Mag.AI-S — Full curriculum (working outline)
Castalia Institute
This document is the working program reference for Magister of Artificial Intelligence in Science (Mag.AI-S). Course books under courses/AINS-S####/ are authoritative for cohort-specific faculty notes; this file tracks program-level structure, themes, and the AI literacy spine.
Executable articles & iNQspace (program-wide)¶
Mag.AI-S does not separate “theory slides” from “lab notebooks” as the primary delivery model. The curriculum is expressed through executable articles: Jupyter notebooks (and linked MyST prose) that are built into each course’s Jupyter Book and run in iNQspace for real data, tools, and lineage.
Jupyter Book — public, structured site per course (
myst.yml,SYLLABUS.md,lectures/*.md,notebooks/*.ipynb).iNQspace — where execution, credentials, SAMWISE, and artifact traceability attach.
Read EXECUTABLE_ARTICLES.md for conventions. AINS-S6005 — Literate Computation & Experiment Logs makes this explicit in the title; every other module still uses notebooks as the default vehicle for artifacts unless the syllabus states otherwise.
AI literacy spine (program-wide)¶
Across all terms, every course engages with technical AI literacy as appropriate to the artifact:
Evaluation — metrics that match the claim; calibration; failure analysis
Grounding & retrieval — when AI consumes literature or domain corpora
Generative workflows — versioning, provenance, and human review gates
Scoped tooling (MCP) — least-privilege access to instruments, literature APIs, lab systems
This is not a separate “AI minor”; it is embedded in measurement, reproducibility, inference, and governance modules.
Term I — World models & integrity¶
| Code | Title | Theme |
|---|---|---|
| AINS-S6001 | Scientific World Models & Executable Hypotheses | Smallest honest world: assumptions ↔ observables ↔ tests |
| AINS-S6002 | Measurement, Uncertainty & Calibration | Error budgets and honest limits |
| AINS-S6003 | Computational Reproducibility & Provenance | Containers, seeds, software bills of materials |
| AINS-S6004 | Data Systems, Documentation & Lineage | Metadata, schemas, audit trails |
| AINS-S6005 | Literate Computation & Experiment Logs | Notebooks and logs as reviewable artifacts |
| AINS-S6006 | Ethics, Integrity & Responsible Conduct | Authorship, misuse, and integrity boundaries |
Term II — Evidence & systems at scale¶
| Code | Title | Theme |
|---|---|---|
| AINS-S6101 | Experiment Design & Causal Reasoning | Identification and design for strong claims |
| AINS-S6102 | Statistical Learning & Model Evaluation | Generalization and evaluation discipline |
| AINS-S6103 | Simulation, Generative Models & Synthetic Data | Synthetic data and simulators under constraints |
| AINS-S6104 | Literature Intelligence, RAG & Evidence Synthesis | Retrieval and synthesis at scale |
| AINS-S6105 | Instruments, Signals & ML for Measurement | Signals, noise, and learned measurement |
| AINS-S6106 | Open Science, Collaboration & Review Stress Tests | Adversarial review and collaboration risk |
Term III — Strategic science¶
| Code | Title | Theme |
|---|---|---|
| AINS-S6201 | Competing Hypotheses & Model Comparison | Comparison and strategic model choice |
| AINS-S6202 | Research Strategy as Policy Under Uncertainty | Policies mapping states to actions |
| AINS-S6203 | Portfolio & Resource Allocation for Research Programs | Budget and risk across initiatives |
| AINS-S6204 | Autonomous Lab & Instrumentation Systems | Governed autonomy in lab workflows |
| AINS-S6205 | Safety, Governance & Control in Scientific AI | Oversight for scientific AI systems |
| AINS-S6206 | Magisterium Thesis (Science) | Capstone deployment, validation, defense |
Prerequisites¶
Default path is sequential within each term unless faculty publish an exception list. Cross-track recognition (e.g. from Mag.AI-Marketing) is not automatic—faculty review.
Artifacts¶
Each course expects a minimum viable artifact appropriate to the module (simulation, pipeline, benchmark report, or governed tool integration). The default delivery is an executable article (notebook + narrative) in the course’s Jupyter Book, with execution and lineage in iNQspace. Exact rubrics are cohort-specific; the constant requirement is inspectable lineage in iNQspace.