What are the risks from AI? 

This week we spotlight the 26th framework of risks from AI included in the AI Risk Repository: AI Verify Foundation. (2023). Summary Report for Binary Classification Model of Credit Risk. AI Verify Foundation. https://aiverifyfoundation.sg/downloads/AI_Verify_Sample_Report.pdf

Paper focus

This paper is a sample summary report demonstrating the use of the AI Verify Testing Framework. This framework is the basis for a toolkit used by organizations to systematically evaluate responsible AI practices during the deployment of their traditional and generative AI applications. The AI Verify Testing Framework was developed through consultation with companies from different sectors and is aligned with other international AI governance frameworks from the ASEAN, European Union, OECD, and USA.

Included risk categories

The AI Verify Testing Framework consists of 11 AI ethical principles, grouped into 5 areas:

1. Transparency on the use of AI and AI systems

• Transparency: Appropriate information is provided to those affected by the AI system

2. Understanding how AI models reach decisions

• Explainability: The AI system’s decisions and behaviors can be understood and interpreted
• Repeatability/reproducibility: The AI system’s results are consistent and can be replicated by a third-party given similar inputs

3. Safety and resilience of AI system

• Safety: The AI system does not result in harm to humans and known risks are mitigated
• Security: The AI system is protected from unauthorized access, disclosure, modification, destruction, or disruption
• Robustness: The AI system is resilient against attacks and can still function despite unexpected inputs

4. Fairness / no unintended discrimination

• Fairness: There is no unintended bias or inappropriate discrimination against individuals or groups
• Data governance: There are good governance practices throughout the data lifecycle

5. Management and oversight of AI system

• Accountability: There is proper management oversight and actors accountable for AI system development
• Human agency and oversight: The AI system does not decrease human ability to make decisions
• Inclusive growth, societal, and environmental well-being: There are beneficial outcomes for people and the planet

Key features of the framework and associated paper

• Implementation and risk mitigation focus: The framework is the basis for a comprehensive toolkit consisting of technical tests and/or process checks that are currently in use

⚠️Disclaimer: This summary highlights a paper included in the MIT AI Risk Repository. We did not author the paper and credit goes to the AI Verify Foundation. For the full details, please refer to the original website and sample report: https://aiverifyfoundation.sg/downloads/AI_Verify_Sample_Report.pdf.

Further engagement 

→ View all the frameworks included in the AI Risk Repository 

→ Sign-up for our project Newsletter