Yep the issue with high bonus is that a while set of parameters decisive for the players aren't to the AI. You can't cripple your opponents you have to strong-arm them and make use of their tactical flaws.



The goal oriented AI seems interesting.

Having worked as a gameplay software engineer for AAA titles, I'll like to volunteer some thoughts about AI in games. Hopefully, the Amplitude gameplay team find this post useful for developing this game that I absolutely love.



Game AIs are all about maintaining illusions of consistent opponents. I found AI agents that following a set of Scripted Plans better at it than an AI that constantly re-evaluate and re-choose plans based on weighted stochastic (something I suspect being at the heart of the EL AI)... A model I worked with before goes like this: the agent starts by choosing a Goal, which maps to a Plan. The Plan in turn maps a set of Actions that span multiple turns. The evaluation period and triggering conditions are all tunable parameters that the gameplay producer can affect. For example, I.e. Aggressive AI opponents only choose aggressive Goals and Plans. It does so a lot less frequently than it evaluates and re-choose Actions. AI with higher competency can be implemented by having it re-evaluate the its Actions on every turn, while AI at lower competency can be implemented by having it evaluate its Actions less frequently.



If a plan is implemented in a scripting language, that would make the AI flexible to changes to meta after it's shipped.



Apologies for being too presumptuous...

From a "clinical" point of view, and with only Endless Legends as a parameter, I think the decision to make the AI aware of the victory conditions would be a significant improvement - given that everything else is adjusted to the fact the AI will be focusing from the start of the game.



Unfortunately, AI opponents on 4x games will be for a long time just a very complicated puzzle the player has to solve. We cannot currently teach game-AI how to play a game.



What I mean is that, probably, an AI which focus on determinate Victory Conditions will have less flexbility to react to things outside that Victory Condition. It will be much harder to beat an economic AI on an economic game, but probably easier to beat it at Diplomacy or even Military.



The way I see it, people who would preffer VC focused AI (me included) hope that the fact that factions are geared towards a certain VC would make them more able to react to unkown threats once they´re playing according to their strenghts. They wouldn´t be more capable of resisting because they were made that way, but as a consequence of their natural gameplay being strong enough inside the game´s mechanics.



For that to work, the mechanics have to be very very well balanced among themselves, and that´s something I honestly think has not yet been achieved in Endless Legend.

Hi BPrado,



AI that factor in Victory Conditions can be implemented in a Goal/Plan/Action model. That said, I think such an AI may run into the problem of being consistent or being in character. For example, imagine a Craver AI in a diplomacy VC only game... Imagine a Roving Clan AI in a game with only supremacy VC on... I think in such games, the decks have been stacked against the AI. Rather than having the gameplay team implements an AI that can win (which as you pointed out is difficult), the AI should strive to immerse the player into the game. i.e. an AI that plays in character of its race without considering VC, and chooses plans that suit its race preferences.

Yes, I suppose that´s another valid option, instead of "VC" flavoured AI, simply "Game-mechanic" flavoured AI - which favours the mechanics related to its strenghts, and end up being a challenge because its potentials are fully explored. That´s probably more or less how to describe Civ 5´s AI, which imo is the best out there in terms of 4x.



edit: I´m really not a fan of editing victory conditions, or most game settings for that matter. I think that taking that into consideration screws up any hope for balance, and if I was lead design of any game ever, this would never be a thing. You ppl should be glad I´m not that successful. I am aware though that professionals have to take everyone into account.

Good to hear you got some people dedicated for AI!

poolslice wrote:
Hi BPrado,



AI that factor in Victory Conditions can be implemented in a Goal/Plan/Action model. That said, I think such an AI may run into the problem of being consistent or being in character. For example, imagine a Craver AI in a diplomacy VC only game... Imagine a Roving Clan AI in a game with only supremacy VC on... I think in such games, the decks have been stacked against the AI. Rather than having the gameplay team implements an AI that can win (which as you pointed out is difficult), the AI should strive to immerse the player into the game. i.e. an AI that plays in character of its race without considering VC, and chooses plans that suit its race preferences.

