Logo Platform
logo amplifiers simplified

AI, what can be done about it?

Reply
Copied to clipboard!
9 years ago
Aug 20, 2015, 9:46:33 PM
wilbefast wrote:
256 empires in the same game? smiley: ohhO:



I'm afraid this would cause a host of UI and performance issues. The issue isn't address-space, we can spare 10 bytes for identifiers. The issue is that the time taken to calculate the next turn isn't linearly dependant on the number of empires, it's polynomial IIRC (or perhaps exponential). So we're unlikely to have more than in previous Endless games I'm afraid.





I'd like us to focus on feature requests here rather than implementation details - figuring out what the best architecture is is our job smiley: smile





Interesting idea! We'll have to see what Jeff thinks of course.





There are a certain amount of things that can be tweaked in Steam\steamapps\common\Endless Space\Public\AI and \Steam\steamapps\common\Endless Legend\Public\AI respectively - best refer to the corresponding mod forum to find out what affects what. We're likely to have parameters made public in ES2 as well.



[HR][/HR]



With regards to the API: yes, it's an interesting idea - I'd like to stress though that my saying "that's an interesting idea" is by no means a promise that we will have any such thing smiley: wink






Sure it's your job, I just like to daydream about such matters until I finally get to release my first game. I like to wonder about such grand issues. My core AI request is: use multithread and opencl wherever possible, even for AI if applicable. Such up every bit of parallel performance you can, not just for graphics-related stuff.



It's great to have such a direct communication with you guys! I love this level of transparency.



AI & early colonization & culture game aspects:



It would be great to have a 'cultural grasp' mechanic in the game where military is not the only way to take over a planet/system. some systems could be culturally shared and maybe even have mixed governments with each faction controlling part of it.



There should be a development limit for multi-racial colonization at early development stages. After the first race settles other races should still have the option to setle in the same planet, as it's still a colony. Going to war should be optional, and those early colonists could evolve into a bitter grudge or a great bridge between factions. It would make cultural tolerance and interchange easier/more tolerable as people get used to it.



There should be a way to make a cultural 'invasion' as well if the other factions have a weak culture. This would give the player bonuses for trade. Cultural domination by some other faction should be a double-edged sword by the culturally colonized granting some form of advantage if the colonized chooses to embrace it as a strategy. A cultural invasion would not overthrow governments but would sediment alliances and trade agreements very solidly. it would be a way to ramp up the costs of going to war for the one who declares it, possibly with a temporary, but huge unhappiness penalty. There should be a whole metagame just about cultural influence.
0Send private message
9 years ago
Aug 21, 2015, 12:12:32 PM
Brazilian_Joe wrote:
Sure it's your job, I just like to daydream about such matters until I finally get to release my first game.


That's fine! It's more in the interest of keeping the thread as a whole on track rather than to you in particular. Perhaps "what can be done about it" is not the best title.



Brazilian_Joe wrote:
Sure it's your job, I just like to daydream about such matters until I finally get to release my first game. I like to wonder about such grand issues. My core AI request is: use multithread and opencl wherever possible, even for AI if applicable. Such up every bit of parallel performance you can, not just for graphics-related stuff.


You be surprised by the degree to which technological choices which seem like non-brainers may actually be sub-optimal because of other factors in the context. The choice between multiple AI threads vs performing calculations over multiple frames (I'm not actually sure what the technical term is for this - maybe time-division multiplexing) vs a single AI thread that does all the work vs doing AI in the main thread with a GUI thread to ensure interface responsiveness... it's not as obvious a choice as one might think. You really need to be up to your neck in the implementation details to make that call. We'd rather you guys remained detached from all that so that you're objective - so you can ask for the things you want to see without censoring yourselves.



That being said 256 empires is a lot :P



Brazilian_Joe wrote:


It's great to have such a direct communication with you guys! I love this level of transparency.



Cheers smiley: smile We do try!





Brazilian_Joe wrote:


AI & early colonization & culture game aspects:



It would be great to have a 'cultural grasp' mechanic in the game where military is not the only way to take over a planet/system. some systems could be culturally shared and maybe even have mixed governments with each faction controlling part of it.



There should be a development limit for multi-racial colonization at early development stages. After the first race settles other races should still have the option to setle in the same planet, as it's still a colony. Going to war should be optional, and those early colonists could evolve into a bitter grudge or a great bridge between factions. It would make cultural tolerance and interchange easier/more tolerable as people get used to it.



