Skip to content

Generative AI for Research (UCT Course)

Key Ideas

  • A freely available, openly licensed 12-week postgraduate course (NQF Level 9) exploring how generative AI is transforming research practice — no prior programming, ML, or computer science background required.
  • Course materials were themselves created and enhanced with AI tools (primarily Claude), and an AI content disclaimer is included — making the course a meta-example of AI-assisted course design.
  • Twelve weeks span: LLM foundations and history, AI ethics (including Ubuntu and African relational ethics), literature review tools, AI writing workflows, data analysis and coding, multimodal AI, AI agents and MCP, and the future of AI in research.
  • Week 11 includes a dedicated unit on Africa's Sovereign AI Capacity: compute as the floor, data and language challenges in African model-building, and policy/institutions/talent.
  • An optional advanced track (adapted from Dominik Lukeš's Oxford e-Research Centre course on AI Agents for Reproducible Research) covers Claude Code as a research environment, reproducible project folders, CLAUDE.md, pre-registration, skills, and end-to-end reproducible workflows. Requires a paid Claude subscription and terminal access.
  • 54 PDFs of readings are curated across the 8 built weeks.
  • The hallucinated citation crisis gets its own dedicated module (Week 5).

Summary

"Generative AI for Research" is a comprehensive, openly available postgraduate course developed by Assoc. Prof. Jonathan Shock at the University of Cape Town. It takes a researcher through the full landscape of AI for academic work — from foundational LLM concepts (transformers, the current landscape as of May 2026) through applied domains including AI-assisted literature review, writing, data analysis, multimodal processing, and agentic tools. The course is notable for its attention to ethics beyond standard western frameworks: Week 4 covers the Ubuntu and "Just AI" ethical frameworks alongside conventional perspectives, and Week 11 examines Africa's specific context for sovereign AI development.

The course is carefully scaffolded for non-technical researchers. It acknowledges what AI does poorly (hallucinated citations get their own module) while building practical skills across all dimensions of the research workflow. The advanced track, adapted from Lukeš's Oxford materials, is a self-contained Claude Code tutorial covering reproducible research workflows, CLAUDE.md, skills, and end-to-end project management — an unusually complete resource for academics wanting to go beyond chatbot use.

Relevance to Economics Research

This course is one of the most complete freely available resources for training researchers to use AI tools across the full research lifecycle. For economists, the advanced Claude Code track directly addresses reproducibility, project scaffolding (CLAUDE.md, session logs, skills), and the dispositions needed to use agentic tools effectively. The week on data analysis and coding (natural language to code, AI-assisted data analysis, verification) is particularly relevant. The course also models good practice by disclosing AI use in course design — a template for how researchers might approach AI disclosure in their own work.