I've been busy with meatspace lately, so I don't remember if I've said something about this before, but I figure it's worth drilling home. In science, most people understandably focus on how theories are supposed to predict and explain the facts that we observe. I, however, feel the need to point out the opposite side of that: A theory also needs to explain why certain things are not seen. It's been said that "a theory that can explain anything explains nothing," and it's my intention to give the reason for that saying.
Let's use the familiar comparison between evolution and Creationism. Evolution has constraints: A generation can only build on or modify what it inherited from its ancestors. Evolution tells us that Crocoducks are absurdly unlikely: Birds evolved from bipedal dinosaurs, a different branch of the tree of life from crocodiles. A Crocoduck would be analogous to (but much more extreme than) a couple siring a distant cousin's child. For that reason, the discovery of a Crocoduck (barring origin by mad science) would completely dismantle everything we know about genetics and evolution.
Creationism, however, can perfectly explain a Crocoduck: God can do whatever he wants, and for some incomprehensible reason was in the mood to slap two very different species together. There is no mechanism of action to limit what God's tools can do. Thus the Crocoduck is explained by Creationism.
Of course, the problem this presents for Creationism is that there are no Crocoducks. There are no lion-eagle hybrids with a six limb body plan. There are no cloven-hooved horses with spiral horns or lion tails. Creationism is useless because it can be used to predict anything, including all the things that will never happen. Creationists have no method for sorting out true predictions from the infinity of fantasy predictions. They can only point out certain "predictions" after the fact has been observed. That, in a nutshell, is why Creationism's unfalsifiability makes it useless.
Often, I call Creationism "The Random Theory of Randomness." If you think in terms of a dice experiment, I think it's quite apt: If you roll a die a million times and get all sixes, it's perfectly reasonable to think the die is weighted in a manner to come up six. That theory has a comparatively narrow set of predictions: If you keep rolling the die, you will get mostly sixes. The theory of randomness, that the die roll just happened to come up all sixes by chance alone, predicts any outcome pretty much equally.
If you rolled the die some more and get a roughly equal tally of all six numbers, that would cast heavy doubt about the die being weighted. If you continued to roll sixes, it would support the weighting theory, but the randomness advocates could just as readily claim that it's possible random chance still favors sixes. Of course, I would think any reasonable person would bank on the weighting theory.