The Value of Procedural Generation

Hand Scripted events, items and layouts are something "large" games can miss. The player often finds the world as wide as an ocean with about half the depth of a puddle, example No Man's Sky.

but using it to generate 95% of your content is a result of misunderstanding how that tool should be used

I don't think you get to tell the world how a certain approach should be used.

I see the problem (especially in the most recent example of No Man's Sky) as being an obsession with every single variation being present in the game so that they can make that magical 18 quadrillion whatevers claim.

If you're going to use procedural generation to make player-ready content, you need to make sure it produces easily distinguishable results almost every time. You need to cut out something a sizable percentage of the results so that the differences between the things you keep are more obvious and more enjoyable.

And then, perhaps most importantly, you need an interesting game to make use of that content. Don't rely on assets of any sort to carry your game, whether procgen or hand-crafted.

I'm sure most of you are sick and tired of seeing "Procedurally Generated" in the blurb of almost every game nowadays

I don't base my game purchases on something as simple as this. I purchase games on the basis of whether it will be fun to play or whether it might be fun to play and occasionally because I want to see how a particular feature has been implemented.

Procedural generation is like any other tool/technique, used well it can enhance a game considerably, used poorly it can diminish the experience, But that can pretty much be said for any of the tools/techniques in game development. However I am in no hurried belief that the variety of applicable uses that procedural generation might bring to a game have all been explored, so arguing a limitation of their use doesn't really hold water with me.

Beyond that, given that the OP seems to be more of an opinion, wouldn't this thread be better served in the lounge rather that in Game Design as it is not actually seeking any feedback directly?

When I read the OP, I was like "How long will it take someone to bring up NMS?" And the answer was, of course, no time at all.

Proc-Gen could be used to generate 100% of your world, with enough iteration and testing. The challenge that NMS in successfully generating incalculable amounts of cohesive traversable art, kinda underlines is that that's not the same as generating a game. Even in the case of NMS, if they had scoped differently (an Earth-sized planet with 18 million different biospheres) with the same algorithm and had a Proc-Gen curator cut out things that didn't fit a particular diversity profile, the world would seem full and gigantic instead of empty and repetitive... with more time to focus on multiplayer and survival aspects, which aren't part of the world, and can't be proc-genned, but are how the game is evaluated most often. And room (and goodwill) for some pretty epic DLC as just one extra factor to the Proc-Gen could be used to fill out a whole solar system of similarly epic planets, and roll out the space game slowly. But, hey, hindsight is 20/20. Still, it's hard not to imagine what a team of 150 could do with this same approach that 15 guys used to make an appreciable however divisive game. In any case, teaching a Proc-Gen system to do the curation and editing that the OP suggests is the big innovation of NMS, and one that only an indie could do, and one that I hope AAA games will start to pick up.

The next step for Proc-Gen, and we've already seen some of this, is level design. We've seen Proc-Gen level design since the early days, and we already have city builders with some basic playability rules. Adding more and more and more level design rules to our city builders could, even with an indie company, create an arbitrarily large number of urban environments that were all playable and as diverse as Beverly Hills and Downtown Nashville.

But that's the rub... playability requires good gameplay. If you can generate a huge universe full of story and lore and all the story-generating systems, and platforming puzzles and secrets and etc... but the character isn't fun to move, or you haven't given collecting all the Proc-gen secrets meaning, then you don't have a good game. Maybe one day there will be a Proc-Gen system that can generate a whole game, a computer that can GM a great game of DnD, but right now, you still have to handcraft a good game yourself, even if you have a system that procedurally generates a perfect copy of real life, like real life, it's not fun unless a game maker makes it fun. Computers don't know what gameplay loops are, and they can't, because they can't empathize with the player and extrapolate from there. They don't even really know what story is, but a skilled game maker can make systems that have story values embedded in them that the player can latch onto.

And that's the other thing, don't be so in love with Proc-Gen that you are unwilling to do handcrafting that's needed out of pride or something. There's no need to make a game that 100% proc gen outside of having something fancy to put on the box, that, again as we learned with NMS, will become loved or hated based on the actual quality of the game.

