AI Governance and Ethics Quiz

AI Governance

• AI and machine learning (ML) systems make decisions faster than humans
• AI should be designed to support organizational goals
• AI must meet ethical standards and not break rules
• Trustworthy AI works responsibly and ethically
• AI is part of information technology (IT), but it's evolving rapidly and less clearly defined than other branches of IT
• AI is treated like an emerging technology due to risks and a lack of established best practices

Need for AI Ethics

• AI will take over some decision-making from humans
• AI systems need to make ethical decisions
• AI ethics prevents unfair AI decisions and requires humans to retain control as AI systems make more decisions
• AI systems must not be allowed to act on their own
• Ensure AI systems do not cause harm
• Keep AI human-centered without limiting benefits
• Ensure AI outcomes stay within appropriate boundaries, such as healthcare, law, and engineering

AI Risks

• Unexplainability: Difficult to understand how AI makes decisions; cannot explain its decision-making process
  • Explainability:
    • Shows how AI makes decisions
    • Builds trust in system performance
    • Helps users accept AI and trust its performance
    • Reduces the risk of unexpected behavior
  • Mitigation:
    • Spend more to understand AI systems
    • Implement human oversight to check AI decisions
• Bias/discrimination: AI results can favor specific groups (gender, race, etc.) due to biased training data
  • Causes: Biased training data
  • Fix: Use balanced training data with equal representation of groups

AI Ethics (Body of Knowledge)

• Developed by SCS to guide AI use
• Designed for AI solution providers, businesses, and consumers
• Guiding principles of AI to promote trust:
  • Transparent, fair, and explainable
  • Human-centric, protecting well-being and safety

AI Ethics BOK Approach to AI

• Internal governance: Implementing structures for accountability for AI decisions and management of risks
• Human-centricity: Setting acceptable risk levels for AI usage; deciding how much human involvement is needed in AI decisions
• Operations management: Building, choosing, and maintaining AI models for data management and ensuring auditability
• Stakeholder communications: Effectively communicating with stakeholders

AI Model Governance Framework

• Practical guidance for organizations to address ethical and governance issues in AI usage
• Not limited to specific algorithms or technologies; applicable to all AI methods
• Applicable across industries and organization sizes
• Objective: Build confidence in responsible AI usage and gain stakeholder trust
• Guiding principles: ensure decision-making is explainable, transparent, and fair; human-centric AI
• Internal Governance:
  • Define roles and responsibilities for managing AI systems
  • Assign specific responsibilities for AI usage
  • Manage AI risks (including in risk management systems)
  • Align with ethical standards
• Operations Management:
  • Minimise bias
  • Balance needs and obligations
  • Focus on data quality
  • Define data origin, use, and transformation (data lineage and provenance)
• Stakeholder Interaction and Communication:
  • Create and share policies to ensure stakeholder awareness
  • Initiate efforts to encourage feedback and improve communication with stakeholders
  • Provide tailored communication styles for different audiences
  • Implement opt-out options, as appropriate

Human Involvement Model

• Depending on the context, AI may be used in tandem with human input or operate independently.
• Human in the loop: (e.g., medical diagnosis, operational tech) Human oversight essential
• Human out of the loop: (e.g., personal assistants) Automatic AI operation; limited human involvement, low impact
• Human over the loop: AI operates, but human intervention may be required

Risk Impact Matrix

• Assesses risk based on potential harm, reversibility, and likelihood
• Defines the level of human involvement in the AI system
• Green: Human out of the loop (low risk, minimal harm)
• Yellow: Human over the loop (some risk, potential harm)
• Orange: Human in the loop (moderate risk, higher harm needs human intervention)
• Dark red/red: Human involvement is inadvisable; high impact, irreversible harm; human intervention needed

Ethical Principles and Requirements

• Respect for human autonomy: Maintain human freedom and enhance skills
• Prevention of harm: Prevent physical and mental harm to humans and the environment
• Fairness: Ensure fair distribution of benefits and costs (including in the case of existing inequalities)
• Explicability: Understanding AI's decisions and capabilities
• Human agency and oversight: Respect human rights and allow control over AI systems
• Technical security and safety: Secure and reliable AI functionalities
• Privacy and data governance: Protection of data and guarantees of quality
• Transparency: Clear, explainable AI to everyone
• Diversity, non-discrimination, fairness: Avoiding biases in favor of specific groups
• Societal/environmental well-being: Promotion of sustainability
• Accountability: Transparent processes, reducing harm, and offering solutions to problems

Data Governance & Bias

• Data bias:
  • Selection bias: Data used may not be representative (incl. potentially biased sampling).
  • Measurement bias: Data may be influenced by the measuring device and result in biased results (favoring some results over others).
• Training, testing, and validation: Using separate data sets—avoid overfitting
• Updating: Regular checks on data for accuracy, quality, and relevance. New data can help improve existing systems.