AI questions

[UltraSonix]

The AI should be able to win using other methods, but I think it'll be too hard to contruct an AI that is smart diplomacy-wise. If it could be done, then the game'll be awesome, but we can't have everything in life... So if not a trade/diplomacy good AI then Firaxis should at least try to build a militaristically-intelligent AI.

------------------
No, in Australia we don't live with kangaroos and koalas in our backyards...

[Bkeela]

I had to add my two cent's worth into this topic, (alas I've not had time to read all the comments; so forgive me if I'm repeating what has already been said), but the reply of Mark_Everson particularly disturbed me, thus I had to comment:

First of all, it is pointless to use Civ II as the comparison for an improved AI. Either SMAC or CTP should be used.

"0) AI needs to expand more quickly, especially at the start. It is waaaay too conservative, and incurs huge "opportunity costs" in size due to slow expansion. The AI should also build cities in niches on the shore that have only a few land squares but could develop into decent sized cities with a harbor..."

The AI employs this technique in SMAC, and as a result, the AI goes expansion crazy. Instead of developing the cities it already has, it just spreads out. Although this may be a good strategy, it is very annoying for the builder/diplomacy minded players.
A better compromise for a weak AI, would be for it concentrate on building a moderately sized civilisation, with quality cities, rather than numerous poor quality cities.

"5) AI should build more diplomats/spies and Use Them. Even used in a ham-handed manner they are pretty nasty. Bribing cities when enemies have a pile of units around the city are especially damaging to the enemy. If you could add logic to let the AIs use the ZOC-cancelling functions of diplos and spies AI attacks and counter-attacks could be Much more devastating."

There has to be a balance between a really tough AI, and the pleasure of playing the game. To me, having the AI bombard you with spies would make the game unplayable. I've heard that with CTP, all these annoying non-combat type units, like the Lawyer unit, have the potential to make the game very frustrating indeed.

Bkeela

[Lord Magnus]

Interesting topic, here's my suggestions on this (this tread is big and I didn't have the time to read it all. Sorry if I repeat some stuff).

The AI likes to spread its units out (in its defences as well as attacking), therefore someone with a large group of units can easily conquer the civ city by city. The number of defensive units should be proportional to a city's population and other factors should be considered. Such as: is the city a capital? How many wonders (that are not obsolete) does this city have? Is this a city on the edge of my empire? Has this city been attacked recently? The AI should consider this when placing its units or building defensive improvements.

The partisans should act like partisans. Whenever I take a city and partisans appear I take them out before I advance. (Aside: In Civ3 to make partisans more effective, they should be invincible in forest or jungle squares or squares that are not being used by the city's workers) On occasion I don't have enough units to take them all out, so they unfortify and pillage terrain improvements. This gives me a chance to mop them up and the damage they do is a minor annoyance. Partisans would be more effective if they avoided confrontation with normal military units and attacked units such as engineers, spies, caravans or weakened units. And they should pillage squares that are out of the way or near cities that appear to have no extra units near them.

The AI should also know that swamp and jungle squares without special resources are worthless and manpower should be invested to clear them.

On a relevant topic, I believe that difficulty levels should be based on AI "competence" as opposed to just handicapping the player's civ or cheating on the AI's part. This may not be feasible until the AI is "competent" and hopefully this is still not the case in Civ3. But basically at chieftain mode, the AI barely tries. As we advance on levels the AI tries more tricks and lets more AI civs play full out until deity mode where all the AI civs will collaborate and destroy you at all costs.

If Firaxis is able to code an AI that "learns," one could give the AI a better edge if there is a networking option that allows AI's from different machines to exchange information in an attempt to create the ultimate civ AI that eats human players for breakfast.

------------------
Learn the mistakes of yesterday to prevent the ones of tomorrow...

[Matthew]

One other thing we should not be able to do against the AI at the highest 2 levels or so. Beat it. At least not at first. In the book that came with Civ II you talked about the new Diety level as something that you were sure could actually be beaten. In doing so you made the game to easy. I'm sure almost everyone had a hard time beating the AI at first, but how long did it take for people to have to come up with exciting and more difficult ways of winning? I mean come on. People are reaching Alpha Centauri in the late 1600's with 1 city against 7 civs and with Raging hoards. People are reaching A.C. with one city while spending most the game in fundamentalism. I'm about to break the 1500 mark on a large world, though with a few more cities. Two people played as the neutrals against the AI's in the world war II scenario and conquered the world.

