Should AIs make errors on purpose to keep human minds sharp?
Introducing occasional errors might keep human users on their toes and engage their cognitive skills more, potentially slowing cognitive decline.
That would only be useful if the user knows it's an error, or if the application were to explicitly tell the user it's an error. Otherwise you are just giving the user false information, which they might accept as fact.