We're working on faction personalities (changes to the stochastic weightings you mentioned) for EL at the moment, so there'll at very least be this kind of being in character. As for the Goal/Plan/Action - can you give an example of what sort of thing this would be for a 4X game? "Plan" is a loaded word - it means very different things to different people, especially if said people are in academia. To my knowledge technologies like STRIPS are great if your branching factor is low (say in an FPS), but in a 4X game there's something of a combinatorial explosion of possibilities. Heuristics may seem a little too "fuzzy" sometimes but the one advantage they have is that they can handle these thousands of possibilities without needing a country's worth of FLOPS to provide reasonable response-times.

Metalynx wrote:
I was trying to understand why you (and possibly others) enjoyed those AI, by looking at technical implementation and numbers.

Let's get back to GalCiv2. Your research showed it only gets resources bonuses. That may be true - but it feels intelligent for several reasons:



- countering player's ship design. That's what real humans are supposed to do. So when the AI does it even only because it cheatingly knows every design in existence it still feels like an intelligent move.

- it recognizes common strats and the triggers a diplomatic response at the very least - for example, when you start massing troops at its borders the AI call you out on your action in a tone and voice appropriate to its faction (that really helps with establishing faction personality). Nowdays even Civ5 does it. AI also responds if you try to culture-flip its colonies or are clearly rushing for wonder victory. And the response is not merely sending a diplomatic note - sometimes it goes to war with that as casus beli.

- small and weak AI factions gang up on strong ones or do not aggress them in the first place

- it goes after the same victory types, not a general chaotic expansion that may or may not lead to a domination victory



My advice, for what it's worth, is to make it feel intelligent by any tricks you can use. I do not have a background in software so to me all AI in games seems to be a plethora of "If X, then Y" triggers, with some additional conditionals for complexity. Put in those triggers that best simulate intelligent human action.



Let's return to the example of countering ship design:



- AI1 is at war with AI2 and already has a particular ship design

- human player decides to attack AI1 and has superior design

- so let that be the trigger that makes the AI1 change its ship design to counter the human player

- immediately retrofitting its entire fleet at the start of the war thanks to its econ bonus would feel too much like cheating

- so make the retrofit happen after an additional condition is fulfilled: like after n or n+1d6 turns/battles (for variety's sake a random element would be better, people would quickly notice if every AI in the game always retrofitted its fleets to counter players after 5 turns/battles)

- for added complexity and personality you can factor in faction preference for certain weapon types or ships classes or whatever



From the perspective of the player (who does not know exactly how the AI is programmed) this would seem like an intelligent reaction by the AI due to battle experience, it would have seemed to have learned from its losses. Not simply obeying a blind predetermined trigger or having difficulty arising from endless fleets of zero upkeep ships equipped with weapons entire eras in advance of the player due to research bonus.

Falkner wrote:


My advice, for what it's worth, is to make it feel intelligent by any tricks you can use. I do not have a background in software so to me all AI in games seems to be a plethora of "If X, then Y" triggers, with some additional conditionals for complexity. Put in those triggers that best simulate intelligent human action.

In practice this is a little to rigid to really work for an 8-empire free-for-all. A better metaphor would be that of a highly complicated and interconnected system of weights and pulleys all tugging eachother in different ways: sometimes well gets pulled far enough to pull a lever and then, as they say, all hell breaks house



Which is not to say your suggestions are not very interesting

Falkner wrote:
Let's get back to GalCiv2. Your research showed it only gets resources bonuses. That may be true - but it feels intelligent for several reasons:

- it recognizes common strats and the triggers a diplomatic response at the very least - for example, when you start massing troops at its borders the AI call you out on your action in a tone and voice appropriate to its faction (that really helps with establishing faction personality). Nowdays even Civ5 does it. AI also responds if you try to culture-flip its colonies or are clearly rushing for wonder victory. And the response is not merely sending a diplomatic note - sometimes it goes to war with that as casus beli.

