Definition
Learning that changes what an institution is permitted to do, as distinct from what a model predicts. The model retrains and the prediction improves, while the challenge never changes the evidentiary rule and the exception never changes the ontology; that is model learning without institutional learning.
It matters because Ethotechnics cares about the second. The test is political: did the failure alter the evidentiary rule, redistribute authority, change who pays for the burden, or constrain what the institution may do next time? If not, the machine may be learning while the institution remains structurally unchanged. See also exception learning.
StructuralAgency
Tests that show it is in place
- A failure that changed no policy, authority, or burden allocation is not reported as learning.
- Retraining and workflow optimization are distinguished from changes to evidentiary rules or authority.
- The institution can name what a past failure is now entitled to constrain.
References
- Exception learning · Theory