparency, and accountability into a structured question format, the RAI Question Bank aids in identifying potential risks, aligns with emerging regulations like the EU AI Act, and enhances overall AI governance.
A key benefit of the RAI Question Bank is its systematic approach to linking lower-level risk questions to higher-level ones and related themes, preventing siloed assessments and ensuring a cohesive evaluation process. Case studies illustrate the practical application of the RAI Question Bank in assessing AI projects, from evaluating risk factors to informing decision-making processes.
1. Introduction
Since the emergence of ChatGPT and other large language models, AI has seen a surge in popularity in recent years. Fueled by remarkable advancements, companies across all industries are rapidly adopting, using, and developing AI systems to enhance their businesses. While this rapid adoption has driven significant market growth in the AI industry and generated excitement for its potential, it has also raised concerns about the responsible development and application of AI, such as hallucination, generating harmful content, and user’s overreliance.
A recent report found that while many companies view AI as a promising technology and actively pursue AI opportunities, only 10% of those surveyed have publicly announced their Responsible AI (RAI) policies. This suggests that many companies are still concerned about their RAI maturity level. Moreover, there have been numerous incidents involving AI across various sectors, raising concerns in areas such as privacy, bias, and safety.
2. Background and Literature Review
The discourse on RAI has gained significant traction in both industry and academia, underscoring the critical need for effective AI risk management to foster RAI practices. Despite the proliferation of studies and frameworks on responsible and safe AI, many remain abstract, lacking concrete measures for risk assessment and management. Our previous mapping study systematically analyzed 16 AI risk assessment and management frameworks worldwide to gain insights into current practices in managing AI risks.
Key trends and areas for improvement in AI risk assessment practices globally have been identified, informing the design and development of our question bank. Increasing global concern is evident as the number of AI risk assessment frameworks worldwide continues to rise, indicating a growing recognition of RAI approaches to assess and mitigate these risks.
3. Methodology
This study was conducted in five dedicated phases from 2022 to 2024. A systematic mapping study was performed to understand the state-of-the-art in AI risk assessment and select reference frameworks for developing the RAI question bank. The frameworks selected were scrutinized, and AI risk questions were synthesized to develop a comprehensive and holistic question bank for AI risk assessment.
The evaluation of our proposed question bank was carried out via two phases of case study involving eight AI projects and the ESG-AI framework development project. The case studies provided valuable insights and feedback for refining the question bank, ensuring its relevance and effectiveness in real-world applications.
4. RAI Question Bank Overview
The RAI Question Bank is a set of risk questions designed to support the ethical development of AI systems. It addresses the needs of various users, including C-level executives, senior managers, developers, and others. The necessity for such a question bank is evident, given the complex and multifaceted nature of AI risks that can arise from various sources, including biased data, algorithmic errors, unintended consequences, and more.
Organized into different levels (Level 1-3), the RAI Question Bank ensures a comprehensive evaluation by linking high-level principles with detailed, technical questions. This structured approach promotes a thorough and consistent approach to responsible AI development.
5. Case Studies
Two case studies were conducted to evaluate the application of the RAI Question Bank. The first case study focused on AI projects within a national research organization in Australia, assessing the ethical risks associated with each project. The second case study involved the integration of Environment, Social and Governance (ESG) factors into AI frameworks for investors. Feedback from stakeholders during these case studies proved invaluable for refining the question bank.
6. Compliance with Regulations
With the emergence of regulations such as the EU AI Act, companies face growing concerns about compliance with current and upcoming regulations, especially for high-risk AI systems. The RAI Question Bank can be utilized to examine RAI practices against legal requirements, creating a comprehensive compliance checklist.
This three-step approach includes identifying legal requirements, mapping them to the RAI Question Bank, and scoring the AI system against these mapped questions, enabling organizations to assess their compliance with corresponding requirements.
7. Conclusion
The RAI Question Bank represents a significant step forward in structured risk assessment of AI systems. By offering a valuable resource and tool for organizations, it promotes a thorough and structured approach to risk assessment, supporting the responsible use and development of AI systems that are not only innovative but also aligned with ethical principles and societal values.