I like this suggestion in particularly - it's very frustrating when our AI does something that's actually rather clever but it is perceived to be stupid because there's no thought-bubble popping up to tell players why this particular action was chosen. Often I'll open the AI's brain to see what it's doing and, upon seeing it's internal state, decide that in fact it is behaving as it should. On the surface though it might have seemed strange. I'm sure Jeff wouldn't object to writing more flavour-text. Well he might, but he's outnumbered by French people here 10 to 1 As our Metalynx and I come to think of it

Falkner wrote:


- countering player's ship design. That's what real humans are supposed to do. So when the AI does it even only because it cheatingly knows every design in existence it still feels like an intelligent move.

Falkner wrote:


- AI1 is at war with AI2 and already has a particular ship design

- human player decides to attack AI1 and has superior design

- so let that be the trigger that makes the AI1 change its ship design to counter the human player

- immediately retrofitting its entire fleet at the start of the war thanks to its econ bonus would feel too much like cheating

- so make the retrofit happen after an additional condition is fulfilled: like after n or n+1d6 turns/battles (for variety's sake a random element would be better, people would quickly notice if every AI in the game always retrofitted its fleets to counter players after 5 turns/battles)

- for added complexity and personality you can factor in faction preference for certain weapon types or ships classes or whatever

- small and weak AI factions gang up on strong ones or do not aggress them in the first place

- it goes after the same victory types, not a general chaotic expansion that may or may not lead to a domination victory

Good idea, although there's a lot to implementing this. First we need to recognise that a specific military "whacking stick" has our name on it. We need, in order to do so, to have a look at all the whacking sticks that all the empires may have at their disposal. Probably the guy we're at war with or the guy on our border is planning on whacking us. Okay. Then we need to figure out what will counter this stick, which is a surprisingly complicated problem to solve automatically.

That said you're right: giving the AI some sort of "crystal ball" it can use to see what's coming at it so that it can prepare accordingly would vastly improve it's ability to defend itself.

Metalynx wrote:


Gal Civ 2 (which was mentioned - don't remember by who)

http://galciv.wikia.com/wiki/Difficulty_level - Essentially same as AoW.

Low difficulties - Intentionally making bad decisions for AI.

Above normal: flat economic bonuses 105% all the way to 400% on extreme difficulty modes. Also the AI cheat by selecting ship designs that directly counter the players ^^

I mentioned Gal Civ 2, and I wanted to interject a bit to clarify a few things.



It is true that AI in extremely high difficulties get bonuses, but that only happens after a certain difficulty level (either "tough" or "challenging", forgot which). In the last difficulty level before the bonuses, the AI relies solely on its "intelligence", which can be boosted by allowing it to use CPU to its max potential. When it does so, the AI becomes surprisingly challenging (this is esp the case with the Twilight of Arnor expansion).



Not only does each faction behave differently according to its strengths and weaknesses, but they also are efficient at pursuing different victory conditions. I've had several scenarios where the Torians stayed out of everyone's business, looking unthreatening and isolated, while b-lining for the science victory.



As was mentioned above, the AI feels alive and intelligent, by reacting to the player's actions, such as expanding military power, or amassing fleets near the borders or expanding cultural influence. Often you have AIs sending you messages warning you about a faction becoming too powerful. Weaker factions gang up against stronger ones, often setting differences aside and stopping wars between them. Bribery is used commonly (esp with the Drath AI) to instigate wars. Stronger factions bully weaker ones for credit and tech. Alliances that make sense persevere, sometimes throughout an entire game...etc.



