Runaway AIs

[haphazard1]

I have been playing some personal games to try to improve my skills and put in practice some of the concepts learned from the recent OSGs. (Thanks to everyone for all the helpful advice and information -- these games have been great learning opportunities! )

This afternoon I got stomped by a runaway AI. Large galaxy, 5 opponents, playing as the Meklar. Started in the southeast quadrant, OK but not great starting region with only 1 world in initial range (a decent Steppe 70) but three more habitable planets within range 5, plus three barren worlds nearby. Of course, the Meklar are poor in planetology, so no Controlled Barren. Slow start, needing range 5 or better planetology tech (which goes so slowly as the machines).

What I did not know until later was that three of the AIs were struggling even worse than I was with expansion. Only two AIs really managed to grow: the Darloks, who were busy expanding right into my face and grabbed those barren worlds...and the humans, who ended up with free run of half the galaxy. I got the early warning that the humans were growing fast, but it was GNN announcing they had hit 18 systems (right as I was colonizing my 6th world ) that really made it clear that this game was not going to end well.

I tried to stick things out for the learning opportunities, and being the Meklar meant I had enough production to not fall TOO horribly behind (although I was behind, and by quite a lot). Poor planteology continued to be brutal -- terraforming tech? What's that? Never heard of it...nothing until +60. So all my Meklar bonus and IRC4 did was keep me sort of even (on a per planet basis) with the much larger terraformed worlds of my rivals. But with the humans having 3 times as many worlds...well, there was not much hope. Their population soared, and despite my best efforts to grow my worlds (I did have Atmospheric and Advanced Soil) it ended with a council loss in 2475.

I thought about trying something military to gain tech and worlds, but my neighbors were the Darloks and Humans who were in alliance. I did not have the tech or fleet strength to crack their defenses. Spying against them was also pretty fruitless, with the Darlok racial bonus and the human's ever-increasing tech lead.

I eventually (about 2440) finally made contact with the Alkari, Bulrathi, and Psilons who were in the far northwest after I teched to range 10 and built a long range colony ship to grab a toxic planet in the southwest corner to extend my range. Was able to get Terraforming+20 from the Psilons in trade (poor brains only had 2 worlds), but the humans were so far ahead by that point that it was not nearly enough. Council defeat followed shortly afterwards.

What is the worst runaway AI you have ever faced? Were you able to make a comeback? Share your stories of overcoming tremendous odds...or at least share the misery of getting stomped by an out of control AI.

[shallow_thought]

My experience on impossible is limited, but one of my recent learning games lead to humilation.

Psilons (easy, right?), small, impossible. I can't remember the start too well, but it wasn't too bad - off at the far east, able to expand out and grab four-five planets with early techs.

There were three opponents. One was the Meklar. They got to six worlds before I could get my second colship out. I thought that their poor planetology tech would mean I had a chance to grab the hostile worlds in the second wave, but no... they developed a huge stack of planetology tech to go with the rest of their goodies and got to 12, then 18 while I topped out at six. And that was on my second play of the start. On the first, the Erratic borg had decided to make me an early target with their huge, duralloy, fusion bomb equipped ships before I could get PSV up. On the second, I managed to survive by tweaking my tech path, and was a clear second (ahead of the one-and two-planet also-rans), but ended up cowering in bunkers hoping that the massive Meklar fleets would pass me by...

In the end I shelved it, and managed to delete the save while tidying up .

Still, I do think that Meklar with good planetology tech can be terrifying. It's rare that they can establish that production edge early and snowball, but when they did, it was terrifying.

[haphazard1]

I agree that if the Meklar manage to get a sizable number of worlds, either through map luck with habitable planets or by unusually good (for them) planetology, they can end up becoming very scary. Their usual production edge means they spend tons of time building factories in most games rather than expanding, but when they manage to do both...look out! Checking the status graphs and seeing the Meklar leading the galaxy in population is generally a very bad sign.

[Sirian]

Although it is not the worst runaway AI that I ever faced in MoO, it was well documented, and gives a good account of the game. Meklar Corner

The worst runaway AI that I ever faced in a 4X game did not occur in Orion. While I expect you may have seen that Meklar report before, THIS GAME might be new to you. Enjoy the read!


The empire-building genre isn't what it used to be. In today's environment, I cannot seem to find (or even work on) a 4X that pleases me personally as a player. Maybe I just outgrew the genre. I held on to the idea that it was still for me, if only the right game came along, but in recent years I have shed the hope. It's been more than 15 years since I played a 4X that tasted like ambrosia to me, and even the old ones that I once loved so dearly have issues I could not stomach to endure any longer.

There's Stellaris at one end, which is a giant sandbox with a lot of empire-building mojo to it, but the opponents are passive, the gameplay boundaries feel klunky, and it's really more of a sandbox planet builder than a 4X game.

On the other end are streamlined games like Civ has become, but also lacking AI opponents that are competent enough to be fun to play with.

Building up your own stuff can be a lot of fun, but there needs to be a grand purpose behind it all, or you just moving sand around with a toy shovel.