But even saying 100% of your game is misleading. 100% of a gameworld isn't a game, and that's the real lesson that Proc Gen has to teach us this year.

I appreciate all of the discussion here. This thread is probably better suited for the lounge, I agree. Is there any way to move it?

In addition, this is just my opinion. By all means, use Proc-Gen all you want, I'm not going to stop you. I like to share my thoughts, that's all.

Love procgen. I have a blog on it and have created an procgen engine of my own.

When should or shouldn't you use it? The answer is the same as with other topics and technologies: it depends.

ProcGen is nothing more than an industrialisation process and so it comes with its pros and cons.

Still, if the concept isn't good, don't blame the execution, although execution has saved a lot of concepts :)

Moved to the Lounge for more subjective discussion.

Procedural generation is great when well done and with artistic input. Take a long at most of the very low size demo scene, really nice visual, often proceduraly generated but hand crafted still for a very good data size / visual quality ratio.

I think procedural worlds often lacking. They have some great surprises, but not near the quantity and quality of surprises that a real designer can put into the same world footprint.
Basically, procedural generation can create great areas, but the greatness is spread farther apart than I want, basically stretching out the world and watering it down.

My ideal use of procedural generation is in tools to help designers make better areas faster.

I think we need to backup a few steps and ask some questions regarding procedural generation:

1) How much engineering time will it cost to develop procedural generation which is indistinguishable in quality to a hand designed world? How much designer time would it cost to get the same or a better result?

2) How much value does it add to the player playing the game? If development resources are a fixed asset you can spend, would the player get more value out of a game designed to be deep but narrow, or a game designed to be shallow by broad?

I think we can look at Spore and No Mans Sky as examples which inform our answers to these questions now. Without those two games, our answers would have been more difficult, so there is great value in these games and looking at the mistakes they made.

I think at the end of the day, it all comes down to giving players a rich experience. We have to be very careful in how we define "rich experience", because a world with infinite objects to interact with doesn't necessarily make the world "rich" if the experience is universally the same. We have to look at what makes the game "interesting" and develop the game to be that: interesting. A game isn't interesting just because the worlds are procedurally generated. A game gets interesting when there's interesting things to do and think about in that game, and a procedurally generated world would just be the scaffolding upon which a richly designed game system rests upon; The procedural generation is pretty much irrelevant as far as I'm concerned. If you have to say "procedural generation" in your marketing pitch and value proposition, you're doing something wrong and either your pitch needs work or your game design needs work. The litmus test is to take away the procuderal generation aspect out of your game, play it extensively, and ask if its still a great game without it.

The litmus test is to take away the procuderal generation aspect out of your game, play it extensively, and ask if its still a great game without it.

Hmm... I'm not sure about this.

Consider the "roguelike" genre: Would a roguelike be much good if it relied on levels that were purely hand-crafted, resulting in the player exploring through exactly the same set of rooms on each run? I doubt it--intuitively, I imagine that most roguelikes would become boring if the levels were the same on each run. Now, one might argue that this should simply imply that roguelikes aren't particularly good games--but they seem to have a fair following, implying that a reasonable number of people enjoy them. To my mind, this implies that they likely have at least some virtues. If I'm right that they would become boring with static levels, I argue that procedural generation is one of the virtues of (most) roguelikes.

I think that, as I believe someone else wrote earlier in the thread, procedural generation is a tool: it can be overused, used to poor effect, etc.--but I hold that it can also be used well. And yes, I think that it can be a central feature--as in roguelikes.

The litmus test is to take away the procuderal generation aspect out of your game, play it extensively, and ask if its still a great game without it.

Hmm... I'm not sure about this.

