I would like to hear what people here think: Would this game benefit from some sort of built-in achievement system?

Obviously this game is about discovery: The community very quickly discovered the futility of creating wiki pages for each individual animal and planet, because we realised that every such seemingly unique discovery was just a recombination of certain known prosaic templates. So people smartly agreed upon animal/plant/planet/system/galaxy/crafting materials/ships shape category names, and documented and discovered these instead.

Certain types of “discoveries” can be literally found in the game files, if we wanted, we could look up how many template pieces exist. We know there is not one more type of exotic planet or of hazardous plant, and there is a fixed number of Hacked Factory Puzzles, Celestial Archive Records, and Remembrance Records, etc., they could simply be read in the game files, correct?

On the other hand, there’s the funny text descriptions of animals, or the mission summaries, or the little snippets you get when scanning sentinels: They sounded unique until we noticed that they are just a recombination of the same x sentences with some words swapped out. That’s a case were I don’t feel I need to track and discover all snippets - but maybe others track them?

The (fixed number of) Observatory and Factory puzzles (and their answers) are one thing that people probably most commonly look up on the wiki. Some are too easy (after you figured out the pattern), but for others, it forces you to learn a language first to get the hints. Not sure in which category these belong - do you feel solving these is a mini-achievement each, or not?

My point is: Most enjoy finding these themselves, and at the same time, we are aware that someone else of us literally has the file open in front of them and could tell us whether there are more or not. — It’s such an odd clash of worlds where one’s fun is the other’s spoiler. (Obviously the data miners here are nice and mark their threads as spoilers. )

In this context, do you think this game would benefit from some sort of built-in achievement system? (Either per account or per game save)

HG halfways started such a system when they created this Discoveries page that lists all craftable and tradeable products. We know to stop looking when the list is filled in with icons.

Only that it doesn’t contain icons that show you “you just discovered the last of x types of exotic planets!” Or what about a built-in “book of records” that looks like x empty lines, and every time we visit a celestial archive or what ever, it fills in one more line? So we feel we have discovered it, and at the same time, we see whether we have “the full set already”, without having to navigate dangerous spoiler threads?

Would you want to know (and have it shown in your discovery page) that you have discovered all of a type?

We have Milestones but they stop at level 10. I would like to see them expanded. They should at least go to level 100 but I think most of us would have exceeded even that. More categories would be welcome. I like having a goal to work toward or a limit to push.

I don’t feel a need for a scorecard that when completely filled out indicates a completion of any sort. I bought NMS when it first came out because it offered a way to explore many new things at my own pace and did not require much in the way of skill.

I like looking for new things - even old things might be worth a photo if they’re in an interesting setting. If I knew that I only had to find one more particular thing and then I found it, it would take some of the enjoyment away knowing that no matter where I went that there would no chance of finding anything new for that particular thing.

There are things that I like to complete, but NMS is not one of them.

@AdaRynin , I fully agree with you. The supposed diversity of the games’ current procgen rapidly becomes predictable, because of the limited number of base parts the system has to work with.

And yet I suspect it doesn’t have to be this way. And I believe the problem is one of scale. For example:

To make a procedural creature, we start with ten different bodies, which can be paired with any of ten sets of legs, ten heads, and ten pairs of arms. Each of these parts can have one of ten colours, and one of ten textures.

Theoretically, then, we have 10x10x10x10x10x10x10x10x10x10x10x10, or 1,000,000,000,000 possible creatures.

But when we try it, we rapidly find that all these creatures look very similar. Subtly diffterent, yes - but all minor variations of a few major themes - and the system loses its ability to surprise.

Now consider music. There are only12 notes available - and yet every piece of music ever written consists of those same 12 notes, arranged in different order.

Consider writing - there are only 26 letters in the western latin alphabet*, and yet those 26 letters make up everything ever written. Every book, play, pamphlet and essay. Every poem and song.

The problem, of course, is organisation. It is child’s play to generate random strings of musical notes - but the results are seldom anything people would want to listen to. It is equally easy to have a computer create random arrangements of letters - but this hardly ever results in an engaging story.

On the one hand, we have building from predetermined parts - which, we have seen, rapidly becomes predictable and boring. On the other hand, we have the random assembly of small component parts, which, whilst definitely unpredictable, creates incomprehensible nonsense.

The answer, I think, lies in AI. An AI that undestands the rules of language, grammar, and causality, could write understandable stories. An AI that understands the rules of scale, rhythm, melody harmony, and phrasing, could generate acceptable music. Programs exist that do these things. I even wrote a few myself, back in the day.

It would be entirely possible to create an AI capable of designing one-off, unique creatures, buildings, flora and landscape features. The problem then becomes one of designing a game that actually works in such an unpredictable world. How does the player interact with buildings that can only be entered by someone two inches tall, and triangular?

*Yes, OK, some alphabets have slightly more or less - it doesn’t affect the principle.