Endless Space 2 will get a look from me. When it gets out of Early Access. Other than that, I am completely dormant on the 4X front as a player, these days.


- Sirian

[Sirian]

Sirian, have you posted somewhere what these MOO or other 4x problems are? Other than AIs which are indeed poorly done, predictable and not fun. From the ones I've played I like MOO best because of its simplicity and at the same time diversity but the AI again is awful. I'm hoping that when Remnants of the precursors is ready the AI will be well modularized and easy to improve upon.

The cracks in MOO (the original) really started to show up under the scrutiny level obtained when I launched the Imperia.

Watching the same game played by different players to similar outcomes, only to end totally different based solely on dicerolls for who voted how in certain elections was one of the most painful. I had previously believed the votes to be more merit-based, but the evidence revealed otherwise -- steadily so, across enough games, to take some of the shine off the whole design for me. (The difference in some of my closest wins and losses, which I had thought to be more merit-based, in the end come down to some good or bad luck. Period. ... That is something I would aim to fix in any 4X design to which I contribute.)

That wasn't the only item, either.


As for other games, there's a bucket of stuff. Civ still has core issues with snowballing, with expansion control, with the franchise obsession with penalizing its core gameplay of fighting wars, with physics issues involved in its multicurrency economy, and with unit stacking issues. That's only a partial list. Every space-based 4X runs afoul of the MOO2 blind alley of having to make all the space combat look good (as a priority over all else), along with nary a one having an AI worth even a grain of salt. There's a recent trend of introducing pure rock paper scissors model in to the combat engines (Endless Space's worst fault, which they doubled down on in patches that actually made their core game less fun to play -- then look at GalCiv3, with its pure R P S combat)... and-- I could go on and on.

I don't expect perfect games, but I am not even seeing any *effort* to improve on core problems within the genre. Evolution for 4X has spun away from improving the strategy to improving the trappings and eye candy only, and that isn't a genre that interests me, as a player. (There is better eye candy to be had in other genres, for games where I want to see pretty.)


- Sirian

[RefSteel]

Interesting points, Sirian. What would a merit-based election system look like? That is, what would AIs take into account? In Remnants of the Precursors (a near clone of MoO's mechanics, rewritten from scratch, currently in alpha testing) RayF tracks AI relations in two dimensions: How much they like each of their rivals, and how much they are awed/intimidated by them. I could imagine a system where the AI abstains unless they are very friendly with one of the nominated parties or greatly intimidated by any race (nominated or not). Then if they vote for anyone, it will always be the one toward whom they feel most friendly. To keep every vote from ending in mass absention (and provide a little information about the state of relations) AI races might also calculate whether there is a chance (at the time of their vote) for either party to win, and if there isn't, they might just vote for whoever they like the most, especially if the gap is large and relations with one of the parties is reasonably friendly. And if they feel sufficiently intimidated (again, by any race) and calculate that one nominated party can't win, but the other still might, they could vote for the possible winner, preferring a New Republic, even led by someone they're not in love with, over the risk of extermination.

(Another hugely important improvement to voting and AI diplomacy is the elimination of "alliance cheese" where AIs ~randomly join and break alliances with each other throughout the game - a big aspect of the randomness that dictates the vote in original MoO....)

Is that the sort of thing you wanted to see, or did you mean something different by "merit"? I definitely liked the idea in your reports, no matter how well or poorly it's supported by the mechanics of the actual game, of AIs recognizing when you've effectively won the game and responding by handing you your Council victory, and it would be neat to see that actually play out in a game.

[Sirian]

I definitely liked the idea in your reports, no matter how well or poorly it's supported by the mechanics of the actual game, of AIs recognizing when you've effectively won the game and responding by handing you your Council victory, and it would be neat to see that actually play out in a game.

Diplomacy is a complicated affair.

Game theory only offers one track of insight, and that's aimed at the multiplayer situation, where every faction is playing to win the game, rather than roleplaying being leaders who have an agenda for their civilization's future well being.

Civ has tried to quantify diplomacy -- something that must be achieved in order to make a video game portray satisfactory diplomatic encounters. But I do not feel the effort did as well (in any of its cases) as the original MOO did with just its dice rolls (which I still do believe were informed by at least a couple of things.)

The problem with MOO1 votes is that you could take the same saved game, play it forward X number of times with identical moves, and see different vote outcomes. That, I believe, is too much on the dice -- at least to the point of the PRNG not operating on the same seeds. Perhaps that is all it would take, as even well-informed intentions to quantify diplomacy by simple metrics have all done worse than the dice-involved method.

In MOO, there were things that could hurt your chances of winning support. They were not permanent, though, as a shift in alliances could see you on their side again and them voting for you, despite being bitter enemies at some earlier point. Civ has never had that. Civ made its penalties "unforgivable sins", which is too much game theory and not enough roleplay.