Consider the "roguelike" genre: Would a roguelike be much good if it relied on levels that were purely hand-crafted, resulting in the player exploring through exactly the same set of rooms on each run? I doubt it--intuitively, I imagine that most roguelikes would become boring if the levels were the same on each run. Now, one might argue that this should simply imply that roguelikes aren't particularly good games--but they seem to have a fair following, implying that a reasonable number of people enjoy them. To my mind, this implies that they likely have at least some virtues. If I'm right that they would become boring with static levels, I argue that procedural generation is one of the virtues of (most) roguelikes.

I think that, as I believe someone else wrote earlier in the thread, procedural generation is a tool: it can be overused, used to poor effect, etc.--but I hold that it can also be used well. And yes, I think that it can be a central feature--as in roguelikes.

At most, that means that roguelikes are an exception from the norm where the expierience itself revolves around the procedurally generated levels and infinity replayability.

I think it could also point in the direction that "ProcGen Heavy Game" =/= Any other "ProcGen Heavy Game".

The big thing, and most probably also most important failure of NMS and Spores design was to hinge EVERYTHING around the idea of ProcGen, but try to retrofit a more mainstream gameplay into it. This is where the shallow depth comes from, from the try of the game designers to make the game infinitely wide on the tightest of budgets.

A roguelike does not try to do that. It is a very focused, narrow expierience that also makes extensive use of ProcGen as a gameplay tool, but doesn't try to simulate a vast expierience with it.

Instead, it uses ProcGen to DEEPEN the expierience, because replayability is actually the core of roguelikes.

If we can take away something, its that NMS and Spores failures to be compelling expieriences for a lot of players is not a sign that ProcGen is bad.

Its just NMS and Spore that tried to fit the square peg into the round hole and failed doing so.

At most, that means that roguelikes are an exception from the norm where the expierience itself revolves around the procedurally generated levels and infinity replayability. I think it could also point in the direction that "ProcGen Heavy Game" =/= Any other "ProcGen Heavy Game".

But if we dismiss roguelikes as simply an exception, what other exceptions might we be missing? More to the point, I'm not really arguing about No Man's Sky, or any other specific game; I'm not arguing that one procedurally generated game is like another. I'm arguing more against the broad suggestion that a game has poor mechanics if the removal of the procedural generation results in unsatisfying gameplay. (And, I think, an earlier suggestion that hand-crafted elements are always preferable--but I think that I may have misremembered the breadth of that statement.)

There is a danger that people will stretch the 'roguelike' term to 'any procgen game they enjoy' as a circular way to claim that procgen only works for roguelikes. Was Diablo a roguelike? In many ways yes, but in many ways no. Was Minecraft? Certainly not. Was the original Elite? Not at all. Spelunky? Civilization's random maps? Or those of the older X-Com games?

When you move beyond the heavily guided experience of a typical modern AAA blockbuster console game, you stop expecting to be led by the hand through a highly focused story and start seeing examples of carefully sculpted gameplay, often which uses lots of procedurally generated content. Those games show us that the mechanics and the content generation methods can be almost entirely orthogonal - it's only in the theatrical game=narrative cases where the algorithms aren't able to generate what is needed, yet.

This is a really fascinating discussion. What is the linchpin between a proc-gen system that makes a wide but shallow game and one whose procgen makes it deeper? If it's just "roguelike," what is it about that particular set of genre conventions that makes it 'work.' If we just say 'revolves around the proc-gen' that's not very descriptive. After all, one could just as easily say that NMS revolves around traveling to different planets for different resources.

I think there's something to the Narrative idea, where a proc-gen systemic narrative is in a way fundamentally different from a crafted narrative. I also think its a cool observation that both NMS and Spore wanted to fill an entire universe, however, neither concept demands that. A planet teeming with life and environments facilitates all the things. But that 'fill a universe' idea is just too tempting, because it is, in theory, possible.

On the taking out the levels bit, I think that applies to all games. If the core gameplay loop isn't fun, if it's not even a little bit fun to move around in an empty room, you may not have a good game on your hands. You can honestly go the other way. If your FPS isn't fun to play on NMS planets, it won't be fun to play in handcrafted arenas either.

