Automated Decision-Making Systems Used by MSD
Kieran Pegler made this Official Information request to Ministry of Social Development
Currently waiting for a response from Ministry of Social Development, they must respond promptly and normally no later than (details and exceptions).
From: Kieran Pegler
Dear Ministry of Social Development,
Under the Official Information Act 1982, I request the following information relating to automated decision-making systems, predictive analytics systems, or AI-assisted systems used by the Ministry of Social Development in relation to beneficiaries.
I request:
1. The names of any external vendors, contractors, or technology providers involved in the supply or operation of automated or AI-assisted systems used for:
* benefit assessment,
* compliance or sanctions,
* debt assessment,
* fraud detection,
* or client risk assessment.
2. Copies of any contracts, procurement documents, cabinet papers, ministerial briefings, or business cases relating to those systems.
3. Any privacy impact assessments, algorithmic impact assessments, risk assessments, or auditing reports relating to those systems.
4. Documents describing:
* whether decisions or recommendations are generated automatically,
* what level of human review is required before action is taken against a beneficiary,
* and what categories of beneficiary data are used by those systems.
5. Copies of any operational guidance, workflow instructions, system prompts, or staff guidance used in relation to generative AI tools or automated decision-support systems affecting beneficiaries.
6. For each identified system, please state whether MSD classifies it as:
* rules-based automation,
* predictive analytics,
* machine learning,
* generative AI,
* or another form of automated decision-support.
If any part of this request is considered too broad, I request assistance to refine the request under section 13 of the Official Information Act rather than refusal.
I request the information in electronic form where possible.
Regards,
Kieran Pegler
Things to do with this request
- Add an annotation (to help the requester or others)
- Download a zip file of all correspondence (note: this contains the same information already available above).
