The initial, straightforward comparison of ALTAI with the UC-specific ethics principles demonstrates that the bottom-up process conducted in AIOLIA, which started from very concrete AI use cases, foregrounds very similar ethics concerns as established frameworks. The high-level principles of frameworks such as ALTAI thus seem well-reflected in practitioner discussions about the practical implementation of AI ethics. However, we also observed that across the diverse use cases contexts, different focus and emphasis was given to either overall ethics concerns (e.g., Human oversight) or specific sub-aspects (e.g., auditability, deskilling, safety).
More pertinent, however, is the view into the components identified as basis for the operationalisation of AI ethics. Across the very different use case contexts, we found how strongly ethics principles and components are interlinked and interdependent. The high degree of overlaps clearly illustrates that:
- ethics principles are not exclusive, separate, or independent from each other;
- ethics principles do not form a division of AI ethics into ‘branches’ or static ‘subfields’; rather their meaning is constantly reconfigured in practical contexts;
- similar components can be relevant for multiple principles;
- ethics principles can even be seen as components of other principles;
- other considerations (human rights, bioethics principles, environmental ethics principles) may enter in specific contexts and be lifted to the same level of importance as AI ethics principles.
The bottom-up data collection process thus elicited important insights into the challenges of determining how exactly to define ethics principles, what they entail, how they differ from other principles, and whether it is ever possible to address one principle without also touching on others.
This means that a tidy principle-based framework is an illusion. In operational practice it is largely dissolved, because implementation is considerably more complex than a managerial checklist. Other components may arise, and practical technical and operational measures may address several principles or components at the same time. Assessing, evidencing and auditing these complex implementation processes is therefore a task that requires ongoing dialogue between ethics experts and AI engineers or researchers, a process that can be achieved through operationalisation in co-creation.
Source: AIOLIA deliverable 3.1