Fuzzy Logic: beyond deduction and intuition

[Arkimedes]

I hope that the AI of C4 uses some of this logic (http://www.azer.com/aiweb/categories/magazine/24_folder/24_articles/24_fuzzywhat.html): not only “yes” or “no” (the classic programming options) but also “maybe yes” and “maybe not”.

This way we might have a challenging AI, that could deceive us, trick us, play with us (back).

This is indeed a challenge to computer programmers. I consider a challenging AI much more important than some excellent graphics or impressive 3D.

Better diplomacy, politics, religion, treason, alliances, corruption, dictatorship, slavery, culture – in a word ROME, could be alive again. Don’t you agree?

Arkimedes

[Keith]

It depends on what you mean by fuzzy logic; I mean specifically.

Some of those items catagories you listed don't really seem to fit into a "citybuilder", IMHO. Those are more suited to a strategy game built around warfare, where you try to rule what you have. I can see it in a game like Rome Total War.

[wodinoneeye]

I hope that the AI of C4 uses some of this logic (http://www.azer.com/aiweb/categories/magazine/24_folder/24_articles/24_fuzzywhat.html): not only “yes” or “no” (the classic programming options) but also “maybe yes” and “maybe not”.

This way we might have a challenging AI, that could deceive us, trick us, play with us (back).

This is indeed a challenge to computer programmers. I consider a challenging AI much more important than some excellent graphics or impressive 3D.

Better diplomacy, politics, religion, treason, alliances, corruption, dictatorship, slavery, culture – in a word ROME, could be alive again. Don’t you agree?

Arkimedes


It not that much of a challenge to the game engine programmers. Implimenting Fuzzy Logic rules into decision processes in games is already a well understood process. For the game mechanics programmers and scripters is where most of the effort is. Balancing a complex web of influences on behaviors can be fairly tedious (handling all the different modal situational factors and their weights and preventing excessive pendulum swings to match a desired simulation...) Ontop of that comes the individual scenarios that each have a different set of situational factors (and different balance points).


It take much more planning and understanding of the underlying mechanisms than the on-off logic that the CotN game used for the higher level scenario stuff. I hope that they lean more into this kind of mechanism to get a richer 'world' game, but wonder if the editor provided to the players will be good enough (the COTN one was abysmal and IT was only to handle data for the simpler on-off type scenario mechanism).

[Arkimedes]

It depends on what you mean by fuzzy logic; I mean specifically.

Some of those items catagories you listed don't really seem to fit into a "citybuilder", IMHO. Those are more suited to a strategy game built around warfare, where you try to rule what you have. I can see it in a game like Rome Total War.

Quite the opposite. Slavery, for example: after important victories over their enemies, the Romans took, as spoils of war, men and women to become salves on a distant land, preferably.

This could be a problem: more labour (salves) could lead to unemployment of the "free" men (roman citizens). More: slaves cost money that could lead to corruption, inflation, etc.

Slaves, more or less bring to the empire their own culture; with time it could influence the roman culture. The roman people had a keen appetite of "exotic" things.

Politics: don’t you think that elections, propaganda and alliances (diplomacy) could be established in order to win power to rule the Roman Empire? If you loose your elections your faction could find it self in troubles to build, declare war or making any alliances. So, it could be the same as loosing the game…

The items above would be possible with the right balanced scripts. But my understanding of “fuzzy logic” is more, much more than balanced scripts. Do you remember that I proposed some sort of path carrier? I tried to join some imponderable items to the game in order to make it more realistic, harder to play-and-win, and less repetitive.

Well, this is a step forward: the computer programmers / designers are more and more under a great stress to do better quality graphics, more realistic and this could turn into a spiral that a few will endure. A few years ago, 1 game per year would be enough to play over and over. In nowadays, after 3 months (or less), no-matter-what-game is “history”.

So, the answer is on a more intelligent AI. It must be much, much more than great scripts. It must learn with the player (like real people do).

For example: if the player uses over and over the same “layouts” while building a city, the “learning AI” could spot some weak points and beat him on their own (player) terms.

Other example: if the player uses systematically slavery as main-labour-power, after a few scenarios this strategy would turn into a Spartacus-time-bomb.

Imagine what strategies (logic and non logic) the AI could learn worldwide if the it was a multiplayer game!?!

Arkimedes

[Keith]

A more intelligent AI would be welcome, but one that would really learn is probably a bit much to expect given the status of the "average computer" it would most likely run on. I'd would like a better, fine-tuned, branching AI that wasn't quite so predictable

They might call it something else, but I don't think that you'll see slavery in Caesar IV. They've carefully avoided such things in all their past games, and considering that they are doing Caesar IV for VUG/Sierra's publication, that topic specific item is more than likely not going to be on the list.

The political aspect of such a game, as you suggest, I feel, is better suited to a a game like Rome Total War were the focus is on extending power, warfare, and ruling the whole empire from Rome. That's not the focus of a citybuilder, since you will be playing as a governor appointed by the Caesar, more than likely. You aren't going to be ruling "the empire" from Rome but rather just your current province.

My guess is that any diplomacy is going to be represented similarly to the way it was done in Zeus, Emperor, and CotN, which is a relatively simple and streamlined method. I would assume that they'll probably use a method that might be a bit closer to CotN where you have to have emissaries to open new trade routes, and maybe they'll have us buy that trade route when it time comes when it becomes available.