I suggest that before you sell the game,or better at a fairly early developement phase you need to find some of the best civ II players of varying styles. IE warmongerer, perfectionist, trader, infinite city sleezer, tribe of the divine style, and OCCC guru, give them a few weeks to test the game out, and make darn sure each and every one of them is still getting his butt waxed at the toughest two levels, and having a real tough time winning at the third toughest level. Don't just make it very difficult to win at the toughest levels, but so hard that in your mind it is barely concievable that someone just might win at the toughest level someday. Then ask the best players how it could be made harder still.

That way it will at least take a few months for someone to figure out how to reach AC in the game by 1200 A.D. at the highest level without ever building a science wonder or a science improvement in despotism...... well, you get the idea.

Then include a powerful AI scripting language so we the players can make it tougher still.

Remember that SMAC is a far newer game than CivII. yet once someone figured out that you could win civ at Diety with one city, how long did it take someone to do the same thing in SMAC?

By the way. If you kill the Camel in civ III I won't buy the game. I'll stick to Civ II.

[This message has been edited by Matthew (edited July 02, 2000).]
[This message has been edited by Matthew (edited July 02, 2000).]

[Sikander]

I realize that you (Fireaxis) may be well on your way with CIV 3, and these comments might be completely useless. Nonetheless...

The AI is not very good at navigating the map. Thus it is inefficient at many tasks which are map dependant like terraforming, moving caravans, tactics etc. My solution:

Change the game! Take this stuff away from the AI and player both to some extent. It's not like a lot of this stuff is at all realistic or fun anyway. Why should it take years to cross the Atlantic with a caravan for the human (boring) when all it does is give the human yet another advantage over the AI who sometimes can't do it at all? The same is true with terraforming. Why have settlers building roads and stuff (which gets mighty dull when you have zillions of them) when the AI is again so incapable of doing this well? Allow improvements to be made to terrain within the city screen through resource allocation.

By simplifying the areas where the AI is at a large disadvantage you can concentrate your efforts in other areas. Civ is not a great wargame and will never be one in anything like it's present state. Moving tactical units on a strategic map is completely unrealistic (as are the movement / time scales used in every era). Why keep this system when it isn't a good simulation, and is so complicated for the AI that it isn't a challenge either?

I would rather reliquish my ability to (unrealistically) micromanage the operational and tactical details of my armies in order to put myself on a more equal footing with the AI. I enjoy the management and building of my empire more than being overwhelmed by the minute details of unit movement. Since this also seems to overwhelm the capabilities of the AI to provide a challenge why not ditch it and spend your time building a more interesting management game? I really like the Social Engineering aspects of SMAC/X. Why not expand this element of the game by having the player build armies (from unit components) which are given objectives which are carried out (or not) by AI generals? This will eliminate some of the myriad player advantage by limiting his ability to beat the AI by serving as the leader of his Civ at all levels against an inferior foe. Let the player deal with incompetent (or treacherous) subordinates. Let the player act as a true emporer must, by first and foremost maintaining political control over his domain, and if successful, then turning his attention to the other Civs in his neighborhood.

By shifting the emphasis of the game from a 6000 long year wargame to a more realistic game of trying to build a long lasting political and cultural entity, you will shift the game into one where the player spends most of his effort against the management model, rather than thrashing an AI which will not be able to match his capabilities in such a complex game in our lifetime.

One more thing. Movement / Combat should be simultaneous. Letting anyone unleash a minimum of 1 year of combat power before allowing a reaction by his opponent is unrealistic, and the complexity caused by the large area, various terrain, large numbers of units etc. means that even Deep Blue is going to perform pathetically against even a novice human player. Check out PACWAR or War in Russia for a better simultaneous movement model.
[This message has been edited by Sikander (edited July 03, 2000).]

[The diplomat]

I like what you say!! Your suggestions are really good. I think they would make for a much more interesting game experience.
Your ideas are better than what I suggested but nonetheless, check out my post on the thread "civ3 needs better AI-would help with micromanagement too".

------------------
No permanent enemies, no permanent friends.
[This message has been edited by The diplomat (edited July 03, 2000).]

[Ralf]

I have never won a whole game over the AI as a result of clever tactical unit maneuverings - individual skirmishes perhaps, but NEVER the overall game.

I have ALWAYS won the games over the AI as a direct result of better logistics = better resource management + better unit- and city-improvement strategies.

Conclusion: A good AI-civ should march on its belly (= better unit- and city-improvement and city-area management, using force-guiding - but editable - scripts/ templates).
DON’T waste (to) much time in trying to mimic human tactical unit maneuverings – instead why not replace some special key-units all together with perhaps more “AI-friendly” screens instead? The later is the Diplomats (or Sikanders?) idea.

