Reminds me of a question I've always had about game AIs: suppose you have a really good one (like this one, or in chess eg), how do you make the easier difficul

Reminds me of a question I've always had about game AIs: suppose you have a really good one (like this one, or in chess eg), how do you make the easier difficulty levels? Just throwing random noise into the decisions would probably lead to non-enjoyable outcomes. Is there a structured approach to it?