Evaluation is creation, at least when it comes to building intelligence for Intelligent Systems. That’s because intelligence creation generally involves an iterative search for effective intelligence: produce a new candidate intelligence, compare it to the previous candidate, and choose the better of the two. To do this, you need to be able to look at a pair of intelligences and answer questions like these:
Which one of these should I use in my Intelligent System?
Which will do a better job of achieving the system’s goals?
Which will cause less trouble for me and my users?