AI governance

What kinds of explanation actually support public accountability?

Especially when technical systems are probabilistic, opaque, or difficult to contest.