Description of potential stakeholders that can be brought together as part of our value-based framework. These stakeholders have been mapped following the two dimensional criteria (type of knowledge —formal, instrumental or personal— and contexts in which this knowledge manifests —ML, data domain, milieu—) outlined by Suresh et al. ([76]). The nature of their knowledge and the purpose of gaining insight for each of them have also been defined.
| Stakeholder | Mapping [76] | Nature of knowledge | Purpose of insight |
|---|---|---|---|
| Development team | ML, Formal + Instrumental + Personal | - ‘‘Knowledge of the math behind the architecture’’ ([76]) - ‘‘Stakeholder involved in an ex-ante impact assessment of the automatic decision system’’ ([77]) |
- Ensure/improve product efficiency and debug ([78]) - Research new functionalities ([78]) |
| Auditing team | Milieu, Formal + Instrumental | - ‘‘Familiarity with broader ML-enabled systems’’([76]) - ‘‘Experts who intervene wither upstream or downstream’’ ([77]) |
- Verify model compliance with legislation ([78]) |
| Data domain experts | Data domain, Formal + Instrumental | - ‘‘Theories relevant to the data domain’’([76]) - ‘‘Professional involved in the operational phase of the automatic decision system’’ ([77])) |
- Gain scientific or domain-specific knowledge ([76], [78]) - Trust the model ([76], [78]) - Act based on the output ([76]) |
| Decision subjects | Data domain + Milieu, Personal | - ‘‘Lived experience and cultural knowledge’’ ([76]) - ‘‘Layperson affected by the outcomes of the automatic decision system’’ ([77]) |
- Understand their situation ([78]) - Verify fair decision ([78]) - Understand how one’s data is being used ([76]) |