Just concentrate on below three things, and you should come up with a hell of a Civ-upgrade:

* The AI city-management problem (reduce AI-involvement - read above).
* The ICS-problem (Infinite Cities Sprawl) – the original extreme A variant.

Variant B: The semi-ICS problem (building 12-18 cities and 2-4 early Wonders BEFORE building even the most basic temple/marketplace structures)
Variant C: The “25 cloned New York’s syndrome” (halfway ICS - then perfectionist-style parallel city-improvement upgrades, the rest of the game. Trying to cram each-and-every improvement into each-and-every city should be much more prohibitive).

* Introducing the “Rice and fall of empires” idea. (alternatively: combat the "bigger is only and always better" problem). Read more in the http://apolyton.net/forums/Forum6/HTML/001537.html thread. Also, use search-command and the keywords “rise and fall”. There are lots of good posts tucked away all over the place.
[This message has been edited by Ralf (edited July 08, 2000).]

[jsorense]

Gord,
Thanks for remembering me. (-:
I am pretty much a lurker here since I have no strong opinions on these topics.
Your posts are always thoughtful and knowledgeable and worth seeking out.

[Sikander]

Ralf,

Your point about superior player economic management is well taken. It is true that my superiority in this realm is what gives me a majority of my edge. Still, one of the reasons that I have this edge is my confidence that I can whip the AI in combat even though I have inferior numbers, and sometimes inferior troops as well. Thus I don't waste time building much military force until I really need it, as I am confident that I can blunt the AI's weak attack and build modern units quickly and efficiently to turn the tide. This allows me to invest early on in infrastructure instead, which usually produces a tech and production lead that the AI cannot compete with.

I do think that too much of the AI programming effort is 'wasted' trying to improve the map based functions of the game. It is easy to see the difficulty here, where AI formers/settlers are terrible in comparison with my own, even in SMAC which is a third generation (at least) product. I don't think that the programmers are stupid, I think that the problem is just very difficult. It is my opinion that more effort in the design of an AI economic model would produce a more interesting and competitive game. Does anyone with programming experience care to weigh in on my supposition that global screen based functions (like spies in MOO2) are simpler for the AI to master than CIV's walk everything from the production line to it's target system?

[PrinceBimz]

The AI don't use aircraft carriers in Civ2 and should be programmed to do so.

[mgblst]

Ok,

now create the AI as best you can, but allow for an option for us to create our own AI, and add it into the game. For example, one would assume that you would have AI procedure, like Get_Next_Move, Change_Tax etc. now just allows us to write our own procedures, put them in a dll, and allow the program to call us, say as a guest AI. This way you could allow different leaders to have different ways of doing things, and we could create our own leader, with its own programming.


This way, we could also upload our fav AI, just as we upload modpacks, scenarios, etc.


This would be quite difficult to do though, depending on how you have it implemented now.

[Lord Magnus]

Mgblst your suggestion would be crucial in creating realistic scenarios. If this was implemented one can create a actual Hitler in a WW2 scenario that will try to conquer Europe according to details taken from the Mien Kamp. Also this can prevent the AI from just arbitrary cancelling alliances as it likes to do in a scenario. The possibilities for a "Custom AI" are numerous. The best way to implement it would probably be as a scripting language.

------------------
Learn the mistakes of yesterday to prevent the ones of tomorrow...

[UltraSonix]

Lord Magnus has hit mentioned a critical problem with scenarios and AI. While AI could be set in civ2 using events so that civs couldn't talk to each other, more control should be given to the user so that AIs can also be programmed to not break alliances etc.

[Gord McLeod]

jsorense: Thanks! I rarely post these days myself, others seem willing to step up and say what needs to be said. This thread is one of the exceptions, there seems to still be a great number of ways being posted to improve the military performance of things, but I really don't want to see the game become horribly misnamed... there's a reason it's called "Civilization" rather than "6,000 Years of Global Warmongering".

[Mad Lord Snapcase]

What I'd like is a good "apprentice" mode. Let me explain. You'd play a regular game, with "Apprentice AI" as a chosen option. Whenever you did something new for the first time, say build a settler, A dialogue box would pop up asking you why you did that particular thing. Say, you use combo boxes to explain that the first two units you build in a new city are always settlers. The AI will now stop prompting you about settlers until you deviate from this rule, in which case it will ask "Why did you not build a settler as outlined by this rule?" Whereupon you'd answer that the other civs are attacking or that you don't need any more cities because you already have x many cities. You'd go on answering more and more complex questions for about half a dozen full games (the more games the better the apprentice will know your playing style). The "apprentice" will be kept in a file, which can be used as a list file to tell the AI what to build and when, how to attack and when, and so on. This can be loaded onto any civilisation.