What kinds of explanation actually support public accountability?
Especially when technical systems are probabilistic, opaque, or difficult to contest.
open
An explanation can be technically accurate and still fail the person who needs it. A model description, confidence score, or list of variables may say something true about a system without making a decision understandable or contestable.
For public accountability, the useful unit of explanation may be larger than the model. It may need to include the institution using it, the alternatives that were considered, the person responsible for the final decision, and the route available for appeal.
The standard should not only be whether an output can be explained, but whether the explanation gives someone meaningful grounds to question what happened next.