I have changed the topic category to be NMS

That “entirely” makes the possible sound a bit too effortless, I think. The problem you have with AI is not what it can do. The problem with AI is what you can train it to do. For training, you need data sets. Which are extremely time-consuming to create, and therefore very expensive. Of the digital commodities out there, training data for AI is among the most valued.
And then we’re not just talking one system here, we’re talking several. First, one that comes up with the basic creature concept based on whatever input parameters, then one that can take that concept and actually generate meshes for it, then probably one to animate the whole thing… The challenge is staggering, and I don’t think it’s one that can realistically be taken on by such a small studio. We’re talking about a job that the likes of Amazon or Google wouldn’t chew down without swallowing pretty hard a couple of times…

Thet depends very much on what you are asking the AI to do. If you are designing (for instance) an AI to talk real-time to medical patients, ask relevant questions, and then provide a realistic diagnosis, then I agree with you. The number and scope of possible variables is so vast that even large organisations would find the prospect frightening.

On the other hand, an AI to predict the flight of a table-tennis ball can be (and has been) programmed into a very simple computer.

NMS populates fantasy worlds with imaginary creatures. The normal laws of physics and biology are largely ignored. We don’t need to worry about how these creatures feed, or shelter, or reproduce. We don’t need to explain what senses they have, how they survive the environment, how they support their own weight, how they manage to fly, burrow, or swim.

The parameters for a creature design AI are therefore much more limited than they would be for a creature that could actually exist

When Hello Games came to test No Man’s Sky, they quickly realised that there were simply too many possible worlds for human testers to visit them and check that they actually worked as intended. So they created software robots to visit millions of planets, and send back photographs. A similar setup could be used to test AI manufactured creatures - spawn them into a test environment, and measure their performance. The test parameters need not be complicated. Such questions as “can it move about the test environment?”, “does it react to NPCs / player characters?” “is it too big for the play area?” and “is it too small to be seen?” would go a long way to provide a learning data set.

TL/DR: The AI need only be as complex as the application requires. NMS does not require great complexity, either in form or behaviour.

Yes. That’s about the scope I was thinking about. There still have to be rules, even if they’re purely aesthetic, and you cannot program the rules, you have to train them, that’s why it’s so much work. You have to teach it aesthetics so it can come up with new stuff, otherwise you’re still as limited as with the programmed, tightly supervised rules and pre-generated parts we have now, except it’ll look… well, probably not as good.

Yes, of course. And, as a backend developer, far be it from me to underestimate the value of solid test procedures. But you need something to test first. Well, not neccessarily, TDD and all that. But at least you need something to test after. And while I have worked on many a piece of code where the tests were more difficult to write than the actual code itself, this definitely isn’t one of them.
Well, ok, technically it is, since the code of a neural network is essentially dead-simple. It’s the bloody data that’s the problem. And now I’m feeling dizzy from walking in circles…

Ooh, AI generated procedural games, now we‘re talking! then I would not care about achievements either. I don’t see it happen any time soon though, still, AI could spawn new art forms one day…

I’m also pondering what the input data would look like. Meshes and animations and textures and navigation meshes don’t sound easily digestible for training. Could I at least generate some geometric stickfigures to walk across a 2D planet…? The idea is quite inspiring.

Funnily enough, in my mind, that’s the most likely direction of Hello Games’ latest project.

I mean, given their dedication to procedural generation, what could be more impossibly ambitious?

http://www.boxcar2d.com/

I’m thinking of a simple game like this. You can up or downvote designs you deem “acceptable”, and the second quality factor is how far the car makes it on the track.

Hmmm… there’s something to that thought… Though AI and procedural generation as we understand it are two very different fields, and I’m not sure how much the built-up expertise in one would help them with the other. Still, they’re kind of crazy, so… who knows?

For what it’s worth, while I don’t think your suggestion is feasible for creating creatures, I do think that something like that with buildings is more in the realm of the possible. The reason being, they have a lot of structured data from player-built bases. If they’d put in a tagging system so you could add tags to your base overall, and individual combination of parts, large parts of the training set would write itself. Still needs a hell of a lot of curation, but that’s doable.

Flash? Just how old is this page??

Managing Director Sean Murray spoke to IGN about the new game being developed by Hello Games, and said “it’s the kind of project that even if we had a thousand people working on it, it’d still seem impossible.”

Sean Murray, quoted in PC Gamer, 28 days ago:

“For a while now we’ve been working on something pretty ambitious in the background,” Murray said. “It’s a small team but we like it that way.”

Murray goes on to say that the scope of the project is probably bigger than you’d expect a small studio to cope with. “Similar to No Man’s Sky, it’s the kind of project that even if we had a thousand people working on it, it’d still seem impossible,”

Sorry, I don’t know, their phpBB3 forum is down, and their mailing list is literally a list of people whom he intends to send emails, but the author was last seen posting in R/equalrightsforrobots on Reddit 11 years ago, we could ask him there.