Is random maps a AI weakening factor?

[Ralf]

The AI simply cannot utilize the map the same way as an experienced civ-player can. How many times havnt you seen AI-civs with innefficient city-locations? Big islands which easily could support 12-15 efficiently places cities, but only housed perhaps 7-8 AI-cities in the end-stages of a Civ-game?

The main problem is the use of randomly generated maps. Maps that, of understandable reasons, cannot be analysed by an AI-programmer. Its generally easier to program a good AI to campaign-type games, because they use 100% known setup-maps (that IS carefully analyzed and integrated to the game, by the developers).

I dont advocate axing the map-generator. What i would like to see is an add-on alternative. I shall try to explain the idea by a real-life example:

In big city railway-stations you often find a big self-standing electrifyed tourist-map. Then you push a button representing any tourist-attraction, a little red diode lights up on the main map.
Now, what i would like to see added in Civ-3, is a bundle of analyzed handmade maps, with lots of imaginary game-developer placed "red diodes" (= representing possible ideal AI-city locations) on each and every of these special maps.

These "diodes" are of course invisible for the player. Then the first AI-city is founded (on such a red diode spot), the AI has of course produce resources equivalent for a settler-unit, in order to expand to a second AI-city.
But instead of trying to make an AI that can send a settler-unit from any AI-city, then effectively trying navigate that settler on an totally unknown map - only with help of clumsy "if A and B equals X, then it will do Y" type of AI-rules - why not instead do this instead:

1/ The first AI-city produce the equivivalent recourses of an settler-unit (no cheating!).

2/ If an potential "red diode" AI-city location is perhaps 6 squares away from the settler-producing city - the AI simply have to wait 6 turns: then that second city pops up automatically on that predesignated square/location.

The AI can ONLY found cities on these predesignated squares, carefully dotted all over these maps. The Human player can (of course) found hes cities anywhere he wants.

The benefit? Perfectly placed AI-cities, more efficient AI civ-expansions, on effectively utilized islands/continents.
The only snag would be that the AI-programmer would come up with an method to sence human-player precence, and never allow the AI-civ to found cities TOO close to already established human player cities.

(I remember SMAC, then that religious ***** founded a city about 2 squares away from mine . Not nice).

[This message has been edited by Ralf (edited October 06, 2000).]

[UltraSonix]

Hummm - to do this, it almost requires the AI to be open source (so that the map-making can understand what to do).

------------------
No, in Australia we don't live with kangaroos and koalas in our backyards... Despite any stupid advertisments you may see to the contrary...

[Adm.Naismith]

Add to the map generator the role to build "city place" as part of the "seed" of terrain generator:
Add them to usable terrain, like:
- plain terrain
- hill
- "good spot for food city"
- grassland
- "good spot for production city"

This way you can write rules that not only cope with realistic succession of sea, shore, plain, hill, mountain (e.g.), but also consider from scratch where city are more profitable.

Add this to your idea of a list of "hidden" predefined city place, and you should have two birds with a stone: easy life for AI and random map.


------------------
Admiral Naismith AKA mcostant

[Dienstag]

I finally got my hands on the original Civilization board game by Avalon Hill over the summer. (I only got to play it against myself )

For those of you who don't know, the map is mostly of the Mediteranean area (apprximately from Corsica to Mesopotamia, and from Crimea down to middle Egypt) and is divided into regions like in Risk. Each region has a number (1 to 5) that is its maximum population, and some regions have a black or white square. The square represents a good city site and makes it much easier to build a city in this region (white squares mean city is vulnerable to floods).

So, the point of all this is that the original Civ board game had a good number of city sites already picked out. How should this affect CivIII? I don't know.

[Ralf]

The problem, Dienstag, isnt about human player-related difficulties in identifying what areas on the Civ-map is suitable for city-building, and how the human player should utilize that map in best possible ways.
If that was the problem we would probably discuss the graphic design of those indevidual terrain-tiles instead.