As for countering player designs, that is not really cheating. The AI chooses to invest in creating new designs or retrofitting existing fleets (which is very expensive). The human player does the same. And what is interesting, is that you can do many creative things with this system. I have once for instance offered a faction missile tech very early on, encouraging it to invest in that tech tree, while giving another faction a particle tech encouraging it to invest in that tree. Then I manipulated both to fight each other, meaning that both started to invest in the appropriate defenses (against particles and missiles), while continuing to invest in the weapons that I provided for them. Meanwhile, I b-lined to maximize my beam technology, and thus when I declared war, I was able to swiftly defeat both as they had no counters to my designs, and did not have the time to shift their tech. This is a classic example of tech manipulation you can do in the game, which makes diplomacy feel a lot more dynamic and open for creativity.



I've seen the AI think in a sophisticated manner. My favorite example was when during the United Planets' summit, a new habitable planet was discovered and who it was going to be given to was put up for a vote. There was not a single faction that had absolute majority, and all factions including myself voted for itself to acquire the planet, with one exception. The Altarians, an ally of mine, voted for me to acquire it, which gave enough votes for me to do so. Turns out the planet was in the middle of their territory and within their sphere of influence. Then they started investing in cultural buildings around it to peacefully acquire it. This shows several things that the AI considered:

- It understood that if it voted for itself, it would not have acquired the planet. But that its vote could give a majority to whoever it chooses.

- It preferred to give it to an ally than a hostile or neutral empire.

- It understood that it can peacefully acquire the planet without necessarily jeopardizing the alliance it had with me.



That is surprisingly complex thinking for an AI, which makes it feel more alive and intelligent, even if it's easy to exploit by someone who knows the game well. I personally have never experienced an AI that is on its level, and I do think it is an example worth exploring.

KnightofPhoenix wrote:
That is surprisingly complex thinking for an AI, which makes it feel more alive and intelligent, even if it's easy to exploit by someone who knows the game well. I personally have never experienced an AI that is on its level, and I do think it is an example worth exploring.

Something I'm going to mention here that you may not have taken into account.



GalCiv factions are all very similar. They do not have very different techs. There is not a lot of asymmetry in the GalCiv games. The combat systems are extremely simple. That has likely helped Stardock enormously in their AI work. GalCiv's systems are very flat (for lack of a better word). Those games lend themselves to better performing AI. I'm not trying to take anything away from Stardock here either. But for the most part, they do not make hugely asymmetrical factions like Amplitude tend to do. And when they do, the results are less than impressive sometimes. See the Dead faction in Fallen Enchantress: Legendary Heroes.



IMO, that is one of the reasons why faction personality and goals are so important in EL. The AI first has to 'understand' its strengths and weaknesses and how to apply and compensate for them. And also it has to know what goals it should strive for versus just expanding for the sake of it. Fancy diplomatic maneuvering is useless to the Cravers or Necrophages. Diplomatic maneuvering is/should be paramount to the Drakken. A strategy based on war should not be the province of the Roving Clans.



One thing I would like to see implemented is AI factions bribing or paying off other factions to eliminate the ones they see as a threat or as enemies.



I had a really cool game in Fallen Enchantress once where I lost really early because I had a slow start and one faction hired another to take care of me. They just kept hitting my cities and I eventually couldn't field a decent defense.

I would really not like having a bland game design like GC3 with no imagination what so ever just to have better A.I.

Slashman wrote:
Something I'm going to mention here that you may not have taken into account.

I actually discussed similar points here: /#/endless-space-2/forum/65-general/thread/18713-i-have-a-idea-about-science-tree



You are absolutely right to say that factions in Gal Civ 2 are fundamentally more similar than in ES or EL (but there is more freedom and possibilities in tech and research), which would make the AI easier to improve I imagine. I understand that it is more challenging to teach the AI to learn the strengths and weaknesses of different factions.



When I praise the AI in Gal Civ 2, it is by no means a jab against ES and EL, which do a number of things much better. I am simply presenting an example of good AI that one can be inspired by, all the while hopefully improving on the things that make Amplitude games awesome in the 4x genre.

