"In one sense, because of the game’s procedural design, the entire universe exists at the moment of its creation. In another sense, because the game only renders a player’s immediate surroundings, nothing exists unless there is a human there to witness it."

"Through the use of procedural generation, No Man’s Sky ensures that each planet will be a surprise, even to the programmers. Every creature, AI-guided alien spacecraft, or landscape is a pseudo-random product of the computer program itself. The universe is essentially as unknown to the people who made it as it is to the people who play in it—and ultimately, it is destined to remain that way."

More at:


New to LessWrong?

New Comment
22 comments, sorted by Click to highlight new comments since: Today at 9:45 AM

Video games with procedural generation of the game universe have existed since forever, what's new here?

At least there was an interesting part reminiscent of Eliezer's Universal Fire:


Matches catch fire because of phosphorus - "safety matches" have phosphorus on the ignition strip; strike-anywhere matches have phosphorus in the match heads. Phosphorus is highly reactive; pure phosphorus glows in the dark and may spontaneously combust. (Henning Brand, who purified phosphorus in 1669, announced that he had discovered Elemental Fire.) Phosphorus is thus also well-suited to its role in adenosine triphosphate, ATP, your body's chief method of storing chemical energy. ATP is sometimes called the "molecular currency". It invigorates your muscles and charges up your neurons. Almost every metabolic reaction in biology relies on ATP, and therefore on the chemical properties of phosphorus.

If a match stops working, so do you. You can't change just one thing.

The surface-level rules, "Matches catch fire when struck," and "Humans need air to breathe," are not obviously connected. It took centuries to discover the connection, and even then, it still seems like some distant fact learned in school, relevant only to a few specialists. It is all too easy to imagine a world where one surface rule holds, and the other doesn't; to suppress our credence in one belief, but not the other. But that is imagination, not reality. If your map breaks into four pieces for easy storage, it doesn't mean the territory is also broken into disconnected parts. Our minds store different surface-level rules in different compartments, but this does not reflect any division in the laws that govern Nature.

We can take the lesson further. Phosphorus derives its behavior from even deeper laws, electrodynamics and chromodynamics. "Phosphorus" is merely our word for electrons and quarks arranged a certain way. You cannot change the chemical properties of phosphorus without changing the laws governing electrons and quarks.

If you stepped into a world where matches failed to strike, you would cease to exist as organized matter.

Reality is laced together a lot more tightly than humans might like to believe.

From the article:

The team programmed some of the physics for aesthetic reasons. For instance, Duncan insisted on permitting moons to orbit closer to their planets than Newtonian physics would allow. When he desired the possibility of green skies, the team had to redesign the periodic table to create atmospheric particles that would diffract light at just the right wavelength.

“Because it’s a simulation,” Murray stated. “there’s so much you can do. You can break the speed of light—no problem. Speed is just a number. Gravity and its effects are just numbers. It’s our universe, so we get to be Gods in a sense.”

Even Gods though, have their limitations. The game’s interconnectivity means that every action has a consequence. Minor adjustments to the source code can cause mountains to unexpectedly turn into lakes, species to mutate, or objects to lose the property of collision and plummet to the center of a planet. “Something as simple as altering the color of a creature,” Murray noted, “can cause the water level to rise.”

I think most proceduraly generated games aren't that deeply interconnected with regard to their laws of physics.

I think most proceduraly generated games aren't that deeply interconnected with regard to their laws of physics.

It's also not clear to me if this is the result of good coding or bad coding. (If you've seen people play around with ROM memory tricks, that's what I have in mind; changing a bit that's used in multiple places for multiple things is different from "we adjusted Planck's constant and now the periodic table looks different because the energy levels come out differently.")

This is a press release though, lots of games were advertised with similar claims that don't live up to expectation when you actually play them.

The reason is that designing an universe with simple and elegant physical laws sounds cool on paper but it is very hard to do if you want to set an actually playable game in it, since most combinations of laws, parameters and initial conditions yield uninteresting "pathological" states. In fact this also applies to the laws of physics of our universe, and it is the reason why some people use the "fine tuning" argument to argue for creationism or multiple universes.

I'm not an expert game programmer, but if I understand correctly, in practice these things use lots of heuristics and hacks to make them work.

since most combinations of laws, parameters and initial conditions yield uninteresting "pathological" states

Another issue is too simple optimums. Human players are great at minmaxing game rules (=physics) and if the optimal behaviour is simple, well, the game's not fun any more.

