Model Documentation
Records of model development, assumptions, and validation.
How the laws define this
Regulatory definitions universally agree that model documentation consists of structured technical records, such as model cards or datasheets, detailing training datasets, training processes, evaluation metrics, and system limitations to enable transparency, oversight, and reproducibility. The instruments diverge primarily regarding specific compliance mechanisms and application scopes. The Política Nacional de Inteligencia Artificial (CONPES 4144 - National Policy for Artificial Intelligence) focuses on publishing non-sensitive descriptions specifically for public administration systems, whereas the Agenda Nacional de Inteligência Artificial (National AI Agenda / ANI) extends requirements to include risk assessments and mitigation measures, particularly for publicly funded models. Meanwhile, the Malaysia Artificial Intelligence Consortium (MAIC) mandates reproducible artefacts like training logs specifically for conformity assessment and provenance tracking within consortium activities. Lastly, Law No. 31814 — Law that promotes the use of Artificial Intelligence in favor of the country's economic and social development uniquely mandates recording system architecture alongside ongoing maintenance and monitoring plans to support explainability.
Synthesised from the 4 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first. Showing 3 of the 4 instruments that define this term; all are listed under Definitions below.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| Colombia | National Policy for Artificial Intelligence | Adopted | Recommends publication of non-sensitive model documentation particularly for AI systems used in public administration. |
| Malaysia | Malaysia Artificial Intelligence Consortium (MAIC) | Adopted | Describes reproducible artefacts for conformity assessment, provenance tracking and transparency of AI models used in consortium activities. |
| Portugal | National Artificial Intelligence Agenda | Draft | Emphasizes support for legal and ethical compliance, particularly for publicly funded models. |
What it triggers
- In Colombia's National Policy for Artificial Intelligence, it calls for publication of non-sensitive model documentation, particularly for systems used in public administration.source
Definitions (4)
Technical and non-technical descriptions (e.g., model cards) that disclose an AI system's purpose, training data characteristics, performance metrics, limitations, and audit trails to enable transparency, oversight and reproducibility. The policy requires publication of non-sensitive model documentation, particularly for systems used in public administration.
The set of reproducible artefacts—such as data descriptions, training logs, evaluation results and model cards—required by MAIC for conformity assessment, provenance tracking and transparency of AI models used in consortium activities.
Structured technical and explanatory documentation that records a model's architecture, training procedures, datasets used, performance metrics, limitations, intended use cases, and maintenance/monitoring plans to enable oversight, reproducibility and explainability.
Model documentation refers to structured records (e.g., model cards, datasheets) that describe datasets, training processes, evaluation metrics, risk assessments and mitigation measures for AI models, intended to ensure transparency, enable audits, and support legal and ethical compliance—particularly for publicly funded models.
Not to be confused with
Model validation focuses on processes to verify model performance, while model documentation records development details like assumptions and metrics.
Model transparency concerns clarity about operation, inputs, outputs and limitations, whereas model documentation provides structured records of development and validation.
Transparency measures encompass broader documentation and registries for AI system visibility, while model documentation specifically records model development details.
Questions and answers
What model documentation is required in Colombia's AI policy?
Colombia's National Policy for Artificial Intelligence recommends publication of non-sensitive model documentation, including technical and non-technical descriptions like model cards. This applies particularly for systems used in public administration to enable transparency and oversight.
What does MAIC require for AI model documentation?
The Malaysia Artificial Intelligence Consortium describes reproducible artefacts such as data descriptions, training logs, evaluation results and model cards. These support conformity assessment, provenance tracking and transparency in consortium activities.
Does Portugal's AI Agenda mandate model documentation?
Portugal's National Artificial Intelligence Agenda, which is in draft, refers to structured records like model cards and datasheets describing datasets, training, metrics and risks. It intends these for transparency and compliance, particularly for publicly funded models.
Related Terms
Model Validation
Processes to verify models perform as designed....
model transparency
Clarity about a model's operation, inputs, outputs and limitations....
Transparency measures
Documentation and registries enabling visibility of AI systems....
Independent Validation / Peer Review
Objective review of model assumptions and performance....
uitlegbaarheid (explainability)
Ability to explain model purpose, inputs and decision logic....