Creating and breaking immersion - did PDX break their formula?

[Stadtpark]

All those threads about how the AI is failing in Stellaris and cheating to make up for it make me think.

The thing about PDX grand strategy games was always to create a sandbox, where complex flavors (history, politics, economics, warfare, religion, internal politics / society) were ingeniously mapped to strict but relatively simpler rules /systems (e.g. the different flavors of EU4 nations and ideas to choose from in the beginning mapped back to all the same game-systems, giving slightly different bonus or malus here and there), so that the mutliple interacting AIs by following the rules would unfold an "emerging alternate history" / "tell complex stories" back to the player simply by following the rules of the game, from very diverse (historic) starting positions.


Then came HoI4 and Stellaris, and both seem to be extending the formula to (or beyond) its breaking point:


HoI4 with its National Focus Trees was implementing "history on rails": more and more scripting that was interfering with the simple "rule-execution from a given starting point"-formula I described above, in order to influence the possible stories the sandbox was able to tell towards either more historic accuracy, or towards more interesting pre-defined outcomes. So not only the starting positions of the different AIs were very deliberately choosen (to match history like in EU4), but also the very paths they would take.


Stellaris tried to go the other way:
At first it seems to start from a clean slate: the starting positions in Stellaris are designed to be relatively equal.
So the complexity / differences in later empires would need to come from more diverse rules - which inevitably would present balancing problems (- whereas the balancing problems in EU4 and HoI4 would come from starting positions): the different flavors this time would map to actually different systems. At launch of Stellaris even the modes of travel were different (remember that?); the diplomatic interaction viabilities would depend on government / ethics preferences (up to the point were you can't even do diplomacy with half the galaxy); the choice of planets to settle / empty space to expand into would depend on habitability and bordergrowth (remember bordergrowth, and how unpredictable it was?); the different economies to build could be based on different ratios of resources, since the POPs would produce and consume different (- machines needing energy instead fo food; POPs eating other POPs as food), even the way to acquire more POPs is different: natural growth vs building robots vs abducting slaves, vs attracting migrants.
The possibilities in Stellaris seem endless, and the AI would need to act very different depending on what sort of governing authority and ethics it has ,and what sorts of POPs it consists of, and what sorts of traditions / ascension paths it would like to choose from, what sort of "victory" it would try to attain. - You'd pretty much need a different strategy for every different compositionof strengths and weaknesses / limitations.

For the player that sort of variety in the game-mechanics is great - with all the great flavors that come with it, he can tell his own story each and every way he likes = maximum immersion.

But after reading too many threads about how the AI fails (and cheats to make up for it), I get the impression that the AI empires not only fail to put up a competition / challenge for the player without cheating, they even seem to fail in ending up coherent stories of their own: maybe the rules are too complex / allow for too many paths for the limited AI to actually choose a consisting one? - The problem with failing AIs is, that they break immersion when they break the rules.

So the same complexity that allows for maximum immersion within the players own empire feeds into a breakdown of immersion as soon as he takes a closer look at the AI-empires and realizes that they are cheating / not following the same rules.

When the AI-Empires behavior and abilities don't stem from the same game-mechanics, but must be sustained against those very same limitations by cheating, the "natural emerging stories" / "alternate histories" that PDX came to be known for, stop actually being "natural emerging" from the background of geography and history and rules. - While trying to introduce a geography (by limiting travel to starlanes early on and grouping stars into clusters), and introducing history / starting conditions (with Advanced Start-AIs and Fallen Empires and End-Game-Crisis), the part of braking the rules / cheating AIs makes the other two points moot! - So please: Improve the fucking AI, so that you can set it to following the same rules again! - PDX - you are gambling away your legacy!

 

[sillyrobot]

The AI can do a whole lot better than it does. Glavius' mod is a testament to that. Unfortunately, a good AI doesn't move boxes off store shelves so there is little business imperative to fix it as opposed to doing something shiny like new humanoid portraits.

I've also noted a seeming desire to build a "correct way to play" that has been working to shrink the sandbox. Part of it comes from a sensible desire to improve game performance by limiting choice (drive system, number of inhabitable planets, default system link levels coupled with limited testing/balancing of secondary games systems against legal values). All of it tends to reduce the possible stories available.

 

[Arutar]

While the A.I. ofc needs fixing you are not seriously suggesting that even the most competent game A.I. today would be able to provide a decent challlenge without any cheats, do you?

The most powerful computers have learned to beat humans at chess and recently majong, which are orders of magnitude less complex than games like Stellaris.

Seing what openAI does in Dota, maybe in 10-15 years ?

 

[Pyoro]

Making a perfect AI presumably wouldn't be that difficult. As soon as the players starts outgrowing everyone else they'd be identified as the main threat and murdered by all AI nations who of course would always play optimal.

What's difficult to make is an AI that's fun to play with. That makes some mistakes but not so many they are hopelessly outclassed. That does different things according to "personality" they are playing but ones that aren't so different they completely sabotage themselves.

I don't think the issue is the formula of the game per se. I just think it's in a sorry state that needs to and hopefully at some point will be fixed.

 

[Fisk]

Keep in mind that literally everything you do rests on your economy, and the economy was recently completely rebuilt. PDX have been constantly improving Stellaris since launch, and from what I've seen they tend towards creating small problems while fixing big ones. We've also seen them improve the AI before, suggesting they'll do it again with the new eco system.

I'm confident every existing problem will be fixed, and new problems will be created. Rinse and repeat until they stop supporting the game.

 

[Stadtpark]

I guess you are right.
Maybe my posting was more motivated by myself getting disillusioned by visiting the forum and reading about cheating AI, than the immersion breaking coming from the game itself: if I could go back into that state were I either wouldn't know or would not care about AI breaking the rules, my immersion could be restored.

Still I think the story about PDX fame coming from artfully applying strict rules to geography and history and letting the sandbox play out in games that are equally wide (lots of atmosphere / flavor) as deep (complex interacting systems) holds true. And failing to deliver on geography, history or in this case the actual application of the rules in competing AIs is not compensated by making the scope of the game even wider by simply adding more and more stuff. - Having actually functioning rules is at least as important as having functioning side / start conditions, and I'm not sure if I shall call the rules functioning, when they are only applied for the player, but circumvented by the AI, just for simpler / better story-telling. - I'd rather have a little more boring stories, but have them emerge from the sandbox itself: if I want pre-planned stories, I buy Virtual Novels or Single-Player-RPGs.

Edit: Maybe we still can have emergent stories / histories with the AI circumventing some rules or getting a lot of boni, but at what quantity or quality would you consider it arbitrary?

I'm ready to accept a "Deus ex Machina" - solution as an end-game-crisis for dramatic purposes (or even as a mid-game stop-gap like awakening Khans or Fallen Empires), but I'm less willing to start with such things in a sandbox-game from the get-go! - "Here for a start we challenge your creativity with limiting rules and circumstances in a sandbox environment", but once you have largely overcome them, we throw in first a little arbitrary challenge and then a large one. - Oh: "And btw.the limitations and rules we put on you in the growing-up / sand-box-phase didn't really count for your "equals" / AI competitors, because we couldn't make them good enough to be competitive..." - that in my book still mostly counts as a failure, and less as a "we gotta keep things thrilling, therefore no-rules-apply"-get out of the prison-card.