Great job. The game has improved. Looking forward to the next upgrade.

Freebird in the birdcage for the moment.

I'd just like to chime in that while the general AI is important, the risk when you start making general spell and building priorities is that the unique nature of each of the factions risks being lost. The suspension of disbelief and story-telling elements are just as important as a "winning" AI. Frankly, I think a skilled human player will be able to beat the AI, regardless of how much tweaking is done. But a good individualized AI that acts uniquely as per the unique society of its faction lets me play with the game to tell a story. Maybe I'll help Pariden bring justice to the world, or help Umber return the Titans to glory. I expect Umber to be following certain research paths and building Titan temples. I don't expect Pariden, say, to ignore magic development and not harvest its shards right away.

Ultimately, if I like Pariden, I'll help it survive. But if it doesn't act like Pariden, I'm just not motivated -- the game loses its color, its story.

Just my two cents as more of a "story" than "strategy" gamer,

-mr. b

Not that I disagree with your point, but you're assuming that factions have a unique nature to begin with. They don't. Hopefully this will be addressed in the future.

I think you should just go for the look at future turns like u did wiht gal civ 2, seirously what gamer doesnt have a quad core cpu? comon now there like 60beans online nowaday. The cost of one pc game. (or 2 if u want a lil better one).

If it really is that slow of a community of cpus, maybe make a option of normal , normal (With half assed computer) ,expert , expert (with half assed cpu) .

or u could just make the people taht dont know how to upgrade a computer wait an extra second or two for the turns to end i dotn mind

or maybe do a Overall user system rating for all of the elemental users and see if the majority has quad cores or not. maybe slap something on elemental, or even on teh forums taht way only the players involvedd with the game more so get to be the input ? (last part may be pushin it lol)

q9550 2.83 intel core2quad

465 gtx nvidia

8gb ddr2800

Win7x64 Enterprise.

Good point above me to , there really is no uniqnes to the races other then the dif between kindom and empire, all them are basically the same as far as i konw minus what there lil startup bonuses are, like 10% tech bonuses, i say we work on the ai now and then people that need a story go play teh campaign, or we work on there individuality after froggy gets a nice structured ai system up and working, weres the fun in a game were u dont have to try when the games set to ridiclous setting, i just rhomper stomper threw with a few good units (single stacked btw). Keep up the good work tho froggy def see that your dedicatedl. Need some game strat tuts, or wlakthroughs let me know im all up on addyz 8) so i dont mind.

Interesting post, thanks. I didn't know the AI in GalCiv2 could predict the future a bit, that sounds very fancy. Good luck with teaching your AI!

Most people realize that it's not as gratifying to play against an AI that "cheats". People want to be challenged based on the effectiveness of their strategy, not simply fight some mindless and artificially inflated horde. That takes the "I" out of "AI".

Anyways, great work. It's not easy spending long hours fleshing out details to something that seems simple in concept.

I'm currently on a dual core laptop.  I'm waiting for solid state drives to come down in price before upgrading laptop, which should be late this year/early next year.  I have to game on laptops.

I do however find it gratifying to stand victorious against even a dumb AI opponent whom has both a tech and numbers advantage over me.  If the AI is unable to be dynamic with it's tactics, and is efficient in its use of combined arms, then I want it to have a numbers, and perhaps a technological advantage.  With Elementals tactical battles, we humans can easily stomp an AI who even has a slight military advantage over us.  We can use combined arms to a lethal affect.  We can alter our tactics on the fly.  The AI can't match the power of the human mind.  So I don't mind a handicapping of the AI. 

Ideally I'd like to see an AI which needs no external help.  But at this moment in time, I don't know that a competitive 4X TBS AI is realistic.  I think there may be some hardware and software limitations which will deny us this for some time yet to come.  That is not to say that crafty coders won't be able to maximize AI potential.  Clearly there is room for innovations such as Brads multi-threading.  Creative workarounds along with the brute force scripted approach will yield some wiser AI.  But to match a human, I think 4X TBS AI is gonna need some hardware innovation first.  For now, I am cool with a handicapped AI.  The AI isn't cheating, it's simply getting a competitive handicap.   Just as with many sports, a handicap is used to makes things interesting for the both the skilled and novice players.  There is little challenge in the skilled player stomping th4e novice.  Give the novice a handicap, and the contest becomes more interesting.

As far as AI in general, while I would like to see some personality, I also don't want it 'locked in'. If I always know that Umber is aggressive and tries a X type of strategy, that's no more interesting than if all factions did the same strategy. The only surprize is who I meet. I'd rather see randomize personalities/strategies so that if I see Umber I don't know if they're going to be aggresive or defensive or what kind of unit they're going for.

