Collection, consent, and provenance
Where data came from, whether people agreed, and whether you can trace a record back to its source—**provenance**.
What it is
Where data came from, whether people agreed, and whether you can trace a record back to its source—provenance.
Why it matters
Illegal or unethical collection creates legal risk and model risk. “We scraped it” is not a strategy.
How it works (plain)
Document: source, license/ToS, consent basis, intended use, retention, and known gaps. Prefer datasheets and intake checklists before training.
Everyday example
A photo dataset without model-release clarity can torpedo a commercial launch.
Try it
For one dataset you touch, write a 5-line provenance card.
Myths
- ⚠️ Myth: Public URL means free to train on.
- ✓ Reality: Terms, copyright, and privacy law still apply—get counsel when unsure.
- ⚠️ Myth: Provenance is bureaucracy.
- ✓ Reality: It’s how you debug and defend systems later.
Sources
- Course 02 data-and-labels; Course 20 policy
- Datasheets for Datasets literature (cite Gebru et al. when teaching)
- NIST AI RMF: https://www.nist.gov/itl/ai-risk-management-framework ↗