Magister of Artificial Intelligence in Science (Mag.AI-S)

Castalia Institute — Magisterium Design Document

Authoritative copy: part of the MyST project (myst.yml).

I. Purpose¶

Definition¶

Mag.AI-Science is a post-graduate program focused on AI-native scientific practice: hypotheses, measurement, reproducibility, instruments, and literature-scale evidence as executable systems, not slide decks. Curriculum delivery is built around executable articles: Jupyter Book / MyST publishes syllabi, lectures, and Jupyter notebooks from this repo; iNQspace is where those notebooks run, connect to data and tools, and accumulate inspectable lineage. See EXECUTABLE_ARTICLES.md.

Core thesis¶

Science is a world model with observables, uncertainty, and accountable lineage.

Therefore:

• understanding science = modeling what you claim, what you measure, and what could falsify you

• learning science = refining those models under evidence

• mastery = designing systems that survive replication and scrutiny

II. Degree philosophy¶

1. Hypotheses before hype¶

Students formalize assumptions, observables, and tests—then perturb before they scale claims.

2. Student as author¶

Students define ontology, choose assumptions explicitly, encode measurement and uncertainty, and own their models. AI does not replace judgment about ethics, authorship, or truthfulness.

3. AI as instrument¶

Parallel to world models, students build competence in evaluating AI in scientific workflows: grounding quality, version discipline, human-in-the-loop review, and red-teaming analysis pipelines. This is embedded across courses (see CURRICULUM_FULL.md, AI literacy spine).

4. Knowledge as metabolism¶

Students continuously ingest signals (instruments, literature, benchmarks), refine models, test assumptions, and ship measurable artifacts.

III. Executable articles (Jupyter Book + iNQspace)¶

The default student-facing unit of work is an executable article: Markdown + code + outputs that can be re-run, published as part of the course Jupyter Book, and executed or extended in iNQspace with logged lineage. Lectures explain; notebooks show the mechanism. Policy and conventions: EXECUTABLE_ARTICLES.md.

IV. Learning model¶

Traditional “template → slide → hope” is rejected in favor of:

hypothesis → instrument / simulation → executable article → log → critique → revise

Artifacts are inspectable: code, data lineage, and documented limits—not narrative alone.

V. Relationship to Mag.AI-Marketing¶

Both programs share Castalia’s Magisterium stack (iNQspace, SAMWISE, faculty defense, MCP policy). Mag.AI-Marketing centers audiences, messages, and channels; Mag.AI-Science centers measurement, reproducibility, and evidence chains. Cross-links live on the Magisterium hub: /magai/.

VI. Publication model¶

• Program hub (this repo root): orientation and reference.

• Course books: one MyST book per AINS-S#### under courses/, each including lectures and notebooks/*.ipynb as executable articles in the same build.

See magisterium.md for deploy paths.