Procedural universes seemed to see a real resurgence from around 2014, with e.g. Elite Dangerous, No Man's Sky, and a quite a few others that have popped up since.

I love a beautiful procedural world, but I think things will get more interesting when games appear with procedural plot structures that are cohesive and reactive.

Then multiplayer versions will appear that weave all player actions into the plot, and those games will suck people in and never let go.

Artificial storytelling has some promising directions for games and there may be some reasons to think that this can have benefit value aligned AI research.

we argue that the traditional goal of AI in games—to win the game—is not the only, nor the most interesting goal. An alternative goal for game AI is to make the human player’s play experience “better.”

-- Beyond Adversarial: The Case for Game AI as Storytelling

Also Storytelling may be the secret to creating ethical artificial intelligence but alas storytelling is hard.

the traditional goal of AI in games—to win the game

That's just not true. The traditional goal of AI in games is to lose in an interesting way after forcing the player to spend some effort.

This calls to mind someone's mod to a Total War game to improve the AI, where a user complained that Portugal had invaded Ireland to seize some territory for itself, instead of leaving it there for England to eventually take. "But if the AI doesn't do that," the modder asked, "how is it going to win?"

I prefer different rules. The AI tries as hard as it can to beat me while I have an unfair advantage.

Sort of the reverse of the typical situation.

I prefer different rules.

In which situation does this preference come into play?

Dwarf Fortress is a great example, of the "The AI only has to win once" variety.

It largely determines which games I play, and for those I want to play which don't adhere to my preferred rules, I modify until they do. (It's much less satisfying than when the game is designed for it, granted.)

I don't know if I'd characterise Dwarf Fortress as a game where "the AI tries as hard as it can to beat me". As far as I understand, the AI for mobs (and dwarves) in the game is pretty rudimentary, not much more than pathfinding -- it's just that the rules of the game (= physics of the simulation) are very unforgiving.

It looks to me to be similar to taking a small open boat across the Atlantic -- can be done, but any mistake or just bad luck can have dire consequences. And yet this is not the case when something tries to beat you, there is no malicious agent involved.

What other games you play that you think offer further examples?

So, what of this isn't true of, say, Minecraft?

Or Dwarf Fortress...

How far afield (not just down) can you go in DF? IIRC, the world in Minecraft, if fully generated, would put Ringworld to shame.

A large DF world is approximately the size of the moon, according to one guesstimate. A pocket world is 153 square miles.

But completely exploring even a pocket world would be an immense undertaking. Scale is much less important than depth of content - the first ocean you find might be exciting, but the tenth is just another ocean. And that is where Dwarf Fortress does quite well, with a finely detailed simulation of the world and its history.

Wow, DF is much much larger than I had thought. There is behavior going on in the background in Minecraft, but from my highly non-expert position on both games I suspect that Dwarf Fortress has more intricate background behavior.

To give some idea of the amount of background detail, here are some bug fixes/reports:

Stopped prisoners in goblin sites from starting no quarter fights with their rescuers Stopped adv goblin performance troupes from attacking strangers while traveling Vampire purges in world generation to control their overfeeding which was stopping cities from growing Stopped cats from dying of alcohol poisoning after walking over damp tavern floors and cleaning themselves (reduced effect) Fixed world generation freeze caused by error in poetry refrains Performance troupes are active in world generation and into play, visiting the fort, can be formed in adventure mode Values can be passed in writing (both modes) and through adventure mode arguments (uses some conversation skills)

The article looks like they're trying to say how awesome their game is.

Any TECHNICAL difference between that and, say, a decent rogue-like algorithm? I have a feeling that it's scaled-up rather than technical up.

Also, couldn't see a GitHub link, so I'm assuming this is proprietary and therefore have no reason to trust whatever they say.

Yeah, it's like they are making a miracle of "the result of rolling a die is a surprise, even to the person who rolled the die".

An example of a technical move forward would be a game world that is so large it must be procedurally generated, that also has the two properties that it is massively multiplayer, and that players can arbitrarily alter the environment.

You'd get the technical challenge of reconciling player-made alterations to the environment with the "untouched" version of the environment according to your generative algorithm. Then you'd get the additional challenge of sharing those changes across lots of different players in real time.

I don't get the sense that either of the two properties (massively multiplayer and alterable environment) are a big part of this game.

If a game with all three properties (procedural generation of a large universe, massively multiplayer, and alterable environment) were to be made, it'd make me take a harder look as simulation arguments.