There should be a way to make a cultural 'invasion' as well if the other factions have a weak culture. This would give the player bonuses for trade. Cultural domination by some other faction should be a double-edged sword by the culturally colonized granting some form of advantage if the colonized chooses to embrace it as a strategy. A cultural invasion would not overthrow governments but would sediment alliances and trade agreements very solidly. it would be a way to ramp up the costs of going to war for the one who declares it, possibly with a temporary, but huge unhappiness penalty. There should be a whole metagame just about cultural influence.


Interesting stuff! Culture, propaganda, brain-drain, etc. All good stuff. Best put this to MeeDoc and Metalynx though in the appropriate GDD thread - or create a thread if you want to go into detail smiley: smile
0Send private message
9 years ago
Aug 22, 2015, 3:29:20 PM
BPrado wrote:


I think it might work much better for tech path than army designs, even because the first heavily affects the latter.



It works best if both the "expert player" tech path and army designs used are integrated. The only reason I did the AI Improvement Mod for Distant Worlds was because the developer provided us with full access to the AI's research orders. Only then did I feel enough of a difference could be made to make the effort to do a Mod. There is no point emulating the tech paths of good players if the designs don't take advantage of the equipment those technologies provides.



BPrado wrote:


Having an efficient army is relative to your opponents, I don´t think there are specific designs that are absolutely better in Endless Legends (with which many people here will disagree).



The AI's units in Endless Legend (for example) often failed to use strategic equipment. Most of the time their armies were relatively small. It's been some time since I played Endless Legend, but I do recall the impact of initiative, enabling my army to often blitz the AI before they could even get a shot in. As a human player, it's obvious that getting better equipment and army sizes is a priority, yet the AI at best was very slow in doing the same. Every game felt like a curbstomp of the AI and this only changed when the bonuses became immersion breaking. It's much more fun to have an AI that can do a decent job with less immersion breaking bonuses.



Only Amplitude can diagnose the root cause of that. I can only speculate. Looking at the xml's there are a huge volume of weights the AI seems to use. Which makes me wonder, how much is actually scripted in the Endless Legend AI? Is it designed to be excessively adaptive? Or is there a "base" script and those weights are used to select variations? The latter seems much more likely to me to be able to emulate good players and put in a good show. I would START with the AI unable to adapt. Get it to kick ass in a scripted fashion emulating as much as you can from good players. Then very slowly, very incrementally, give it some various abilities to adapt ... but without killing the heart of the script i.e. building decent armies / fleets etc.



With Distant Worlds the AI research orders and ship designs are fixed for each race ... there is no adaptability at all (although there is a lot of variation between races). And despite playing the game for many moons I still get beaten from time to time (without feeling like there are immersion breaking bonuses) because the AI is completely focused on it's objectives. It uses my own research orders and ship designs against me. Would I like to be able to Mod some adaptability? For sure. But this very simple scripted approach is light years ahead of Endless Legend and Endless Space AI in terms of performance.



So far. I encourage Elegant Simplicity.
0Send private message
9 years ago
Aug 26, 2015, 1:43:10 AM
wilbefast wrote:
Interesting idea, and there were several academics asking for open API for game AI at Nucl.ai. I'll discuss it with the team smiley: smile



What would constitute a "baseline AI" for you? We'd rather not give away all our technology (ie. open-source our AI code) and given the choice it's probably better to spend the time making said technology better rather than writing a new, simplistic version as an example for modders.




If players could plug in AI modules (that weren't just parameter tweaks) then it would really open the door to a lot of improvement.



I'm sure you've discussed this among yourselves and have done more research than I have, but I would be surprised if Sid Meier's Next Game is going to be interested in lifting any code-- especially any code encumbered by even a liberal license. And, just maybe, the people at Amplitude are more interested in helping independent, starting developers than they are in protecting their intellectual property.



There's an argument to be made beyond economics. I used to be involved in the Quake modding scene. iD, by opening up so many details of their engine, launched a lot of people into the business of game development. Which is a good thing. But even beyond that, they created a game where people could enjoy the game by learning how to model, how to design levels, how to code. With their decisions, iD created avenues for people to learn and actualize.