Also, one hitch with personality based AI is that they -aren't- as interesting as something which can be more flexible. Humans can be creative, true, but they can also be very predictable as well. After all, most of us probably build a hut and workshop within the first few turns. If the AI is set up to do the logical thing, it takes away predictability for it to do something outlandish...but that may be equally effective or at least, very interesting for the player to see. Similarly, if the AI is thinking too much, it may just be bland. An example is from the review of GTA4 - the reviewer noted that in MP, players would always do certain things and tend to play in certain ways. The AI on the other hand would provide really interesting situations such as chasing someone up on to a suspended I-beam and having a gunfight there. 

As for the questions...

1. On Challenging or higher, warfare 1 should be one of the first few techs (depending on research speed). Certainly on these difficulties at around 100, Warfare 1 should be being researched, just started, or just done as it is at this point that the starting equipment (assuming no faction bonuses) will significantly drop in effectiveness against mobs. This can be delayed for more risky epansion and such, but only if there isn't immediate danger such as a powerful mob or mob stack (bears with their high HP, highwaymen with their high damage, etc)

5. By turn 100, 4 single-soldier units is likely enough to provide a basic militia (it will deter the mob AI and will at least give pause to a player especially if roaming allied units are near). On cities in danger (whether currently or potentially), 6-8 is enough to deter mob AI and will be a significant obstactle. However, these numbers are not enough to be aggressive with (as in, sending out more units to support field units or perform sorties). This is also the bare minimum - while it may deter a strong stack (especially with field suport), this amount will probably fall under a strong stack (even though it will probably hurt te stack significantly). This also won't survive under a concentrated attack (two or more stacks attacking on the same turn or within a few turns). Groups-of-solders aren't needed yet at this point.

6. Heal and bursting regeneration is helpful within the first 200 hurts. For weaker champions who can't stand toe to toe, they can 'add' hp to other units and let them take on a few more units. However since heal does not scale at all, this will prove less useful later on (6 hp when you have 200 hp and the other guy is doing 20 in 1 attack is unlikely to be very useful).

7. Even a few points in armor and dodge can be very effective in the current set up. the sand golem is functionally invincible for the first 100 turns (a 6/7 creature), if you need a general gauge. Based on 'feel' it takes around a 4 point different to be effective (4 more damage than armor) or at least, seem that way.

Food is -not a high priority; it's a seond or third priority. On a decent start (2 food sources near by), one can usually hold off on building the second food source for a while. But generally, it is the mobs and production time that limit growth, not food.

Population is also -not- a high priority assuming you don't spam relentlessly. While this might not maximize other areas, it does allow for greater overall options as one can always build in different ways depending on the situation rather than, functionally, always needing huts and food to do something.

One thought would be that the AI always tries to save a little for the future; rather than using everything and then cutting back, use up almost everything and 'dip' into the savings when it's desperate

I'll be talking about how modders can do "stuff" to the AI via the XML more in the coming year once Stardock's new Mod Czar starts.

Hmmm, ... it does not bode well, as the modding god has now been reduced the modding czar...

Hi Santa Frog, are we getting a new EXE in a cute .gift wrap tonight or have we been naughty kids ?

What is a mod czar exacalactly?

I'm going to buck the trend and disagree, on principle, that there is a limit to how much an AI can challenge a player without cheating.  A game operates on a set of rules and constraints (in the case of a TBS game, typically a lot of rules and constraints), which computers are exceptionally good at following.  Things just go south because we impatient players don't want to wait 20 minutes between turns for the computer to fully evaluate every facet of strategy  We just need (significantly) stronger processors to see better AI.

Brad, has anyone tinkered with using GPU processing to boost the processing power of the AI (CUDA, GPGPU, etc)?  It seems to me like a fragmented "auditor/agent" style AI with hundreds or thousands of monitors evaluating a handful of inputs to provide data to a master process would be relatively well-suited to the kind of massively parallel processing that the video card brings to the table.  For ideal applications I see people get 100-300x speed increases over quad core processors (I got 100x for one of my own), so it may be worth taking a look at... if not for Elemental, for a future strategy game.

Edit: Of course, AI on the GPU takes cycles away from graphics... but I'd personally prefer smarter to prettier for my strategy games, and I'm sure that's most strategy players.

The graphics arent that hot to begin with so I say go with it. I'd prefer strategy as well

@Brad: Great post! I love how you include pictures. Lookin forward to future AI matches like you did with GalCiv2 

An AI could easily beat a human if the game was designed for AIs and not for humans. If the game gave advantage to extremely complex calculations, then the computer would beat a human incapable of such. IF damage calculation hade tons of variables interacting, then an AI could predict optimal gear for a unit better than any human, but the game would be no fun.

