AI-P07 — AI for Knowledge Management

Course Code: AI-P07  |  School: School of Artificial Intelligence  |  Cluster: Level 3 — Applied & Sectoral Practice

Level: Intermediate  |  Duration: 5 weeks · 18–22 learning hours  |  Language: English  |  Certificate: Professional Certificate (non-degree)  |  Format: Self-paced with AI support under human supervision

Overview

Knowledge management has a long record of expensive failure. Repositories are built, populated once, and abandoned, because contributing knowledge costs the contributor and benefits someone else. Retrieval-augmented systems change the retrieval half of that problem substantially. They do not change the contribution half at all, and an organisation that mistakes the first for the second will build an expensive index of its own out-of-date documents.

This course treats the retrieval architecture as the easier problem and the institutional problem as the real one. Learners audit what their organisation actually knows and where that knowledge sits, distinguish codifiable from tacit knowledge honestly, design a retrieval-grounded system with citation to source, and then design the contribution and curation arrangements that determine whether it stays alive.

The technical content covers what a retrieval-grounded system does, why grounding reduces but does not eliminate fabrication, how chunking and metadata affect answer quality, and how to evaluate a knowledge system against real questions rather than demonstrations. Access control receives particular attention, because a retrieval system that ignores document permissions is a data breach with a friendly interface.

Learning outcomes

On completion, a successful learner will be able to:

1. Audit organisational knowledge, distinguishing what is codified, what is codifiable and what is tacit.
2. Explain how a retrieval-grounded system works, and why grounding constrains but does not remove fabrication risk.
3. Design a knowledge architecture with source citation, metadata, currency indicators and permission inheritance.
4. Evaluate answer quality against a test set of real questions with known correct answers.
5. Design contribution and curation arrangements with named owners and review cycles.
6. Plan for decay: detecting stale content, retiring superseded documents, and preventing confident answers from obsolete sources.

Who this course is for

Knowledge and information managers, librarians, chief information officers, heads of policy and secretariat functions, and consultants implementing knowledge systems.

Prerequisites

AI-F03 recommended for the technical foundation. Experience of information or knowledge management practice. Learners need access to a real document corpus they are permitted to use.

Syllabus

Module 1 — What the organisation actually knows

Focus. Auditing knowledge rather than documents. The distinction between codified, codifiable and tacit knowledge, and the honest recognition that the most valuable knowledge in most institutions is in people.

Lessons. 1.1 Knowledge audit method. 1.2 Codified, codifiable, tacit. 1.3 Where the important knowledge actually lives. 1.4 Why previous repositories failed here.

Core reading. Ikujiro Nonaka & Hirotaka Takeuchi, The Knowledge-Creating Company (New York: Oxford University Press, 1995), chapters 2–3. Thomas H. Davenport & Laurence Prusak, Working Knowledge (Boston: Harvard Business School Press, 1998), chapters 1–2.

Deliverable. Knowledge audit with the codified, codifiable and tacit portions estimated and evidenced.

Module 2 — Retrieval-grounded systems, honestly described

Focus. How retrieval augmentation works, what grounding achieves, and the residual failure modes: retrieval of the wrong passage, plausible interpolation between passages, and confident answers from superseded sources.

Lessons. 2.1 Retrieval augmentation in plain terms. 2.2 What grounding fixes. 2.3 Residual failure modes. 2.4 Chunking, embedding and metadata effects on answer quality.

Core reading. Patrick Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks”, Advances in Neural Information Processing Systems 33 (2020). Maryam Alavi & Dorothy E. Leidner, “Knowledge Management and Knowledge Management Systems: Conceptual Foundations and Research Issues”, MIS Quarterly 25, no. 1 (2001): 107–136.

Deliverable. Technical brief explaining the proposed architecture to a non-technical executive, including its limits.

Module 3 — Architecture, citation and permission

Focus. Designing for traceability: every answer cites its source passage; every document carries currency and authority metadata; retrieval respects the permissions of the underlying document rather than the querying interface.

Lessons. 3.1 Source citation as a hard requirement. 3.2 Currency and authority metadata. 3.3 Permission inheritance and the breach risk. 3.4 Handling contradictory sources.