It's hard to say whether that was a wise economic decision for iD. There is certainly a lot of competition that wouldn't exist without that choice. Of course, it's also helped to earn them a core of devoted fans and sold a lot of copies of their engine. But economics shouldn't be the only motivation. For many people, helping others learn is at least as important as making money.



iD's not alone. I've had similar experiences with other games. ToME is probably a good recent example of a game that, by opening up all of its code, introduced a lot of people to art, code, game design. (And it's also French, if I'm not mistaken, so you might have even heard of it :P)



I understand that there are other constraints. But I hope that you, and the rest of Amplitude, are willing to entertain this idea.
0Send private message
9 years ago
Aug 27, 2015, 5:19:40 AM
I agree with BPrado a bit. No one has really devoted a lot of resources to a smart AI in these kinds of games. I recently played Beyond Earth and found the AI to be a super-duper joke.



However, I think it's possible. There was an AI competition on CodeCombat's platform about a year ago and a number of people created fairly decent AI. So, to natev's point, if there was a way to add some real logic to a game to create a home-brewed AI, people would do it.
0Send private message
9 years ago
Aug 27, 2015, 7:58:33 AM
natev wrote:
I would be surprised if Sid Meier's Next Game is going to be interested in lifting any code-- especially any code encumbered by even a liberal license


Hate to play devil's advocate (especially since I'm actually pretty pro FOSS) but if they did use our code who would know? This is the games industry: it's one giant technologically Mexican Standoff where everyone is saying to everyone else "ok, if you put down your competitive advantage I'll put down mine...". Which is ironic because I think most game developers are all for sharing tips and tricks.
0Send private message
9 years ago
Aug 31, 2015, 11:11:45 AM
I assume the Amplitude AI Team are monitoring the Stardock Gal Civ III forums where there have been plenty of interesting AI threads recently. The lowdown from my perspective is that there seems to this desire to have a very adaptive AI. As admirable as that is both the Endless game and Gal Civ III have AI that really struggles to do some of the basics. It certainly seems to me there is a need to walk before you can run, and if that means lots of scripts at release, so be it.



http://forums.galciv3.com/470928/page/1/#3585268
0Send private message
9 years ago
Sep 4, 2015, 6:50:51 AM
wilbefast wrote:
Hate to play devil's advocate (especially since I'm actually pretty pro FOSS) but if they did use our code who would know?




You're right, you wouldn't know. I mean, you could litter it with obfuscated, tricky tells, but that'd just be more work for you.



But whatever code you have is going to be less portable to, for example, Civilization VI than Civilization V's code is, and Civ's developers are going to have a much better understanding of their own code than they will of yours. Even beyond the fact that they'd be risking millions of dollars in judgment should anybody with any knowledge of the theft ever feel any moral pangs, or maybe just get angry about being denied a raise.



I understand that's not as true of a small developer, where they might not have an existing code base, and only one person ever needs to know about the theft. But even though this the games industry, with its share of silly lawsuits, this is also the games industry where there's a long history of developers intentionally declining to patent every imaginable "innovation"--can you imagine how much pain you'd be having if each successive 4x developer had tried to patent everything they could?-- the games industry where developers routinely share information about their design process on sites like gamasutra, the games industry where developers support and help each other rather than sabotage each other. Id's decisions trained just about every competitor they've ever faced. Unreal has released all of their code, to anyone! In a lot of ways, the games industry is an amazing industry where developers are routinely more interested in improving the field than they are in shutting out competition. Maybe, as a small developer itself, Amplitude is more interested in helping other small developers than they are in limiting competition.



Now, I'm not saying that there aren't any economic risks involved in that. There are risks. But part of the reason for behaving in the way that the greatest developers have behaved, the developers that are remembered, is because one cares as much about improving the quality of video games as one does about shutting out competition. Because one thinks its worthwhile to learn how to code or create digital art and wants to put the opportunity to do so in front of more people.
0Send private message
9 years ago
Sep 12, 2015, 10:12:39 PM
Are you seriously thinking Firaxis will steal code and ideas from other games in order to make the next Civ better?! That's frankly absurd. Just look at Beyond Earth, they disabled parts of their own Civ5 code. And I do mean disabled, not cut because the code is still there.





The last thing you have to worry about is Firaxis trying to do better.
0Send private message
9 years ago
Sep 13, 2015, 6:14:56 AM
I don´t understand what this has to do with anything, Falkner.