Great graphics is always going to be a major component of any game. Often it is what get's people "in the door" and attracts their interest at the very first. So I don't agree with the more than common assumption that the emphasis on great graphics is necessarilly bad for any game. Of course the game must have some depth to keep people interested in it. The fact that people can't keep an interest in something for more than a few months at a time before moving on to something else is not always the fault of the game, but rather the short attention span that many gamers have developed over the years. They've become spoiled by the number of game introductions each year.

[MarkDuffy]

There are many, especially the new players, who believe TM already has used fuzzy logic in COTN!

:D

[Arkimedes]

A more intelligent AI would be welcome, but one that would really learn is probably a bit much to expect given the status of the "average computer" it would most likely run on. I'd would like a better, fine-tuned, branching AI that wasn't quite so predictable

A more intelligent AI must be not only reactive but also proactive. If it could "underground" monitor the player methods and then "strike" making his life miserable - that's my point.

C3 was an imperfect game, too unstable: somehow difficult to play because the market ladies weren't doing their job, or the perfects were going always on the wrong way. On my point of view, this was the secret of its success: turning the player life miserable.

This could be achieved also by intelligent methods and intelligent AI.

A have a P4 at 2.66 GHz, 1 GB Ram and a NVIDIA GeForce FX5200 (256 MB). This is already an old (low specs) computer concerning the recent released games.

I like good graphics but I can't (I won't) afford to bye a new computer every 6 months.

Arkimedes

[wodinoneeye]

A more intelligent AI must be not only reactive but also proactive. If it could "underground" monitor the player methods and then "strike" making his life miserable - that's my point.

C3 was an imperfect game, too unstable: somehow difficult to play because the market ladies weren't doing their job, or the perfects were going always on the wrong way. On my point of view, this was the secret of its success: turning the player life miserable.

This could be achieved also by intelligent methods and intelligent AI.

A have a P4 at 2.66 GHz, 1 GB Ram and a NVIDIA GeForce FX5200 (256 MB). This is already an old (low specs) computer concerning the recent released games.

I like good graphics but I can't (I won't) afford to bye a new computer every 6 months.

Arkimedes

The solution you would get would not be a generalized (real) AI mechanism.
Instead it would be a some kind of self tuning pattern matcher (with preprogrammed patterns to match the game specifics/scenario). Im not really sure that the average player would play a scenario enough times for it to tune/learn enough to make it worth while (the individual scenarios are just too many hours long in a City Builder).

A better use of the effort might be to have the game have more sets of strategies/countermeasures to use against the player and probably scramble them each game to offer more diversity in the way the game plays (also if the game actually plays as an opponent -- a risk/resource manager that makes halfway intelligent reactions to the players moves-- again using precanned reactions, but applying them differently/optionally).

[Arkimedes]

The solution you would get would not be a generalized (real) AI mechanism.
Instead it would be a some kind of self tuning pattern matcher (with preprogrammed patterns to match the game specifics/scenario). Im not really sure that the average player would play a scenario enough times for it to tune/learn enough to make it worth while (the individual scenarios are just too many hours long in a City Builder)

Long hours? Well, I played C3 it till dawn!!! Jeese, how I loved that game!

On a multiplayer game via internet (??) it would be possible to monitor strategies/reactions of different players on real-time and then use it against them.

Maybe, just maybe, it would make the patches useless because the game could "recover/cure" itself.

If some reaction or pro-action of the AI didn't workout right on a precise situation then the AI wouldn't risk it anymore, at least with that player.

Imagine the situation: "Arkimedes? humm, I know you! You're dead!" (AI thoughts).

Lets face it: the major objective of any game is to beat the gamer on their (game) own rules. So, on fair terms, it must make the player life miserable! :)

Arkimedes

[wodinoneeye]

Long hours? Well, I played C3 it till dawn!!! Jeese, how I loved that game!

On a multiplayer game via internet (??) it would be possible to monitor strategies/reactions of different players on real-time and then use it against them.

Maybe, just maybe, it would make the patches useless because the game could "recover/cure" itself.

If some reaction or pro-action of the AI didn't workout right on a precise situation then the AI wouldn't risk it anymore, at least with that player.

Imagine the situation: "Arkimedes? humm, I know you! You're dead!" (AI thoughts).

Lets face it: the major objective of any game is to beat the gamer on their (game) own rules. So, on fair terms, it must make the player life miserable! :)

Arkimedes



There arent that many 'solutions' available to particular problems (taking CotN as an example) and combinatorics of more complex situations quickly gets out of hand (which is why AI for this is so difficult -- trying to determine which action in a chain of causality is the critical factors as well as measuring success).

To do any such system in a current game all the aspects are precanned
(set up specificly for the game mechanics) and allow very little learning capability. An AI to handle generalized solutions takes a magnitude more code and two or more magnitudes of effort to build the AI data set.


Good games are actually designed to be fun/interesting to the majority of players as well as challenging. Making a game frustrating/a misery usually spells its doom (only a minority of players are masochists). Its actually very easy to make a game 'harder'. Making it adapt to the player successfully is much more difficult.