The problem is about artificial intelligence (the AI), and its inability to compete with us human players - at least in such a complex and option-divided game like Civilization.
This thread and http://apolyton.net/forums/Forum6/HTML/001531.html? is about ways and ideas in how to "level the road" for that AI somewhat. Making it easier for it to compete against us.

And it really IS an important problem because playing Civ through the internet can never be anything else then a secondary option. There is simply to many "if´s" and "but´s" then playing such a complex and longlasting game like Civilization online in multiplaying mode, with real human counterparts.
Most of us know the problem of quitters, the problem of different timezones, the problem of trying to gather enough contenders at the right time, the problem of unnexpected changed prioritys that causes one contender to leave for several days (or for good), the problem of slow unreliable connections, and so on.

Im not saying we shouldnt have the multiplayer option in Civ-3. It shall of course be added. But just as an excuse in exchange for that same samey and uneffective AI we experienced in SMAC? Noooooo thanks!

[Nick-oh]

A thread entitles "Where's all the Farmers" which ran it's course a few weeks ago (I think it was started by Hannibal3; I apologise if I'm wrong) was primarily concerned with the lack of rural population, but contained some ideas which would apply here.

The part of the thread applicable was about the way cities are founded, and the gist of it went something like this.

Land squares would each have to contain a certain rural population.

In regions with large amounts of exploitable resources, the rural population would migrate to these places as they provide thim with a means of living.

When an area grew to a certain population, a city would form there. (How and when that city is brought into your civ is for another thread)

This wouldn't mean you have no control on where cities go, because designating an area for settlement subsidies and building infrastructure in an area would attract people.

Now to how this is relevant to AI city placing:

The AI would now concentrate it's city building (via the construction of roads, military instalations, etc)into areas where the rural population has been attracted toward (i.e. "good city sites")

I hope that is clear enough because I have no more time today.

------------------
"Common sense is not so common" - Voltaire

[Ralf]

Well, why not - infact that was a very good idea, Adm.Naismith!

However, instead of using single indevidual terrain-squares as guidance for AI "city-place" seeds (that one good square could after all be surrounded with generally bad terrain - or the other way around; a whole area consists only of really good-terrain squares), the map-generator should be able to calculate average values from multiple of potential 21-square AI city-areas (= city-places).
The lower life-sustain threshold-values for such AI-city generated places should of course be manipulatable by the human player - giving him some control over what the map-generator AI should accept, or not.

Infact, once the continents of the map is generated, the map-generator then could calculate potential 21-square "city-place" (or rather "city-areas") locations from up-left to down-right, shoulder-to-shoulder over the whole map. Only those areas who had its central square on terra-firma would be account for.
Above by itself however, would give rather bad results because small 6-10 square island/capes would often be left out, and also not less important: the general AI-city distribution would be relatively squarish and unimaginative (= ugly).

The shoulder-to-shoulder generation of AI city-areas should therefore be alloved to overlap, disjoin, disfigure each others areas max 1 square-row (the outer 12 squares - the 9 inner ones are untouchables). in random ways.
Also, this first map AI city-place sweep is followed by a second one that distribute its potential city-places in an alltogether different fasion. Again, this is perhaps followed by a third sweep.
If the human player chooses that, some few city-places are plucked out between each new sweep, in order to give the end-result a more natural feel.
Finally, any second/third sweep coustal city-place (or "red diode") should always overide any first sweep 2-3 squares-away inland city-place. The reason for the latter is that the terrain on small/medium-sized island can be utilized much better, if coustal city-places have priority.

Above idea, together with the map-EDITOR´s ability to allow the player to manually pin-point invisible "red diode" AI-city places on hand-made maps (not to be confused with placing visible AI-cities in scenarious), really would give the Civ-player "the best of both worlds".

[This message has been edited by Ralf (edited October 06, 2000).]

[Ralf]

I have read Hannibal3´s thread http://apolyton.net/forums/Forum6/HTML/001651.html To be honest though, i didnt like the idea that much.