On surprises, I think proc-gen can be used to do surprises, though it's much cheaper for a normal sized game to handcraft. Procedural art is a well pioneered technology that can be used to really fill out an area in terms of sculptures, architecture, signage/billboards and associated landmarks. Rarely used attributes or rarely surpassed thresholds on common entities, whether they be weapons, NPCs or something else could be used to create rare and super-rare items with incredible infrequency. The rarity of these items does lend heavily to proc-gen as a tool that is curated by a person as opposed to runtime use.

Is it just me, or is the dream of proc-gen the ability of a small team to make a AAA-sized quality game? But I guess that goes back to the scoping issue, as AAA games don't aim or claim to fill an entire universe.

The litmus test is to take away the procuderal generation aspect out of your game, play it extensively, and ask if its still a great game without it.

Hmm... I'm not sure about this.

Consider the "roguelike" genre: Would a roguelike be much good if it relied on levels that were purely hand-crafted, resulting in the player exploring through exactly the same set of rooms on each run? I doubt it--intuitively, I imagine that most roguelikes would become boring if the levels were the same on each run. Now, one might argue that this should simply imply that roguelikes aren't particularly good games--but they seem to have a fair following, implying that a reasonable number of people enjoy them. To my mind, this implies that they likely have at least some virtues. If I'm right that they would become boring with static levels, I argue that procedural generation is one of the virtues of (most) roguelikes.

I think that, as I believe someone else wrote earlier in the thread, procedural generation is a tool: it can be overused, used to poor effect, etc.--but I hold that it can also be used well. And yes, I think that it can be a central feature--as in roguelikes.

This is a good question, and I think it's testable. So, how do you test a roguelike game without procedural generation?

Get a brand new person who has never played the game before. Load a level. You could even do this as an blind A-B test, where you push a button and one person randomly gets a procedurally generated level and another gets a designed level, or a preset level. Now, they play the game for the first time. Because its the first time, any merits of a procedurally generated level are wiped away (variety in replayability). So, now you have to ask, "Is the game fun or boring?". If you add procedurally generated levels, all you're doing is adding replayability to a game which should have a core game play mechanic which isn't the procedural generation. If the game is fundamentally boring, then no amount of procgen is going to fix that underlying problem: congrats, you can now create ten million boring levels. If the game is fundamentally fun, then the procgen is not the core element of the game play, but it does add replay value.

I'd much rather have a game with 1 level which has a really, really fun game play mechanic. The developer nailed it. The game is deep. It's interesting. Whatever they did, it's fun and working. The developer spent most of their project "budget" (time, people, resources) generating this fun game experience. Great! Now, hire a level designer for three months and have them design five large levels, randomly pick a level when the game starts, and you've got replay value! Or, hire a hand full of designers and have them build about 40 maps, then let players choose a map to pick. Again, loads of *cheap* replay value. A lot of RTS games do this quite well.

On the other hand, some companies spend most of their project "budget" working on building robust procedurally generated worlds. At the end of the project, you can say, "Great, the worlds look convincingly different! But... you spent your whole project budget developing this part of the tech instead of building a deep game. You've got thousands of hours of worlds to explore, but three hours of interesting game play before it gets boring and tedious." Just think, what if No Man's Sky had spent 90% of their project budget making the core part of their game REALLY fun, skipped the procedural world generation all together, and just hired a couple people to hand design 5 different planets in the same star system, and they created a multiplayer element where players join factions and battle for control of territory on each planet to control precious resources? And you can craft things and build houses. It would have been the game everyone dreamed it would be.

I now think that heavy procedural generation is a trap. It's easy to get excited about, but it wastes time and is inferior to a carefully designed set of levels / worlds.

One aspect being missed from the above is that procedural generation cuts costs. It makes things practical that were not otherwise practical for the same team. It is certainly possible that - replayability aside - 20 hand-designed levels could be more interesting than 20 procedurally-generated levels, given the same designers being involved in both cases. But it's not possible that, assuming a competent code team, the 20 hand-designed levels could be made in the same space of time.

