Translation
Translated from the Spanish source, which is the normative text. Status: draft pending human review. Differences are resolved in the Spanish text.
Start with the documentation¶
Status: reading guide for ASILO 0.1, MILPA 0.1 and SCAFFOLD 0.1; see their editions.
1. Understand¶
- Choose a methodology by the question you want to solve.
- Learn its purpose: purpose of ASILO, purpose of MILPA or software engineering with artificial intelligence (AI).
- AI scope: a model, one assistant or several according to the task.
- Minimal vocabulary: what distinguishes principle, requirement, profile, example and implementation. Scales and history are consulted later.
2. Apply to a small task¶
Follow Apply the methods to your first project: identify the project, locate its sources, scope a task, review the proposal and keep what you learned. You can do it with your current tool; the guide does not require installing an SDK (software development kit) nor does it accept a new profile.
Open the detail only when you need it: manifest, structure, context, architecture and security or evaluation. The use cases show how the task changes, not the rules.
3. Learn the rules¶
- ASILO and MILPA principles: six principles for each of those methods.
- Requirements: verifiable obligations with their residual risk. MILPA 0.1 rules · ASILO 0.1 rules · SCAFFOLD 0.1 rules.
- Profiles: technological contexts and each profile’s approval state.
4. Look things up¶
- Reference: index of rules, vocabulary and records.
- Artifacts: what to record and which requirement it answers.
- External frameworks: OWASP (Open Worldwide Application Security Project), NIST (US National Institute of Standards and Technology) and MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) with versions.
- Normative language: what MUST, SHOULD and MAY mean.
5. Versions¶
- Versions and changes: editions in force, history and how the rules change.
- Method and software: what is documented here and what in each software repository.
A rule becomes accepted only with a recorded human decision and an identified version.