Forgive me if this is all obvious or obviously wrong.
The feedback that we can receive can be composed variously of emotional content and "factual" content. If you press a button and see Mario jump, there's no emotional content intrinsic to the feedback you have received. You impose your own valuation of the jump after you notice that it is getting you into (or out of) a tight spot*. If you get to the flagpole and see the fireworks, there is some emotional content. If you grab a star, you get a visual effect that factually informs you that your state has changed, but you also get a rockin' sound effect that has some emotional content.
*But how did you know to evaluate that tight spot as a bad thing that you wanted to avoid or get out of?
Imagine chess. If you are playing normally, you like it when you win and you dislike losing. There are states (checkmates or stalemates) that are rather trivially associated with a sentiment (like or dislike), and this association is more or less built into the game. As you learn how the pieces move, and what kinds of positions lead to what other kinds of positions, the sentiments associated with ending positions percolate back into earlier states of the game that might lead to (or remind you of) them. It doesn't take long to realize that losing your queen stinks- but that is a sentiment that you impose on the game; it is not built into the game.
Now, this process of attaching sentiments to chess positions (or events) is not merely a mechanical process of exploring the graph of possible games; if it were, we'd all play like computers, and we don't. The process of attaching sentiments involves the use of heuristics- some learned from others, some derived on an individual basis- that help you roughly determine how closely associated a particular circumstance or event should be with a win, loss, or tie.
If your heuristics are really good, they are effectively equivalent to traversing the graph of possible games, and the game (or opponents) can't surprise you in interesting ways. You've "beaten" the game, in a sense.
There are a couple interesting things about chess. One is that no one has good enough heuristics that they can claim to have "beaten" chess. The other is that the emotional feedback you get is reserved for the very end of the game. The rest of the feedback you get is factual, and the sentiments you feel in the middle of the game are your own appraisals.
Most of the games we make have emotional feedback at intermediate points throughout the game. I think +Will Robinson
's issue is not necessarily with loops themselves, but with certain uses/methods of intermediate emotional feedback.
There is a difference between wanting and liking. Addicts (among other things) want without necessarily liking. Is it the case that liking
a game experience is associated with sentiments arrived at through a subjective appraisal via heuristics, whereas wanting
to play a game is more associated with sentiments induced by emotional feedback? Those ideas may be amenable to empirical investigation.