Metalynx wrote:
Now, I'm no AI expert, but you say this is a general flaw in too many strategy games. So as an exercise, could you mention a few examples of games - preferably with the scope of a 4X - that uses this kind of methodology?

From my understanding, the methodology we used in Endless Legend is similar to that of most other prominent games of the same genre, such as Total War and Civ. So also as a general request to you all: What games of the same scope/genre has an AI you consider good? What parts of that AI is giving you the impression it is working well?

Total war games are shunned as having a usually terrible strategic AI though, so please don't take it as an example. For instance in Shogun 2 Total war, at the moment you had a certain number of territory in your possession all the AIs, even your ally would declare war on you. This is a perfectly bad example of mixing AI logic with game mechanics.



Paradox Grand Strategy games, though having a quite different gameplay, have a really good AI. Please look into how it works in Europa Universalis IV.



Handicap must also be a different setting than AI difficulty. AI difficulty should mostly determine how aggressive the AI plays, while Handicap is your usual set of bonus/maluses.



Finally, in some total war games (again terrible strategic AI) some modders managed to mitigate the terribleness of the AI by teaching it how to correctly build armies. They weren't able to change the AI decision making about how to use theses army because those parameters are usually hardcoded, but once the AI had better army templates it was able to operate with more success.

In ES1 the common strategy for AI was to stack tons and tons of 1 ship fleets, they weren't able to do anything against the player, but abused the fact your fleet were only able to do one fight per turn (a rule that didn't make any sense in that case), which basically turned war for a system into a massive grind.

If the AI simply knew how to build better fleets it would have dramatically increase its ability to challenge the player.

KnightofPhoenix wrote:
I actually discussed similar points here: /#/endless-space-2/forum/65-general/thread/18713-i-have-a-idea-about-science-tree



You are absolutely right to say that factions in Gal Civ 2 are fundamentally more similar than in ES or EL (but there is more freedom and possibilities in tech and research), which would make the AI easier to improve I imagine. I understand that it is more challenging to teach the AI to learn the strengths and weaknesses of different factions.



When I praise the AI in Gal Civ 2, it is by no means a jab against ES and EL, which do a number of things much better. I am simply presenting an example of good AI that one can be inspired by, all the while hopefully improving on the things that make Amplitude games awesome in the 4x genre.

I understand what you're saying and I do agree. The design philosophies are different for each developer. Amplitude is very much centered on lore, immersion, interface and unique factions. Those are implemented and designed first I imagine before any thought is given to how they would perform as an AI.



GalCiv 2 has a more expansive tech tree, but is also more generic overall. There's a give and take to a certain extent. I suppose the question is: Do we want to lose some of the uniqueness of the factions that we have come to know and love in the Endless Universe so we can make a better AI for the game overall? That's a tough question because AFAIK there are no races in GalCiv that can teleport whole fleets to any owned system/planet. There are no races that can take the entirety of a developed world's infrastructure and move it somewhere else. There are no races that can't use a specific type of resource and further have that same resource hinder them just be being present in their system. There are no races that can only have one system/city.



The Age of Wonders 3 route was to have unique races but without abilities/traits that are so far off the beaten path that they hinder the AI greatly. Then you get to pick a class which distinguishes your faction further with unique spells and bonuses and units. Then you pick a further 3 specializations which give you more unique units, spells and bonuses. This works well for a number of reasons:



1) AoW 3 only has 3 possible victory conditions and none of them are economic or diplomacy based.



2) Even though there are some seriously powerful units and spells in the game. The balance is eerily good no matter what class you play or what specializations you pick. Even so-called 'gimped' combinations are perfectly viable because there are so many different ways to get what you need in the game. If your class lacks a hard-hitting tier 4 unit, you can still make friends with giants and dragons and other fantastical NPC factions.



3) None of the victory conditions favor a specific faction.



The downside of AoW 3 is that it is much more military oriented than any of the other 4Xs listed except maybe for Sword of the Stars. Its diplomacy works but is very bare bones. This does make its AI factions behave more 'sensibly' in diplomatic relations though.