Why would they rewrite routines that work exactly the same way if they have them written already? Is it such a large problem that they might have forgotten to erase irrelevant parts of it (like they might have actual reasons for it)? Or is the problem here that the AI is mostly the same? In any case, doesn´t seem like there´s an open source issue. Seeing you quote Delnar Ersike pokes me, because apart from strictly technical stuff, like this, I don´t remember ever agreeing with anything i´ve seen him write about Civ5.
0Send private message
9 years ago
Oct 28, 2015, 7:14:22 PM
@willbefast

Are there any news on this topic?



If ES2 AI system will be similar to ES maybe it's possible to start some contest for limited AI that handle some obvious weak points of ES AI.(ship design and fleet composition management).

Contest 1 should be easy to implement as standalone application. Each next contest is expanded version of previous one.



Contest 1 - mid-game generic fleets. (simplified)

- generic race

- 2 small, 2 medium ship hulls available

- mid-game tech arbitrary picked for contest

- AI design one up to 10 command points and 2500 production fleet

- 10 rounds after each battle AI could redesign fleet with knowledge of past rounds.



Contest 2 - AI generic empire war.

- generic race.

- AI is given some amount of science(let's limit to 3 possible values early, mid and late game) and production(enough to build around 60 command points of ships).

- AI assign science pool to techs(simplified tech tree with only combat oriented techs) and use unlocked parts, command points and extra space.

- AI create multiple fleets that participate in battle. (diffrent AI could have diffrent amout of fleets and cp depending on ship designs)

- Battle: AI alternate attacks(first AI attack with one fleet then second AI attack etc.) till every existing fleet have used it's attack. side with higher fleet condtruction value after battle win.

- 10 rounds in each round AI can redesign it's fleet(can't change techs) with knowledge of past rounds.



Contest 3. AI empire war with races:

- similar to contest 2, however AI choose race.

- Race grant unique amount science and production and also provide bonuses to ships. (some bonuses that depend on game state have to be adjusted to it's typical value)
0Send private message
9 years ago
Oct 29, 2015, 4:36:49 AM
Falkner wrote:
Are you seriously thinking Firaxis will steal code and ideas from other games in order to make the next Civ better?! That's frankly absurd. Just look at Beyond Earth, they disabled parts of their own Civ5 code. And I do mean disabled, not cut because the code is still there.





The last thing you have to worry about is Firaxis trying to do better.




I'm not surprised at all.

It took me 30 minutes to realize that BE was just a reskinned version of Civ 5. It felt like I'm playing modded Civ 5 the whole time, rather an actual brand new game.
0Send private message
9 years ago
Oct 29, 2015, 9:24:24 AM
Mer wrote:
@willbefast

Are there any news on this topic?




Well re- FOSS: not a discussion we really want to be getting into here and now - my spider-sense is telling me that it's likely to wax polemical real fast smiley: rollblue



For my part I've been working on AI for EL lately, with MeeDoc and Metalynx taking the baton where ES2 AI design is concerned. I'll let them fill you in on exactly where things are as I don't think I'd do it justice smiley: smile
0Send private message
9 years ago
Oct 31, 2015, 4:38:01 AM
Icemania wrote:
It works best if both the "expert player" tech path and army designs used are integrated. The only reason I did the AI Improvement Mod for Distant Worlds was because the developer provided us with full access to the AI's research orders. Only then did I feel enough of a difference could be made to make the effort to do a Mod. There is no point emulating the tech paths of good players if the designs don't take advantage of the equipment those technologies provides.





The AI's units in Endless Legend (for example) often failed to use strategic equipment. Most of the time their armies were relatively small. It's been some time since I played Endless Legend, but I do recall the impact of initiative, enabling my army to often blitz the AI before they could even get a shot in. As a human player, it's obvious that getting better equipment and army sizes is a priority, yet the AI at best was very slow in doing the same. Every game felt like a curbstomp of the AI and this only changed when the bonuses became immersion breaking. It's much more fun to have an AI that can do a decent job with less immersion breaking bonuses.



Only Amplitude can diagnose the root cause of that. I can only speculate. Looking at the xml's there are a huge volume of weights the AI seems to use. Which makes me wonder, how much is actually scripted in the Endless Legend AI? Is it designed to be excessively adaptive? Or is there a "base" script and those weights are used to select variations? The latter seems much more likely to me to be able to emulate good players and put in a good show. I would START with the AI unable to adapt. Get it to kick ass in a scripted fashion emulating as much as you can from good players. Then very slowly, very incrementally, give it some various abilities to adapt ... but without killing the heart of the script i.e. building decent armies / fleets etc.



