AI Governance for Workflow Automation

Lecture 2 defines AI governance as a structured framework of policies and standards that guide the responsible management of AI systems throughout their lifecycle. It explores key frameworks, including Singapore's IMDA Model AI Governance Framework and the MAS FEAT Principles, and compares them with global standards like the EU AI Act and the NIST AI Risk Management Framework. The module also addresses the evolution of governance from traditional machine learning to generative AI, highlighting new risks such as hallucinations and copyright concerns. It emphasizes the need for multi-stakeholder collaboration and presents the business case for AI governance, focusing on risk reduction and increased trust.

Lecture 3 details the risks associated with AI agents and workflow automation. It categorizes a wide range of AI-specific risks, including those related to data quality and accuracy, security and privacy, compliance and legal issues, operations, ethics, and resilience. It also covers risks to the environment and to various groups of people, such as employees, service users, and third-party partners. The module provides a comprehensive risk register that outlines specific risks and corresponding mitigation strategies, such as human-in-the-loop interventions, input sanitization, and bias testing.

Lecture 4 focuses on establishing a clear governance structure for AI workflow automation. It identifies the key internal stakeholders, including data owners, business unit leaders, IT and data science teams, risk and compliance officers, and internal audit. The module introduces the concept of an AI Governance Review Committee, a multi-disciplinary body chaired by a senior executive, and the use of subcommittees for different governance aspects. A RACI (Responsible, Accountable, Consulted, Informed) matrix is presented as a practical tool for mapping out roles and responsibilities across the AI workflow lifecycle.

Lecture 5 breaks down the AI agent deployment lifecycle into distinct phases, from problem definition to retirement. It highlights the importance of integrating governance checkpoints at each stage to ensure accountability, transparency, and fairness. The module explores typical AI workflow system designs and introduces key risk control mechanisms and guardrails, such as human-in-the-loop interventions for sensitive tasks and defense-in-depth security measures. It also explains the use of "Data Cards" and "Model Cards" as best practices for data provenance and transparent documentation.

Lecture 6 provides real-world examples and case studies of AI governance in action, particularly within the APAC region. It highlights the financial services sector as a leading adopter of AI, showcasing how Singaporean banks like DBS, OCBC, and UOB have leveraged AI for areas like credit scoring and customer service while adhering to governance principles. The module also covers use cases in IT service support automation, such as ServiceNow's AI agents, and customer service. It concludes with a discussion of key success metrics and lessons learned from regional implementations, emphasizing the importance of strong governance foundations.

Lecture 7 addresses the critical importance of effective communication and legal awareness in AI governance. It outlines a framework for tailoring communication to various stakeholders, including executive leaders, technical teams, regulators, and the public. The module also covers support systems and programs for employees transitioning to AI-augmented roles. It uses high-profile AI failures, such as the DPD Chatbot and Air Canada legal cases, to illustrate common failure patterns like insufficient testing and lack of human oversight. The key takeaway is the need for proactive governance to prevent such incidents and define clear accountability.