Or, hire a hand full of designers and have them build about 40 maps, then let players choose a map to pick. Again, loads of *cheap* replay value.

I'm guessing from this statement that you have never hired a designer. :) They are cheaper than coders, for sure, but not so cheap that getting them to make 40 good maps is 'cheap' compared to procedurally generating them.

just hired a couple people to hand design 5 different planets in the same star system [...] It would have been the game everyone dreamed it would be.

I don't know who you've been speaking to, but I don't think there would have been even 10% of the same interest in this game if they'd said "you can explore 5 hand-crafted planets". As for factions and multiplayer... that was never what it was about. EVE Online exists for that sort of thing. If you want to understand No Man's Sky you need to understand the history of games like Elite. It doesn't gel well with the modern heavily guided theme-park experience, but that's a failure of marketing and the runaway hype machine more than anything else. I know people who really enjoy the NMS gameplay for what it is, like they enjoy other open-ended and exploration-focused games.

----------------------------------

Let me approach this from another direction. There's still a belief that procgen is somehow replacing designers when really it is about freeing them up to work on higher level aspects. Exactly the same thing happened with the art pipeline. Go back about 25 years, and artists had to place every pixel themselves. Later, they might model a character in 3D and render it out as sprites, so they would only edit pixels to clean up a render, perhaps add detail, that sort of thing - the rest was handled by the lighting and texturing algorithms. In 3D engines, it's gone even further - the artist might produce the model and the texture maps, but they don't get any control over the final pixels on-screen - they just have to trust the algorithms will render their data in a way that works for them. But freeing them from pixel tyranny means they can be more productive in other ways - changing one texture will improve multiple assets. One animation can animate multiple models. Altering a light value will affect the whole scene, environment, props, and characters. Artists gave up some fine control but can now deliver whole 3D experiences with far more content in them.

Procedural generation is the same thing for designers. Instead of designers hand-placing every wall or door, they should be able to tell the system how walls and doors should be placed, and the system does the work for them. If they build a room and want to fill it with props, they should be able to have the game auto-generate appropriate props based on the type of room. If they designate an area as a dangerous forest, the game should be able to populate it with outlaws and wolves without anyone needing to go and drop in spawn points or planned encounters. If the designer knows that orcs and elves hate each other, the system should be able to create quests based on this antagonism, without requiring a designer to manually place each scripted event and write every bit of dialogue. The tools can amplify the designer's ideas, not replace them.

This is a good question, and I think it's testable. So, how do you test a roguelike game without procedural generation? Get a brand new person who has never played the game before. Load a level. You could even do this as an blind A-B test, where you push a button and one person randomly gets a procedurally generated level and another gets a designed level, or a preset level. Now, they play the game for the first time. Because its the first time, any merits of a procedurally generated level are wiped away (variety in replayability). So, now you have to ask, "Is the game fun or boring?".

Ah, I see! I think that I misread your previous post, then (and my apologies for that).

I believe that I read you as suggesting that the test would be, essentially: "If we remove procedural generation entirely from the game, is it still an engaging game?"

If I understand you correctly now, it's rather: "Regardless of how a given set of elements was generated, is the moment-to-moment (or perhaps the "single-level-or-equivalent") experience engaging?"

These are similar, but not quite the same--a roguelike might easily fail the first (since it would likely get boring quickly) but, as you point out, pass the second (since playing a given level is, hopefully, engaging).

I now think that heavy procedural generation is a trap. It's easy to get excited about, but it wastes time and is inferior to a carefully designed set of levels / worlds.

While any one level is likely to be inferior, the overall result of having a large number of level-configurations might provide a preferable experience than the result of having a small number of really good levels. Once again, roguelikes are an example here: a roguelike with five excellent levels probably won't be up to much, while one that can generate a hundred decent levels could be a lot of fun.

That said, I do think that there are games, or types of game, in which a single, brief experience is very worthwhile. Conversely, I think that there are cases in which it's important to have plenty of content, and cases in which deep gameplay might feel wasted if there isn't enough content in which to experience it.