What would a merit-based vote look like? That's a great question. There are countless answers, more than one of which could provide players a great deal of entertainment and value. I'm just not of a mind to do the leg work on that kind of design and hand it over to others. My days of being a pure fan are behind me now.

I got the chance to work on the most recent Master of Orion, and I got to help them improve their design. But it was an advisory role that started not long before release, so I could do only so much. Perhaps somewhere I'll get a better chance at having a larger impact. For sure, I do not think the genre on the whole has made a good showing for itself. I did have fun with Stellaris -- it has some merit -- but it doesn't scratch the itch that MOO1 did. Endless Space came closer but is one of those games who got less fun with better AI, by way of better highlighting the flaws in their design rather than letting you trample over them, as it did initially.

The genre drove down a blind alley in its infancy and is just stuck there. That's the bottom line. And I have more interest in spending time with my family than I do in chasing the dream of trying to reboot the space 4X genre.


- Sirian

[RFS-81]

Civ has tried to quantify diplomacy -- something that must be achieved in order to make a video game portray satisfactory diplomatic encounters. But I do not feel the effort did as well (in any of its cases) as the original MOO did with just its dice rolls (which I still do believe were informed by at least a couple of things.)

According to the strategy guide, the dice rolls are informed by your relations. If an AI is allied with one candidate, and not the other, or at war with one and not the other, the vote is deterministic. Otherwise, if you have good relations, there is a chance that they vote for you, and if you have bad relations, there's a chance that they vote for your opponent. So it's indeed merit-based in some sense (if you break treaties left and right, you could end up handing free votes to your opponent) but still very luck-dependent. I don't know if fixing the PRNG seed really would make it better. In the end, the outcome of the game is still determined by a die, you're just much less likely to notice it because reloading doesn't make a difference. One could argue that perception is everything, and if the player believes that the outcome is more merit-based than it really is, that's good enough, but I don't like that way of thinking.

I got the chance to work on the most recent Master of Orion, and I got to help them improve their design. But it was an advisory role that started not long before release, so I could do only so much. Perhaps somewhere I'll get a better chance at having a larger impact. For sure, I do not think the genre on the whole has made a good showing for itself. I did have fun with Stellaris -- it has some merit -- but it doesn't scratch the itch that MOO1 did. Endless Space came closer but is one of those games who got less fun with better AI, by way of better highlighting the flaws in their design rather than letting you trample over them, as it did initially.

I played Endless Space a bit (don't know if that was before or after they improved the AI), and it seemed decent, but I can't say I know the game in depth. What's wrong with the design?

[Sirian]

I played Endless Space a bit (don't know if that was before or after they improved the AI), and it seemed decent, but I can't say I know the game in depth. What's wrong with the design?

ES uses a pure rock paper scissors combat model. That's the flaw. It's not subtle or obscure.

The AI used to defend against at most two of the three at a time. When they upgraded the AI, they improved its capability to defend, allowing it to defend against all three, and making its shipbuilding more responsiive to whatever the player is using. These two upgrades made the AI smarter and play stronger, but it turned a combat system that was shallow and dull in to one that is mind-killing and tedious.


Endless space used a clever one-two design punch to get around the "must have eye candy for space combat" law that MOO2 wrote in to the genre.
1) Not really have combat. (Combat was SO shallow and simple, it didn't take up much of the player's time. Player time was largely focused on the economy, which is quite decent, making it a fun empire builder.)
2) They used cutscenes to portray the combat, which again is SO simple, it's doable. Throw some pretty models, they make ONE pass by each other, three rounds of combat during the pass: win, lose or draw.

The art isn't spectacular and the fights are boring, but you can watch capital ships fire pretty stuff at one another, and some ships blow up.

With the original AI, you could prosecute a war in a traditional way. With the new AI, it's like the borg: fire your phases a couple of times, kill some stuff, oops they've adapted. You have to modulate your phasers again. Truly local wars are still reasonably decent, but anything with travel time becomes increasingly tedious. There are all manner of ways to combine arms, rather than mass producing a single design, but the overhead of that, from research overhead to manufacturing overhead to unit support overhead to logistical overhead just sucks all the fun out of the experience. At least for me. It's even true on defense, too.


The game was a fun empire builder, and then it turned in to tedious tactical micromanagement exercise. The AI got better and the game got (much) worse.

In my opinion, they should have stuck with the strength of their game, which was leaving combat painless and invisible and requiring little attention. The rest of the game was pretty good. The combat never was nor could it be (which is what what good about the initial version), but then they forced you to pay a LOT of attention to the combat. For me, it was the death of the title. I'd rather play Stellaris at this point. (Not high praise.)


- Sirian

[Sirian]

I said that I would check out Endless Space 2, but there are multiple warnings about a fatal crash bug that halts and destroys games. Assuming they fix that, it looks worth a shot.

The game didn't come on sale with the Steam summer sale, though. It was just released to full release, so it may be a while before it does go on sale. I may or may not wait that out, but for sure I want a fatal crash bug fixed before buying. If anyone hears more about this, especially news of a fix, let me know. Thanks.


- Sirian