It's a little different:
1) Reduce 'soft' duplication. Overhead of having many small logging calls everywhere is probably higher than we all want to admit
2) Reduce small variances. Common logging format is useful (as a baseline-- additional logging is probably also useful)
I'm not terribly familiar with AOP but the cognitive overhead IMO is reduced if you just have to act as a client to logging (via annotation) than bespoke implementations at each call site.