Decision Pipeline #1

Created on: June 6, 2019, 1:07 p.m.

Provenance

Provenance-based Explanations for Data Controllers

We recorded the provenance of the decision pipeline, from which some explanations about it can be generated. An organisation may want to make sure that the creation, validation, and deployment of the pipeline were carried out appropriately. Some example questions include:

Q: Is the decision pipeline sufficiently accurate?

Organisations must assure that the performance of the system is satisfactory.

The company pipeline was assessed by a data engineer (staff/259) and has the level of accuracy of 79.58%.

  • A performant decision pipeline will contribute to the efficiency of a business and, ultimately, its profitability.
  • This explanation would help determine whether the data is processed fairly (Article 5(1)(a)) (although this would not lead to a complete fairness assessment). Ultimately its implementation would be useful for accountability purposes.

Back to list of questions

Q: Who was involved in creating this pipeline?

As part of their own governance and to support accountability, organisations must keep track of who did what and when in their internal processes.

Responsibilities for the AI pipeline were that a data engineer (staff/259) selected file (loans_filtered.xz), that a data engineer (staff/259) fitted data for the company pipeline (1558649326/5011959424), that manager (staff/37) approved the company pipeline (pipeline/1) and that a data engineer (staff/259) split file (loans_train.xz).

  • This explanation would help determine whether the data is processed fairly and transparently (Article 5(1)(a)) (although this would not lead to a complete fairness assessment). Ultimately its implementation would be useful for accountability purposes.

Back to list of questions