Game content sells games. Take a look at advertising material for games and you'll see boasts such as "40+ weapons", "7 vast worlds to explore" or "over 700 cars". What if you could smash your competition out of the water by saying "unlimited worlds to explore". Holy smokes, that sounds great.
The key to this is procedural content generation (PCG). It boils down to getting artists to make up lots of little parts, then assembling the parts on the fly in a unique way. Whether these are literally "parts" or more abstract things such as colour pallets depends on the game, but that's the gist of it.
For example, instead of creating a cat and a dog, you could create a cat head, cat legs, and a cat body - and the same for the dog - then assemble them randomly to create several cat-dog hybrid creatures. It wasn't much more work than just making the cat and the dog, but by mixing and matching, you've suddenly got "over 20 creatures" to proudly print on the box art. It almost sounds too good to be true, right?
And, in some ways, it is too good to be true.
Procedural content has been around for a long time. Bethesda is a huge player in the role-playing game genre, and some of its earliest games (Arena and Daggerfall) had worlds that were procedurally generated by placing a few pre-made assets around and generating names. Since then, they've scaled those procedural features right back, and now it's only used to generate insignificant side-quests. If it's so good, why would they be moving away from it?
I think the answer is that you can't really fool the player into thinking you've got more content than you really have. In the short term, yes, but humans are exceptionally good at finding patterns, so it doesn't take long to realise that your 20 mutant cat-dogs are made up of the same few parts. Remember mixing bright paints together as a kid and getting a mucky grey-brown colour each time? The same sort of thing happens here - mixing and matching tends to actually kill diversity rather than create it.
In my opinion, there is one potential get-out-of-jail-free card, one which a game called Spore used several years ago with reasonable success. Players are good at spotting patterns, and they also like to stand out from the crowd - why not combine these two things to get players to make the game content?
Yes, it works to some degree. You get some extremely inventive creatures in Spore, and they have a certain consistency in their designs that you don't tend to get from purely artificial PCG (no mucky grey-brown here). But as a group, humans can't help pushing the boundary. Even with community moderation, you get a lot of swastika monsters and walking phalli terrorising the planets of Spore. Sigh. It's the same in every other game I've seen with player-made content. I guess, as a species, we just can't help it.
So there you go. That's the trade-off. Do you stick with traditional finite content, have 20 thousand variations on brown, or risk the full glory of immaturity in full bloom? There is no magic bullet.
"No Man's Sky is a game about exploration and survival in an infinite procedurally generated universe," it says on Steam, and setting aside your personal feelings about how successfully that vision was ultimately realized, Hello Games can at least point to the fact that its algorithm did indeed spit out a gazillion planets (for people to grouse about). More astute analysis of how it all happened can now be found at 3DGameDevBlog, which has posted an in-depth breakdown of No Man's Sky's procedural generation. The report gets bogged down in detail fairly quickly with code snippets and whatnot, and there's a certain amount of guesswork going on as well. But what emerges is a picture of a game that may have fallen short of its pre-release billing, but is nonetheless a remarkable technical accomplishment.