Course Objective & Scope
Historical inequalities in training data or subtle stereotypes in large language models can quietly create systematic discrimination in lending, pricing, hiring and customer service—while opaque 'black-box' decisions fail regulatory scrutiny. This course equips business, risk and compliance leaders to govern algorithmic bias, fairness and explainability (XAI).
It covers how bias enters through data, labels, proxy variables and feedback loops, the mathematical trade-offs between competing fairness definitions, and how to translate local/global explainability into stakeholder-appropriate explanations. Participants produce a Fairness & Explainability Assessment Plan for an enterprise use case.
Who Should Attend?
- Leaders in HR, retail banking/lending, insurance, marketing and customer operations
- Compliance, legal, ethics, internal control and Model Risk Management (MRM) specialists
- AI product managers and business analysts working with data science teams
- Teams handling customer appeals and regulatory inquiries
Key Takeaways
- Identify historical, representation, measurement and proxy bias across the AI lifecycle
- Choose the right fairness metric (e.g., demographic parity vs equal opportunity) for a specific business context
- Specify bias evaluation tests for both tabular ML models and generative LLMs
- Balance model accuracy against interpretability using intrinsic or post-hoc XAI methods
- Design stakeholder-specific explanations for customers, internal reviewers and auditors
Curriculum
Where Algorithmic Bias Comes From Across the Lifecycle
- Historical bias, sampling gaps, label bias and self-reinforcing feedback loops
- The proxy variable trap: why dropping protected attributes does not guarantee fairness
- Cultural, linguistic and demographic stereotypes in Large Language Models
Fairness Definitions and Executive Trade-Offs
- Plain-language guide to demographic parity, equal opportunity and equalized odds
- Why no model can satisfy every mathematical definition of fairness simultaneously
- Selecting and documenting the justifiable fairness criterion for lending, HR or triage
Governing Bias Audits and Mitigation Strategies
- Disaggregated performance evaluation across demographic and customer subgroups
- Pre-processing, in-processing and post-processing mitigation options
Explainable AI (XAI): Opening the Black Box
- Inherently interpretable models versus post-hoc explanations (SHAP, LIME, counterfactuals)
- Source citations and retrieval traces in RAG versus LLM post-hoc rationalisations
- Avoiding the trap of plausible-sounding but unfaithful LLM explanations
Designing Stakeholder Explanations and Appeal Workflows
- Tailoring explanations for affected customers, internal operators and regulators
- Building an actionable adverse-action notice and human appeal process
Other Courses in This Category
-
Design an internal control framework and audit programme for AI systems covering data, model, prompt/RAG, guardrail and oversight controls, evidence collection and control effectiveness testing.
- Live Course
- 5 modules
-
Establish end-to-end AI traceability across prompts, retrieved context, model versions and human overrides, and build an AI incident response playbook for hallucinations, drift, jailbreaks and data leaks.
- Live Course
- 5 modules
-
Implement an AI Management System (AIMS) under ISO/IEC 42001 Clauses 4–10 and Annex A/B controls, define scope, draft the Statement of Applicability (SoA), and prepare the organisation for audit.
- Live Course
- 5 modules
-
Plan, conduct and report internal audits of an ISO/IEC 42001 AI Management System (AIMS) aligned with ISO 19011, verify Clause 4–10 and Annex A evidence, and write clear nonconformity reports.
- Live Course
- 5 modules
-
Operationalise the NIST AI Risk Management Framework (Govern, Map, Measure, Manage) and Playbook, implement TEVV (Testing, Evaluation, Verification and Validation) practices, and build a target AI risk profile.
- Live Course
- 5 modules
-
Integrate ISO/IEC 42001 (AIMS) into an existing ISO/IEC 27001 ISMS without duplicating governance overhead: align Harmonized Structure clauses, unify risk registers and SoAs, and run combined audits.
- Live Course
- 5 modules
Get in touch