Slashman wrote:
I suppose the question is: Do we want to lose some of the uniqueness of the factions that we have come to know and love in the Endless Universe so we can make a better AI for the game overall?

My answer to that would be a definite no. But I am not sure if such a dilemna applies, at least not as strictly as to be completely restrictive. But of course I know nothing about AI design in technical terms.



I think it is still possible to add new diplomatic options, such as bribery, and teach the AI how to do it. Another thing the AI can be taught, I imagine, is a better grasp of alliances and the balance of power, aka weaker factions gang up on stronger ones (which i have not seen happening unless i convince them to). The AI tends to act erratically, and it is an unreliable ally.



These are 'basic' things which the AI can be taught, without worrying too about faction diversity, I imagine.



The tricky part would be to make the AI play very differently based on factions. A craver or Hissho AI should not turtle for instance, an amoeba should be an especially reliable ally and diplomatic power...etc.



I think a certain balance between AI performance and faction diversity can be reached, but I am speaking as a layman of course.

wilbefast wrote:
We're working on faction personalities (changes to the stochastic weightings you mentioned) for EL at the moment, so there'll at very least be this kind of being in character. As for the Goal/Plan/Action - can you give an example of what sort of thing this would be for a 4X game? "Plan" is a loaded word - it means very different things to different people, especially if said people are in academia. To my knowledge technologies like STRIPS are great if your branching factor is low (say in an FPS), but in a 4X game there's something of a combinatorial explosion of possibilities. Heuristics may seem a little too "fuzzy" sometimes but the one advantage they have is that they can handle these thousands of possibilities without needing a country's worth of FLOPS to provide reasonable response-times.

Hi wilbefast,



Before I reply, I'll first try to understand what your comments are... The reason why at game design phase of EL, a heuristic function based approach was chosen over the traditional planning approach to avoid large computation complexity, right?



Okay. Now that I think I understand you, here are some of my replies...



A traditional, planning based approach AI does have the problem of large computational complexity. There are many technical solutions to mitigate that... Here are a few suggestions...

• Limited data from the game world: have the decision functions take a narrow perspective from the actual game world. That has the effect of reducing branch factors.
• Decision node unification: typically, in a dynamically constructed decision tree, two paths often lead to the same intermediate node/decision. During the AI planning process, you can try to recognize/hash the decision node and input so that you don't evaluation the same decision node multiple times...
• Decision Tree Evaluation Sequence Optimization: this is just a fancy way of saying 'order your decision from large branch out factor to low branch out factor'. This way, your choices and things to evaluate is narrowing...
• Parallelization: Apologies I haven't done this... I should open up my PC process manager to check how many of my i7 core/hyperthread are being used during a game of EL.
• Multi-tick computation: Unlike an FPS that needs to meet a frame rate, TBS games can do the computation across multiple ticks, as the human brain is thinking about what to do, the AI agents can think about what to do (as I am sure it does, since I see units moved after a while if I don't move my units right away).
  
  
  
  

Mind you, I am not completely against a weight based approach to AI. I think of it as this... there needs to be a balance between playing against a deterministic decision tree (which would be boring as the AI would play out the same way) vs a pure loaded dice (which is what a pure weight based approach boils down to...). I think the balance lies the frequency of evaluation... Just some quick run-down then about what I am thinking about, without having to write a detail design doc...



Tick 1: AI rolls a race loaded dice to decide what Goal to have. ex. the EL Cultists decides to Goal( limited game world ) = 'Improve Influence', because that how I would play the Cultistis...

Tick 2: A Goal can be achieved by a set of hard-scripted Plans. ex. the EL Cultistics does Plan( 'Improve Influence', limited game world, Cultist ) = 'military'.

Tick 3: The 'military' Plan is just a series of Actions that consists of finding the first available unit group, find a village with influence bonus (kanji, I think...), pacify it, and then assimilate it.



