Are you talking about LLMs or symbolic manipulation systems? The latter are certainly very logical, the former... not so much.