An example:

If each item of gear would slow down the soldier, but also adding armour and attack depending on the item. Items could also act differently depending on the unit wearing them (based on classes like scouts, rangers and such). This could be made insanely complex to give the AI and advantage. The problem is that the whole designing unit thing would be a black box for humans.

Re: AI Saving up

In Kohan 2 - there's a value called "saving time limit" where the AI will consider it's current income rate per minute * (saving_time_limit / 60) as how much gold it could have at the end of that limit. If that result is higher than the gold it costs to perform that gold (say researching a unit tech) then the AI will not only save that gold, but ear mark it. So if there's was something else it could do now, it won't UNLESS that goal's cost would not sap too much gold.

Would it be possible to do something like that in Elemental? Something a "population saving limit" where it considers how fast it's getting citizens, how many it would need, and whether or not it can achieve that amount when by the time specified in the populating saving limit threshold?

That would seem to help with the AI not wanting to wait for future big goals like major buildings or larger war parties.

AI population for troops issue:

I think that this could be partially fixed by making population more of a local resource - as in the city has to have enough free population in order to build the structure or train the troops. You are already tracking the local population for cities in the city size/city level bar, so why not have a %population employed? The way the game is set up right now, I can have 1 giant city with thousands of people in it and lots of tiny cities everywhere that do not have enough people locally to staff their improvements using the mega-city's population to run everything. If you shifted more towards a 'population as a local resource' model, you could have the AI decide whether or not it wants to build an improvement or train a group of soldiers in each city it has. The problem with changing to the local resource model is that individual city populations do not grow fast enough to support a 'local population resource' model with the current population costs on structures - perhaps a fix could be that population growth becomes (local prestige)*(city size)? Prestige seems to be small enough anyway unless you go out of your way to build prestige boosters in each city, and if it were seen to be a problem you could cap the number of prestige boosters or have a prestige penalty for higher level cities.

As an example, assuming a level 5 city has 900 people as its pop cap and is using (for my convenience) 15 city tiles on housing (60 people per house), if the average pop use/tile is 20 then the city structures are using only 700 of the local population, leaving 200 available for recruitment. Also, I am fairly certain that an average of 20 people used per tile is ridiculously high since most of the structures that get built are going to be prestige boosters (for an Empire, I think it is 10 pop per monument), tech structures (5 pop each for everyone), and houses (which provide pop).

Side note - if I build a garrison/barracks in a city, why don't I at least get some minor combat troops based on my warfare/conquest tech level for having such a structure in the city? The building requires some population, if I can entrust these population people to train my troops why can't I have them help defend the city? And I cannot think of any reason why I would build a garrison and then fail to garrison it unless it was in a bad location or I did not have the money - which could be in the building upkeep. This could also take some of the pressure off of the AI pop count as it might not see as great a need for troop training if its Barracks gave it something like (5 or 10)*(warfare/conquest tech level) combat rating of 'city patrolmen.'

Whoa, whoa, whoa.  I'd just like to ask, for a moment, why on earth the AI has to "save up population"?  I have never, outside of the first five turns of the game, had to wait on population growth.  Never.  Not once.  It's a complete and total non-issue.  How on earth is the AI having trouble with this given their immense prestige boosts on higher difficulties?  I genuinely do not understand this at all.  I've always felt that population was pretty meaningless as a restriction, as currently designed, given how easy it is to come by.

In an attempt to answer my own mystification, I have noticed that the AI tends to avoid building housing.  It ought to be doing roughly the same calculation that I routinely make when considering what to build next.  Specifically: "my prestige in this city is X, a unit of housing takes Y turns to build, I'm considering building something else that take Z turns to finish, and I have A room in my population cap.  If X*(Y+Z) > A, then build housing first."  That avoids putting prestige to waste, which in turn gives both abundant population units and good cash flow from taxes.  Obviously critical buildings are exceptions, but there has to be a standard before there can be exceptions.

I'll be talking about how modders can do "stuff" to the AI via the XML more in the coming year once Stardock's new Mod Czar starts.

Thanks for the heads up.

I've had to do it frequently.

That bad, some population structures are rather underpowered- take the tower of sorcery for the empire.

It provides 4 arcane knowledge- but requires more people then 4 arcane labs.  (it's a little quicker to get up, but requires lvl 2 city)

Pop requirement should be reduced on that building.

I run into pop issues frequently, if I don't luck out on food.