Once decided, the Plan and its Actions just get carried out across multiple turns. A series of interrupt events can be registered to fire and interrupt the Plan. Higher difficulty AI may want to interrupt and re-evaluate its goals more frequent, to reflect new changes in the world since the Goal and Plan were made.



During AI development of my game, I used to have debug text on the screen to show me, at each tick, what the AIs were thinking. I always thought that the presentation engine could take advantage of all those debug text in diplomatic screens, in game news cast, audio, espionage reports, (won't it be great that as the Shadow faction of the up coming expansion, you can actually see what the AI is planning to do?)...



Anyhow, I feel that my posts are going too deep. Sorry about that..

KnightofPhoenix wrote:
Another thing the AI can be taught, I imagine, is a better grasp of [...]the balance of power

Yep, you touched a really important subject here, for me at least. Many (if not every single) strategy games base their philosophy on political realism - with atomized states that act strictly according to a "national interest" in a permanent struggle for power - yet they often forget the most important consequence of it, which is the balance of power. Each faction´s score makes no sense in a vacuum, they have to be constantly acting and reacting according to a balance, and I don´t see that happening in Endless Legend.



Again, a close comparison would be Civ5, which does that well enough.

darkath wrote:
This is a perfectly bad example of mixing AI logic with game mechanics.

This is exactly what I mean when I say that a "Game Master" wouldn't be "kosher"

It's worth noting though that you can never completely separate the two: most 4X players play against the AI most of the time, so when we're designing systems we have to keep in mind whether the AI will be able to use them correctly. It's also why the trade-off between the exoticism of the factions and the quality of the AI is one worth giving some serious thought.

darkath wrote:
If the AI simply knew how to build better fleets it would have dramatically increase its ability to challenge the player.

Interesting theory - does this seem legitimate to everyone?

poolslice wrote:

• Limited data from the game world: have the decision functions take a narrow perspective from the actual game world. That has the effect of reducing branch factors.
• Decision node unification: typically, in a dynamically constructed decision tree, two paths often lead to the same intermediate node/decision. During the AI planning process, you can try to recognize/hash the decision node and input so that you don't evaluation the same decision node multiple times...
• Decision Tree Evaluation Sequence Optimization: this is just a fancy way of saying 'order your decision from large branch out factor to low branch out factor'. This way, your choices and things to evaluate is narrowing...
• Parallelization: Apologies I haven't done this... I should open up my PC process manager to check how many of my i7 core/hyperthread are being used during a game of EL.
• Multi-tick computation: Unlike an FPS that needs to meet a frame rate, TBS games can do the computation across multiple ticks, as the human brain is thinking about what to do, the AI agents can think about what to do (as I am sure it does, since I see units moved after a while if I don't move my units right away).
  
  

Nice Thanks for the in-depth answer!



Intuition tells me that these optimizations may not be sufficient, but I should probably look into these things and do the math before making any judgement calls. Planning could well prove to be the right solution at a very high level (working towards victory) but the danger there are the "steps" in the plan being too far removed from the actual actions perform in the game. IE are as many way to have one more cities as there are to skin a cat. A much "safer" place that planning could (and may well) be used is quest resolution. This is something we're looking into at the moment.



Got to get back to EL now for a little while I'm afraid - I'll be back

A game should only include features that the AI is able to use well, too. The things that I, as player, am able to use and the AI is not, are "legal" cheating. Naturally, the player can choose not to use them yet there are things one is not always able to detect.



Slashman already proposed Sword of the Stars (1) as an example of good AI. With rare exceptions, it is one of the best AIs in space strategy games. It needs to use very different ways of travel and two very different economic systems and it does that pretty good.



However, the game does let the player use systems the AI is unable to. Things like per-pixel targeting and so forth. Not that I'd like the AI do it, as it could surely be better at it than the player, but I tend to avoid a system like that, too.



So, please let the player only use the features the AI is able to use, too.