This next bit borrows (by which I mean steals) heavily from two books, in this order: Raph Koster's A Theory of Fun and Laurence Gonzales' Deep Survival, both of which discuss the root of the human urge to engage in challenging (or even risky) fun.
Complex social animals like tigers, dolphins and humanfolk spend fewer of their formative months in the womb, so they emerge into the world less able to fend for themselves than, say, fish or lizards or bats. On the one hand, our young come out without any special wiring that lets them immediately stalk prey or flee predators or engage in complicated locomotion. On the other hand, our young learn how to do all those things in a fundamentally different way, taking advantage of their more sophisticated brains to construct progressively more detailed and accurate models of how the world works and how they can optimize their time inside it (however brutal and short that time might be).
And this is where play comes in.
Humans are predisposed to repetitive experimental behavior; activity that gives us a chance to try out our abilities in a (relatively) safe way, where the consequence of failure might be a bruised tailbone or forehead-smacking sheepishness instead of abrupt annihilation. We need this probing behavior to build our mental map of the world. Each tentative attempt at testing a new skill is like a sonar ping, giving our nervous systems a fleeting glimpse of the local universe, revealing a little more detail and understanding of the rules that govern it. If we only try something once or twice, this glimpse fades phantom-like, leaving us with only a limited grasp of the cause and effect that governs the activity in question. A little "beginner's luck" might create an early impression of what's going on mechanically, but without any follow-up, this impression isn't reinforced and it, too will fade. Its lessons will not be transfered to the language centers of the brain, or the neuroendocrine responses of the body's muscles. In other words, we get rusty. If, however, we persist at our play, applying some discipline to the trial-and-error, a curious thing happens.
A child growing up on a farm might learn the skill of vaulting over a wooden fence. He does this by internalizing the specific combination of biomechanics, vision, proprioception and knowledge of "fence-ness" that enables him to predictably plant his hands on the top of the fence, while automatically throwing his weight forward and to the side, at the same time pushing off the ground with his foot in just the right order so as to clear the top. But what's interesting is that the child in question can apply his skill at vaulting generally to obstacles that aren't fences, on ground that's dry or muddy, on an uphill or downhill slope, maybe even in the dark when his eyes can't contribute much information. In other words, months or years of playing around on fences has given rise to a generalized mental model for "vaulting" that does not depend on a precise set of conditions to be usable. In fact the fuzzier the model is, the more useful it is. Rather than having to approach a new "vaulting challenge" by making a complete set of physics calculations to account for each step in the vaulting sequence, the child can rely on his mental map to supply him with subconscious feedback that reflects the degree to which the new situation differs from those that make up his past experience. This feedback comes in the form of feelings, pulses of neurochemical information that aren't processed by the higher brain at all, but by the more primitive part of the nervous system that governs our conditioned responses.
Over the last few epochs, humans have evolved a broad range of complex but recognizable emotions. They give us a mental vocabulary with which to interpret our bodies' unconscious reactions to various stimuli, both pleasant and threatening. Setting aside some of the complications that have emerged from that development - like love, xenophobia, or spiritual epiphany - one consequence is that we humans have cultivated a gut-level appreciation for conditions that push the boundaries of our mental maps' predictive power. In other words, we get more satisfaction out of successfully applying our experience to situations that, strictly speaking, fall just outside of it.
An individual approaches a new challenge by first perceiving all the ways in which the challenge is familiar. The closer the task is to something the individual has done previously, the less challenging he perceives it. But the more the task deviates from the familiar parameters recorded in the individual's mental map, the more it carries with it an inherent risk of failure. Humans, who excel at matching patterns, receive an automatic kick of adrenaline when what's in front of them doesn't match. Of course even allowing for variations in memory and frame of reference, if the challenge in question is too different from their mental model, the individual will stop intuiting any resemblance at all.
There's a sweet spot to this, one that every game designer at some point finds himself obsessing over: Success under conditions seemingly identical to past experience rewards the player with diminishing satisfaction; success under conditions so different from past experience as to be unrecognizable possibly only rewards the player with a sense of relief, of having randomly survived by the seat of his pants.
And thus we come to the problem of Familiarity in Gameplay.