[artifabrian]

I'd suggest it really depends on the exact problem space you are working on.

Some NLP related problems can be solved at a similar level of quality of humans - (humans actually make a lot mistakes as well).