1-1. No, we don't need more asset creators, we need assets that are more than minimally fleshed out. For instance, there's character generation programs for making a basic RPG sprite that walks in 4 directions. But why no attack animations? Why no expressions? Why no jumping or rolling sprites? Why are we failing to procedurally generate our sprites/animations, rather than churning out an endless recap of stylistically mismatched placeholder stuff?
1-2. Immediately, no. Ultimately, yes. What I mean by this is that all artists use particular methods to produce their art - this is why they have classes, how-to guides, and so forth. Procedurally-generated art, while POSSIBLE; is just as methodical as real artists are. We just don't effectively implement these methods, because we're more focused on the gameplay mechanics than the details thereof. But that said, nobody is going to devalue an artist's genuinely good work, either.
1-3. I really don't think so. If anything, they just reveal constraints that are already there; although the method of PGC can be constrained a little more specifically than the mechanics of the games would be, because algorhythms are fairly predictable regardless of their implementation. You either have high core design with less randomization, or high randomization with less core design. Unfortunately, this is a real tradeoff. That said - people are still replaying unchanging games like classic Mario, Sonic, Zelda, Metroid titles today with unrelenting fervor - so the constraints are less relevant/weighty than you give them credit for being. Even Spelunky/Binding of Isaac levels get predictable with overexposure; but that doesn't stop the main gameplay from still being fun.
2-1. Yes, and no. See, there is a difference between GOOD PGC and pure randomization; which comes from understanding gameplay flow and variation. I've been working on some PGC projects the past 5 years; and the system I'm most focusing on involves a deliberate changeup between three distinct focuses: Navigational challenges, Trap Avoidance, and Enemy Combat. While it's only one of about 5-6 level generation methods I'm experimenting with, they're made to interconnect from one type to the other, so that there isn't a singular focus. The singular focus on one aspect over the others, or attempts to overlap them is what results in the bland-feeling results you normally expect of PGC games; but by offering a variety between them, you get stuff like Rogue Legacy, that despite it's randomization, doesn't really get too stale too quickly.
There's also the issue with scope. A good PGC game shouldn't play all of its cards in a single session, but have plenty of various experiences to draw from; and it shouldn't aspire to be 2-hour-long sessions of dungeon crawl, but more optimally either a huge world to explore methodically (ala Terraria) or quick, arcade-like sessions of 20-45 minutes to perform through (Spelunky, Isaac). There's some middle ground to experiment and play around with, I'm sure; as well as what to keep persistent vs. what to replace/randomize.
3-1. Apples and oranges. Would the amount of content in say, Sonic 2, change at all if you simply randomized the order of the levels/setpieces? No, not really. Ultimately, the randomization serves only the replay value and the predictability vs. unpredictability of level/character design. If you use PGC to randomize a game's physics, and then build levels based on the randomized physics, the level design ultimately doesn't change at all - the same variables are calculated in the game's layouts, and the "change" is really just an optical illusion.
Ultimately, it isn't THAT much different; you just have a different kind of bug/troubleshooting process involved now.
3-2. I suppose there is; but I think it's more about experimental people vs. concrete people. People with fluid personalities gravitate towards gaming experiences with that kind of variance and fluidity about it; people more matter-of-fact probably gravitate away from it, since they want mastery of a predictable outcome as part of their internal reward experience. "Science theory" vs. "Scientific applications;" that kind of thing.
3-3. I have found that it makes it considerably difficult to explain to other people that aren't already "on the level;" and makes for more creative implementation to be necessary.
3-4. Now THIS is a brilliant question. This really requires two, very carefully weighed variables: a fundamental, deeper-logic understanding of specific game design (say, "Super Mario design" compared with "Platformer design"); AND the SAME understanding of the outcome you're going for. Using the SMB example, you'll notice that 99% of the levels' content takes place on three horizontal levels - each separated by exactly the height that Mario can jump to traverse between. So, aiming for a PGC SMB experience; your level generator needs to take into consideration these two factors: how to design your levels in vertical layers as such; and in a way that is contingent with your characters' physics.
EDIT: Oh, maybe a third. You need to understand a sense of flow, and how the pieces of your game come together to create (or defy) that.
Sometimes, this means working the physics calculation of your game "backwards" - rather than taking a default xspeed/yspeed and seeing what your physics end up being; we need to devise a way of taking an intended "maxjumpheight" global, and recalculate your initial yspeed based on THAT. Then, you change your global for your physics; and it's already recalculated for you!
But I don't think we do that yet. I feel dumb for not doing that yet, I know.
3-5. Nope. In PGC, it's "Polish" that loops back to "Content," since you're going to want to polish up each part of your implementation before moving onto the next, usually. Tests and Gameplay Layer are also reversed.
And personally, I've had trouble working with PGC concepts *before* asset creation. I at least need placeholders to get the right feel for collisions, etc.
4-1. Personally, my long-term goal is to make a game about gaming itself; that comprises a various selection of gameplay types and experiences together like a cohesive whole. Still kind of torn between a few metagame approaches, and I still have a lot of implementation to learn in doing so. But eventually, the metagame will just generate games within itself, to buy, to play, to compete in, that kind of thing; all from a set of common assets that make contextual sense.
But in the end, I really just expect to get demakes of Spelunky, Rogue Legacy, TBoI, Spiral Knights, maybe some Street Fighter/King of Fighters clones with mixed-up movesets, and a DDR/GH/RB kind of thing with strategically engineered songs/charts for the sake of being fun and full of flow.
Hope that's helpful!