Building Trustworthy AI Systems: Transparency, Explainability, and Control with ISO/IEC TR 24028
1h 18mIntermediate2025-01-29
Authors
Lyron Andrews
Course details
As the global regulatory and legal environments work to catch up to the rapidly changing and evolving world of AI, many mandates have started to require proof of trustworthiness related to developing and deploying AI systems. In this course, instructor Lyron Andrews outlines how you can use the guidance from the ISO/IEC TR 24028 standard to leverage AI tools according to principles of trustworthiness, which include explainability, transparency, and controllability.
Learning objectives
Analyze approaches to establishing trust in AI systems through transparency, explainability, and controllability.
Evaluate engineering pitfalls, threats, and risks associated with AI systems, along with potential mitigation techniques.
Apply methods to assess and improve the availability, resiliency, reliability, accuracy, safety, security, and privacy of AI systems.
Explain the concepts of trustworthiness in AI, including explainability, transparency, and controllability, in the context of evolving global regulatory and legal environments.
Design AI systems with careful consideration of explainability, transparency, and controllability to enhance their trustworthiness.
Learning objectives
Analyze approaches to establishing trust in AI systems through transparency, explainability, and controllability.
Evaluate engineering pitfalls, threats, and risks associated with AI systems, along with potential mitigation techniques.
Apply methods to assess and improve the availability, resiliency, reliability, accuracy, safety, security, and privacy of AI systems.
Explain the concepts of trustworthiness in AI, including explainability, transparency, and controllability, in the context of evolving global regulatory and legal environments.
Design AI systems with careful consideration of explainability, transparency, and controllability to enhance their trustworthiness.
Skills covered
Responsible AIEthics and LawData PrivacyGovernance, Risk, and ComplianceArtificial Intelligence (AI)CybersecurityData ScienceBusiness Analysis and StrategyOne-Off
Concepts
Introduction
- Building trustworthy AI systems
Existing Frameworks Applicable to Trustworthiness (Clauses 5-5.5)
- Recognition of layers of trust (Clause 5.1)
- Software and data quality standards (Clause 5.2)
- Application of risk management (Clause 5.4)
Stakeholders and Recognition of High-Level Concerns (Clauses 6-7.2)
- Stakeholders, types, assets, and values (Clauses 6-6.4)
- Responsibility, accountability, and governance (Clause 7.1)
- Safety (Clause 7.2)
Vulnerabilities, Threats, and Challenges (Clauses 8-8.10)
- AI-specific security threats (Clauses 8-8.2)
- AI-specific privacy threats (Clause 8.3)
- Bias, unpredictability, and opaqueness (Clauses 8.4-8.6)
- Systems specification and implementation challenges (Clauses 8.7-8.8.4)
- Challenges related to use (Clauses 8.9-8.10)
Mitigation Measures
- Transparency (Clauses 9-9.2)
- Explainability (Clauses 9.3-9.3.7)
- Controllability, bias, and privacy (Clauses 9.4-9.6)
- Reliability, resilience, and robustness (Clauses 9.7-9.9)
- Testing (Clauses 9.10-9.10.2.7)
- Evaluation (Clauses 9.10.3-9.10.5)
- Use and applicability (Clauses 9.11-9.11.4)
Conclusion
- Continue building trustworthy AI systems