JwumaConsent and licensing for field-captured data
· 2 min read · provenance, compliance
Consent has to be obtained at capture, by the person capturing and recorded against the specific item. Consent gathered afterwards is usually impossible to obtain and always weaker evidentially. A defensible record ties a named consent to a specific file, with scope and date.
The short version
- Consent obtained after capture is often impossible and always weaker.
- Consent should be recorded per item, not as a blanket collection statement.
- Scope matters: consent for research use is not consent for commercial training.
- Withdrawal handling has to be designed before collection begins.
- Capture briefs should state what may not be captured at all.
Scoping something like this?
We would rather talk about the problem than quote against a volume figure.
See capabilitiesWhy must consent happen at capture?
Because afterwards you cannot find the person. A photograph of a market stall taken three months ago has no route back to the stallholder, and any consent process that assumes otherwise is theatre.
It also has to be obtained by the person holding the camera, which means the capture brief must tell them how. Leaving consent to a contributor's judgement produces inconsistent records and, more often, none.
What does a defensible consent record contain?
A link to the specific item, the scope consented to, the date and enough identification to honour a withdrawal request later. Scope is the field most often left vague, and it is the one that matters: consent for a research dataset does not cover commercial model training, and conflating them is the most common defect.
Where consent cannot be obtained, the brief should say the subject is not to be captured at all rather than leaving it to be resolved later.
- Link to the specific captured item
- Explicit scope of use consented to
- Date and capture context
- A route to honour later withdrawal
- A stated list of what must not be captured
How should withdrawal be handled?
Designed before collection, not after a request arrives. You need to know how you would locate every item associated with a person and what happens to a model already trained on them.
The honest position is that removal from a dataset and removal from a trained model are different problems with different costs. Suppliers who imply otherwise have not thought about it.
People also ask
When should consent be collected for captured training data?
At the moment of capture, by the person capturing. Consent sought afterwards is usually impossible to obtain and evidentially weaker even when it is.
Does research consent cover commercial model training?
No. Scope has to be explicit, and conflating research consent with commercial training consent is one of the most common defects in field-collected datasets.
What happens if someone withdraws consent later?
You need a route to locate every item linked to them, designed before collection begins. Removing data from a dataset and removing its influence from a trained model are separate problems with different costs.