The rules theory breaks down quickly under consideration, however. For example, what about the difficulty of getting computers to identify handwritten letters? It has proved impossible to specify a set of features that would clearly mark off, say, the letter "A" from all other letters.
It's not just letters that are difficult, but concrete things like animals as well. Think about a chicken. What about a frozen chicken in the supermarket -- is it still a chicken? When you take it apart and eat it, is there a point at which it ceases to be a chicken?
And what about a game -- what is the common feature of all games? Games are related to other games, many have overlapping features, but there is no one feature that we can point to and say, "This defines a game."
So maybe concepts are not built up of features. Concepts simply tell us which features are important. This leads to a contradiction -- we must already have a concept to know what features to look for. When you're learning what a dog is, how do you know what is relevant? You don't. You start with one dog. Gradually you refine your understanding. In the end, concepts are defined relative to a prototype, a "best example."
Interestingly, there is a high degree of agreement across societies as to what a "best example" of each thing is. The best example of a cup, a dog, or the colour red, are the same across the world. When we get near the fuzzy edges, say a picture with an animal that could be either a dog or a cat, people give various answers, but the closer a dog is to the "best example", the more unanimous the reply. (And incidentally, the best example of a dog looks like a beagle.)