With Distant Worlds the AI research orders and ship designs are fixed for each race ... there is no adaptability at all (although there is a lot of variation between races). And despite playing the game for many moons I still get beaten from time to time (without feeling like there are immersion breaking bonuses) because the AI is completely focused on it's objectives. It uses my own research orders and ship designs against me. Would I like to be able to Mod some adaptability? For sure. But this very simple scripted approach is light years ahead of Endless Legend and Endless Space AI in terms of performance.



So far. I encourage Elegant Simplicity.




I would say that excessive heuristics can be detrimental to good decisions made by the AI. This affects plans as well with weighting.



On a (fan maintained/developed) Paradox Interactive published game I worked on, we had plans the AI follows based on certain flavor events and it does result in competent AI behavior. There was even a mod that made very competent and challenging behavior, as long as the game turned out a certain way ('historical' scenario). The plans would be .ai files that basically changed heuristic weighting on things like what to build, where to put units, diplomacy, etc. It would work really well for small theaters, but if there were a large amount of factors to consider (ex. multiple theaters) then the AI begins to fall apart. Another thing is that the AI becomes very sensitive to changes (ex. if we improve a rival faction's AI, then a formerly great AI might fall apart. Or if we change unit strengths, or if the player does something out of the box etc). It was also very hard for us to optimize the AI because there were too many factors used in the heuristics. When we saw weird behavior it became a challenge to figure out (did not have access to the internals of the engine, we had 1 developer and he refused to share the code) what exactly was the cause of the weird behavior (ex I even tested setting a heuristic of what I wanted to ridiculously high values but that didn't affect anything). It's hard as a human to find the correct heuristic value to set things.



I agree with the suggestion of updating plans based on the best plans from top players and having AI follow plans. It's important to have plans be updated due to changes in the game, and having this be dynamic would greatly improve the ai.



Due to mods that affect stats of things, As someone who likes to mod from time to time, it would be even better if the game could pull dynamically generate plans based on what mods you have active (popular mods supported via hash or something, while personal dev ones you can have the AI start with baseline and only pull from games played by the local player).



As for what games have great AI.



An old Russian game https://en.wikipedia.org/wiki/General_%28Freeware_game%29 had really strong AI on first glance (but there were AI quirks which could eventually be figured out). I think the simple game design helps it. Probably one of the most challenging AI I've faced



Battle for Wesnoth can have competent AI on a turn to turn basis (not necessarily the entire mission) depending on the scenario. The AI is far superior to Fire Emblem, Elvish Legacy, and most other hex games. Can be very challenging (if playing a 'no-loss' playthrough)



Nobunaga's Ambition:Sphere of Influence has generally bad AI. It's economic AI is competent at managing the economy but can't min-max things. I also ignore the 'suggestions' the economic adviser gives because they are generally very bad. Also I wonder if the AI there is cheating. However it is far far more competent at directing troops to where they should be than Total War in the strategic map, and it (sometimes) pulls off some very good moves in tactical as well (if the unit is led by a good general). The diplomatic AI in the game is horrific and too simple.



CK2/Paradox games while often heralded as great examples of diplomatic AI does have areas where it fails however (ex. a wife I gave a kingdom to who is my lover and has +200 relations supports a faction to depose me....). I think that CK2/Paradox games fall to the victim of having a bit too many modifiers, but what it does correctly is transparency and having (generally) logical AI, aside from the random oversight. Again, too many factors means that it becomes hard to account for.



One thing that I've found in ES after taking over AI planets is that the AI does a poor job of min/maxing what they have with terraforming, moving pops around to optimize growth or ind, etc. Especially notable with the Imperium Aeterna mod. As a player I also found it annoying to have to constantly do the same build logic every turn (go to all my systems, look for anomalies, terraform X planet to Y type, move pops around to maximize Z, etc), especially with all the clicking required (ways to reduce clicking would be awesome, clicking is awful for us coders and our wrists :/ Maybe there are already key bindings I'm too lazy to lookup...hm..). Perhaps if you added a way to have custom automate plan logic, and then analyzed the types of plans players who tend to win use, and then dynamically update the AI based on that?



