Technical

Dataset Provenance

Traceable metadata on the origin and composition of training data.

How the laws define this

Both instruments establish that dataset provenance requires documenting data sources, collection methods, and preprocessing steps for datasets used to train or validate artificial intelligence systems, with the shared goal of evaluating bias risks and supporting legal compliance. Where they diverge is in their specific scope and stated regulatory purposes. Law No. 31814 — Law that promotes the use of Artificial Intelligence in favor of the country's economic and social development (Ley que promueve el uso de la inteligencia artificial) explicitly emphasizes evaluating representativeness, ethical provenance, data quality, and compliance with data protection rules. In contrast, the Regulation of Artificial Intelligence Bill, 2024 expands the required tracking to include data curation and lineage, while explicitly tying the documentation to fulfilling transparency obligations under the bill rather than focusing on data protection rules or representativeness.

Synthesised from the 2 statutory definitions below. Each one is quoted in full, with its source.

What it triggers

In Pakistan, under the Regulation of Artificial Intelligence Bill, 2024, documentation of data sources, collection methods, preprocessing, curation and lineage would have been required to support bias assessment, legal compliance, and transparency obligations.source
Under Law No. 31814 — Law that promotes the use of Artificial Intelligence in favor of the country's economic and social development, information and records detailing the sources, collection methods, preprocessing steps, representativeness, and legal/ethical provenance of datasets must be provided to assess data quality, bias risks and compliance with data protection rules.

Definitions (2)

Information and records detailing the sources, collection methods, preprocessing steps, representativeness, and legal/ethical provenance of datasets used to train or validate AI systems, used to assess data quality, bias risks and compliance with data protection rules.

Law No. 31814 — Law that promotes the use of Artificial Intelligence in favor of the country's economic and social development (Ley que promueve el uso de la inteligencia artificial)Definition 1 of 2

Documentation of data sources, collection methods, preprocessing, curation and lineage used to train or validate AI systems, intended to support bias assessment, legal compliance, and transparency obligations under the bill.

Regulation of Artificial Intelligence Bill, 2024Definition 2 of 2

Not to be confused with

data provenance/lineage

Data provenance/lineage refers specifically to the record of data origins and processing history, while dataset provenance encompasses broader traceable metadata on origin and composition including documentation for compliance and bias assessment.

Data Provenance

Data provenance focuses on documentation of data origin, collection, and transformations, whereas dataset provenance includes additional details like preprocessing, curation, lineage, representativeness, and legal/ethical aspects for AI training validation.

Questions and answers

What duties does dataset provenance impose under the Pakistan AI Bill?

The Regulation of Artificial Intelligence Bill, 2024 would have required documentation of data sources, collection methods, preprocessing, curation and lineage to support bias assessment, legal compliance, and transparency obligations. This bill was withdrawn.

What is required for dataset provenance in Law No. 31814?

Law No. 31814 requires information and records detailing the sources, collection methods, preprocessing steps, representativeness, and legal/ethical provenance of datasets used to train or validate AI systems. This is used to assess data quality, bias risks and compliance with data protection rules.

How does dataset provenance help with AI compliance?

Dataset provenance provides traceable metadata on the origin and composition of training data, including documentation for bias assessment, legal compliance, transparency, data quality, and bias risks as defined in the instruments.