Core reading. Timnit Gebru et al., “Datasheets for Datasets”, Communications of the ACM 64, no. 12 (2021), for the documentation discipline applied to corpora. NIST, AI Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (2023), MAP function.

Deliverable. Architecture specification with metadata schema and permission model.

Module 4 — Evaluation against real questions

Focus. Building a test set of questions the organisation actually asks, with known correct answers, and measuring the system against it. Why demonstrations mislead and adversarial questions are essential.

Lessons. 4.1 Constructing a question test set. 4.2 Scoring correctness, groundedness and citation accuracy. 4.3 Adversarial and out-of-scope questions. 4.4 Reporting results without inflation.

Core reading. Ziwei Ji et al., “Survey of Hallucination in Natural Language Generation”, ACM Computing Surveys 55, no. 12 (2023).

Deliverable. Evaluation report on at least thirty real questions, including out-of-scope cases.

Module 5 — Contribution, curation and decay

Focus. The institutional half. Who contributes and what it costs them, who curates and under what authority, how staleness is detected, and how superseded documents are retired so the system does not answer confidently from them.

Lessons. 5.1 Contribution incentives that survive the first quarter. 5.2 Curation authority and named owners. 5.3 Detecting staleness. 5.4 Retirement, supersession and communities of practice.

Core reading. Etienne Wenger, Communities of Practice: Learning, Meaning, and Identity (Cambridge: Cambridge University Press, 1998), chapters 1 and 4. David J. Snowden & Mary E. Boone, “A Leader’s Framework for Decision Making”, Harvard Business Review 85, no. 11 (November 2007): 68–76.

Deliverable. Final submission: architecture specification, evaluation report and a governance plan covering contribution, curation and retirement.

Assessment

Pass mark 70 per cent. All assessed components must be attempted. Every mark in this course is issued by a human assessor; no assessment outcome is generated automatically.

Rubric criteria

Each assessed artefact is marked against four criteria at four levels (distinction, pass with merit, pass, fail).

1. Honesty about tacit knowledge: does the audit resist the temptation to treat everything valuable as codifiable?
2. Traceability: does every answer path lead to a citable source, and does the permission model hold?
3. Evaluation quality: is the test set drawn from real questions, including ones the system should refuse?
4. Sustainability: are contribution and curation arrangements specific, owned, and plausible beyond launch?

Reading list

Core. Ikujiro Nonaka & Hirotaka Takeuchi, The Knowledge-Creating Company (New York: Oxford University Press, 1995). Thomas H. Davenport & Laurence Prusak, Working Knowledge: How Organizations Manage What They Know (Boston: Harvard Business School Press, 1998).

Peer-reviewed. Maryam Alavi & Dorothy E. Leidner, “Knowledge Management and Knowledge Management Systems”, MIS Quarterly 25, no. 1 (2001). Patrick Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks”, NeurIPS 33 (2020). Ziwei Ji et al., “Survey of Hallucination in Natural Language Generation”, ACM Computing Surveys 55, no. 12 (2023). Timnit Gebru et al., “Datasheets for Datasets”, Communications of the ACM 64, no. 12 (2021).

Practice. Etienne Wenger, Communities of Practice (Cambridge: Cambridge University Press, 1998). David J. Snowden & Mary E. Boone, “A Leader’s Framework for Decision Making”, Harvard Business Review (November 2007). NIST, AI Risk Management Framework (AI RMF 1.0) (2023).

All items are published works identifiable by author, title and publisher. Learners obtain them through an institutional library or the publisher. The Academy does not distribute copyrighted texts.

Academic integrity and use of AI

Generative tools may be used in producing assessed work under three conditions. Use must be disclosed in a short statement appended to each submission, naming the tool and the task it performed. Any factual or technical claim originating from a generative tool must be verified against a citable source before it enters assessed work, and the verification must be evidenced. The analytical judgement in each artefact must be the learner’s own and must be defensible in a short follow-up. Corpora used in assessed work must be material the learner is permitted to process. Confidential documents belonging to third parties may not be uploaded to any external service.

Instructor: pending owner confirmation. Pricing: pending owner approval. Reference list verified against publisher records; any later addition is marked for verification before publication.