AI military design was sort've competent in Imperium Aeterna mod. It would react to what it was weak against and change loadout...eventually (often too late). The problem was that it didn't understand at all how to use tactics though in that mod (and which ones were generally OP/too strong when unlocked at a specific tech level) so I was able to steamroll the AI by use of tactics. The AI was also too predictable. (Early game the AI favored kinetics so I would always rush missiles. Mid game the AI favored missiles so I would get the anti-missle tactic that the AI tended to not counter while transitioning to beam/missile mix. Late game my max level hero from the AI sending spam fleets and dying would destroy everything with a 1/1/1 loadout)



Another thing the AI could do is specialization of systems and planets. Eg whenever I took over a system it's not clear the AI had a goal for that system which leveraged it's location and strengths. For me, I wouldn't build buildings that increased border range in backwater systems that are far away from the border and who's systems into it were all controlled by me (saves some Dust). I also had designated shipbuilding systems based on population potential (ex a small system may only be one if it has many Lava/Volcanic worlds and it is in a critically strategic location, and eventually when I pushed the border back, I would terraform it away from being a shipbuilding system). I had designated Dust generation systems, and Science. Each specialized system had a different build priority (aside from a general priority of Food for pop and Happiness) to bring up the critical FIDS asap. If a system had a gas giant, I would lookup what bonuses that gas giant provided and unless required for another purpose, designate the system for that purpose. Of course, with specialization, sometimes I would shift all my systems completely to something (ex all research if about to get a critical tech needed to increase approval or counter an AI invading with specific weapon type, dust if I need to buy out some ships to defend against an unexpected invasion) if needed, so AI should take that into account.



I guess another strategy I have against the AI is being as lean as possible but having a massive industrial potential (I'm a quality type of player in most 4X games...or a turtler in Sins of a Solar Empire: Distant Stars mod). Because you can have interdiction fleets, I would have 1 (high quality with rare resources parts) fleet hold each critical lanes while having targeted expansions early on to grab key resources while minimizing chokepoints. With shipbuilding worlds I can be safe in knowing I can rapidly build out a response fleet if needed tooled to counter the types of ships the AI has while not paying for fleet upkeep (meaning I can have low tax, which increases happiness, which increases FIDS, so I can grow my system faster). The AI doesn't do this afaik due to the huge amount of small fleets I see it with. Similarly the AI tended to not be aggressive until after a certain point, so I purposely avoided building military (always redesign scouts and colonizer at start to have nothing but engines to be cheaper) and I would time my research so that right prior to needing military I would have the 3rd type of hull unlocked so I could build with more weapons and armor than what the AI spent Dust on.



Lastly if you really wanted the AI to play to win, have it destroy all good buildings if it's about to lose a territory. This is a general way to ruin the TW AI as well (raid, destroy all buildings, give the territory back, watch the AI waste all money on rebuilding)
0Send private message
9 years ago
Nov 1, 2015, 9:17:52 PM
wilbefast wrote:
Well re- FOSS: not a discussion we really want to be getting into here and now - my spider-sense is telling me that it's likely to wax polemical real fast smiley: rollblue



For my part I've been working on AI for EL lately, with MeeDoc and Metalynx taking the baton where ES2 AI design is concerned. I'll let them fill you in on exactly where things are as I don't think I'd do it justice smiley: smile


It's great to hear that EL will benefit from the AI development!
0Send private message
9 years ago
Nov 1, 2015, 10:14:07 PM
This, perhaps, isn't relevant due to how far development is but reading this thread I kept seeing people say how they wanted a consistent AI that was still unpredictable. Perhaps there could be a sliding scale of personality types from Passive to Aggressive. Each point along the scale would mean a little more aggressive and make slightly different moves than the personalities next to it on the scale. Then have each faction weighted based on their personality as to what types of personality they get. Then have that choice weighted by game difficulty.



I'd also like to see the AI interact with other AI more without prompting in some visible way. I rarely see AI form an alliance or invite players to an alliance. I rarely see the AI go to war with each other without player involvement.



I like the fact you, Amplitude Studios, try to program the AI as if it weren't in the game. I've noticed this across ES & EL and it really helps immersion. Perhaps the middle ground though is making the AI work toward general dominance over other factions by playing to their strengths. That way they're working toward an endgame without working a goal.



Not to be specie-ist but I'd like the AI to notice that certain species are potentially existential threats and perhaps make choices relevant to that. For example: In an Ovoid Galaxy I border the Cravers and the Sowers. The Sowers should make a choice betwixt helping me eliminate the Cravers (because, if given the opportunity, will become to powerful for any one faction to stop) and letting the Cravers mess me up (to conquer me while I'm weak). The choice should then be weighted by the faction alignment.