- First of all this idea deprives the human player from the pleasure of topographically modeling his growing empire completely by himself. (Hannibal3 quote: "Instead of having to be constantly concerned about moving into those new territories, the game does it for you").
Well, i really WANT to be concerned about moving into new territories! Thats one of the fun parts of the game! I really d o n t want the game-AI doing it for me.

- The AI-problem! Artificial intelligence has its non-negotiable human-thinking limitations. Its can only be really effective then its restricted to calculating with "if A and B equals X, then it will do Y" type of objective non-fuzzy rules.
Instead of comming up with a more-workload-on-the-AI suggestions that previously had been done by the human player (hence; "Instead...the game does it for you") - we should rather come up with ideas that can BYPASS the AI as much as possible.

Thats the whole basic concept behind this thread and my other thread; "Arguments why its nearly impossible to create an almost human AI".
The two main areas that the Civ-3 AI programmer should primarily concern himself with is unit-moving and diplomacy. Not because the AI does that very well, but because the AI-developer - to the best of his ability - HAS to do it. The reason for the latter is that premises and circumstances in above areas, are constantly changing in unpredictable ways, from turn to turn (theres no "firm point" so to speak, from which the programmer can use automatic "premade templates" or "blindread maps").

The map-generator "Red diode" city place idea is just such a "premade template" that can bypass the workload of the AI. This idea really would make AI city-spreading more effective in theory.
Perhaps there are stumbling blocks when it comes to trying to put this idea into practice? Perhaps someone who read this and have more hands-on AI-programming experience than i have, could comment on that.

[This message has been edited by Ralf (edited October 07, 2000).]

[DarkCloud]

Map generating instructions:

CityPlace #SQUARES from enemysettle$, if more than 10 Place
CityPlace #SQUARES from friendlycity$, if more than 3, Place; if avail continent spaces=30 then place less than 2 away.

[Ralf]

Check out these links that i found:

Computers/Artificial Intelligence: http://dmoz.org/Computers/Artificial_Intelligence
Computers/Artificial Intelligence/Games: http://dmoz.org/Computers/Artificial_Intelligence/Games
GameAI.com: http://www.gameai.com/ai.html

By the way - http://dmoz.org/ is one of the very best links-archives out there!
Quote: "The Open Directory Project's goal is to produce the most comprehensive directory of the web, by relying on a vast army of volunteer editors"

[This message has been edited by Ralf (edited October 08, 2000).]

[Dienstag]

I just offered the Civ Board Game info in case it might have been useful. I personally think the AI should work along the same lines as the human. If the AI knows where the "red diodes" are, then I should have the option of knowing too. On the other hand, is it really so hard to program an AI that builds cities reasonably intelligently?

The AI civ's size, personality, and availability of land will influence whether or not it builds a settler. City location will depend on distance from other cities and a sincere effort to maximize resources. The AI should be able to figure out (not be told) if it's on an island, and decide whether or not it wants to build cities closer together accordingly. I don't think this would be very difficult to program, or be insanely hard on our computers.

While I'm at it, wouldn't it be nice to have an AI that could pass the Turing Test with Civ (that is, you could play a game against it, and not know that it wasn't human) This would involve messages and so forth, and I don't expect this kind of thing anytime soon. Oh well.

[Ralf]

quote: Originally posted by Dienstag on 10-09-2000 02:05 PM I personally think the AI should work along the same lines as the human.

Thats a good one! (by the way, have actually read the linked thread in my previous post to you?)

[Dienstag]

Ralf, I had read the aforesaid thread long ago, and I just now skimmed through it again. Many good ideas there, and many ideas that I'm sure I could disagree with. I guess I'm one of the more optomistic people in terms of improving the AI without resorting to templates and such. I'd (try to) program it myself if I had the opportunity. In the end, it doesn't matter how they do it, as long as the AI improves to the point that we can consistantly play a challenging game.