I strongly oppose factions becoming more likely to align based solely on whether they're the same specie or not. It should be totally based on which faction has been the most beneficial to you. To do otherwise is to promote specie-ism.



I'm also generally opposed to 2 factions in lesser standing allying against a stronger empire unless the strong empire is trying to conquer them both. Its not how I play and its not how the real world works. Example: I'm the Automatons (Good aligned), I'll often invite weaker empires to ally with me to not only protect them (as the Automatons would do) but to allow me to benefit economically from them while I go after the empire directly behind me in strength. In the real world small nations like Estonia and Latvia actively sought out N.A.T.O. membership (strongly financed and supplied by the powerful United States and the moderately powerful allies of the United States like France) to counter what they see as aggression from the Russian Federation. Strong factions should also work to secure alliances to maintain parity while building a strong infrastucture. This serves the dual purpose of allowing for pursuance of other victory types while being able to pursue a military conflict should one arise. In ES2 it'd be neat to see populations become more passive if you're allied with peaceful factions while making theirs slightly more warlike and vice-versa.



P.S. I'd like the ability to buy ships from other empires and for the conditions to meet a diplomatic or economic victory to be much more transparent. As it stands I have no idea how to optimize my diplomacy points and need to go into the code to get an inkling of the values I have to reach to succeed.
0Send private message
9 years ago
Nov 1, 2015, 11:21:02 PM
limith wrote:


Nobunaga's Ambition:Sphere of Influence has generally bad AI. It's economic AI is competent at managing the economy but can't min-max things. I also ignore the 'suggestions' the economic adviser gives because they are generally very bad. Also I wonder if the AI there is cheating. However it is far far more competent at directing troops to where they should be than Total War in the strategic map, and it (sometimes) pulls off some very good moves in tactical as well (if the unit is led by a good general). The diplomatic AI in the game is horrific and too simple.



CK2/Paradox games while often heralded as great examples of diplomatic AI does have areas where it fails however (ex. a wife I gave a kingdom to who is my lover and has +200 relations supports a faction to depose me....). I think that CK2/Paradox games fall to the victim of having a bit too many modifiers, but what it does correctly is transparency and having (generally) logical AI, aside from the random oversight. Again, too many factors means that it becomes hard to account for.





Agreed on Nobunaga's Ambition. I was surprised by the flanking maneuvers and pincer attacks the AI did on occasions, as well as how quickly they are to react and mobilize against any sort of movement close to their borders. I have also noticed them fortifying key strategic areas and actually knowing to put its armies there when the time comes. But it is very true that the AI does not know how to manage cities and is too passive in diplomacy even on max difficulty. NA: SOI is in that sense very much a wargame. It has cool diplomatic features, like establishing and coordinating coalitions, but diplomacy is at the end of the day a military tool and nothing more. In its defense though, the period is called Sengoku Jidai.



But I find that flaw in Paradox games a bit more jarring (barring Hearts of Iron, of course). In my admittedly limited experience, I've found diplomacy in Paradox games to also be a military tool more than anything else, and that at heart the games are wargames.



There are few games that allow one to play as a "soft power", relying on diplomacy as its primary foreign policy tool as opposed to supplementing its military. I hope that ES2 is that sort of game, with an AI that can be more reactive, responsive, and pro-active in diplomacy, as well as have different personalities that view diplomacy differently.
0Send private message
9 years ago
Nov 9, 2015, 9:54:41 AM
Selein wrote:
This, perhaps, isn't relevant due to how far development is but reading this thread I kept seeing people say how they wanted a consistent AI that was still unpredictable. Perhaps there could be a sliding scale of personality types from Passive to Aggressive. Each point along the scale would mean a little more aggressive and make slightly different moves than the personalities next to it on the scale. Then have each faction weighted based on their personality as to what types of personality they get. Then have that choice weighted by game difficulty.



I'd also like to see the AI interact with other AI more without prompting in some visible way. I rarely see AI form an alliance or invite players to an alliance. I rarely see the AI go to war with each other without player involvement.



I like the fact you, Amplitude Studios, try to program the AI as if it weren't in the game. I've noticed this across ES & EL and it really helps immersion. Perhaps the middle ground though is making the AI work toward general dominance over other factions by playing to their strengths. That way they're working toward an endgame without working a goal.


Lately we've done a lot of (as yet unreleased) work on EL making the AI play differently depending on the faction. This work will of course continue for ES2 smiley: smile



Selein wrote:


Not to be specie-ist but I'd like the AI to notice that certain species are potentially existential threats and perhaps make choices relevant to that. For example: In an Ovoid Galaxy I border the Cravers and the Sowers. The Sowers should make a choice betwixt helping me eliminate the Cravers (because, if given the opportunity, will become to powerful for any one faction to stop) and letting the Cravers mess me up (to conquer me while I'm weak). The choice should then be weighted by the faction alignment.



I strongly oppose factions becoming more likely to align based solely on whether they're the same specie or not. It should be totally based on which faction has been the most beneficial to you. To do otherwise is to promote specie-ism.



Racial profiling is indeed a horrible, horrible thing, but in a sense it's the responsibility of media-creators address this kind of real-world issue... one might even argue that it would be racist to pretend that racism doesn't exist by not to including racism in ones comic/film/game/book. It's really a matter of how you address this subject-matter: are certain enlightened cultures above profiling? Are others bitter about past affronts? Who is naive and who is cynical?



What's great about the science fiction, space operatic genre is that it provides a safe space to discuss these follies of mankind: Foundation, Star Trek, Battlestar, Dune, ... they all spend a lot of time philosophising smiley: smile



limith wrote:
I would say that excessive heuristics can be detrimental to good decisions made by the AI.


limith wrote:
I agree with the suggestion of updating plans based on the best plans from top players and having AI follow plans.


Ah... well it's very easy to say "just have the AI do what and expert would do" - indeed this is essentially the principle behind an early type of AI called an "expert system". These systems largely failed though due to, among other things, a lack of flexibility. It's also very time-consuming to distil expert knowledge into rules that an automated system can follow. Heuristic systems have tended to replace rule-based ones, and increasingly machine-learning techniques are being used to find the heuristic weightings through example (supervised) or through trail and error (unsupervised).



limith wrote:


Battle for Wesnoth can have competent AI on a turn to turn basis (not necessarily the entire mission) depending on the scenario. The AI is far superior to Fire Emblem, Elvish Legacy, and most other hex games. Can be very challenging (if playing a 'no-loss' playthrough)



I had a look into their AI: funnily enough it's not so different from ours - I'd have thought they'd use ExpectiMax but in fact even Wesnoth is too complex for that class of algorithm. Writing a Wesnoth AI is hard, but even if it weren't the complexity of Endless Space is at a different order of magnitude, and the difficulty of writing good AIs doesn't scale linearly I'm afraid - more like exponentially.



limith wrote:


CK2/Paradox games while often heralded as great examples of diplomatic AI does have areas where it fails however (ex. a wife I gave a kingdom to who is my lover and has +200 relations supports a faction to depose me....). I think that CK2/Paradox games fall to the victim of having a bit too many modifiers, but what it does correctly is transparency and having (generally) logical AI, aside from the random oversight. Again, too many factors means that it becomes hard to account for.



Your wife may have been plotting your downfall for other, hidden, reasons: as you suggest yourself transparency and perceived logicalness are often essentially the same thing - we were staggered by how much the pereceived intelligence of Endless Legend's AI went up when we added diplomatic feedback messages (the AI saying "I see what you did there and I'm (not) happy about it")



Falkner wrote:
It's great to hear that EL will benefit from the AI development!


Thanks smiley: biggrin
0Send private message
8 years ago
Jul 10, 2016, 11:42:12 AM

By the way, will ES2 be 64-bit with multi-core support? It would really enlarge the possibilities for AI improvement and, if nothing else, would help with turn time. 


GalCiv3 is already 64-bit with multi-core support, the sequel to AoW3 will have it and if I'm not mistaken Civ6 will also be doing it. Would be really silly to release a 4X game in 2017 still constrained by 32-bit architecture when even smartphones will have moved beyond it.

0Send private message
8 years ago
Jul 18, 2016, 12:31:43 PM
Falkner wrote:

By the way, will ES2 be 64-bit with multi-core support? It would really enlarge the possibilities for AI improvement and, if nothing else, would help with turn time. 


GalCiv3 is already 64-bit with multi-core support, the sequel to AoW3 will have it and if I'm not mistaken Civ6 will also be doing it. Would be really silly to release a 4X game in 2017 still constrained by 32-bit architecture when even smartphones will have moved beyond it.

ES2 will be 64 bits-only. Whether and how it handles multi-core will mostly depend on Unity and whether it handles AI, graphics etc on separate threads. 

0Send private message
?

Click here to login

Reply
Comment