That Cheating AI

[apatheist]

Zurai, ok, I buy that. Good argument.

 

[deo]

A human player knows that it will need some tundra to have a chance to get some oil or uranium in the late game.

Well why not telling the AI that it should settle or conquer some tundra, desert etc. to get oil for later?

The problem i think is that making a very good AI requires much much time so that the developers can find out how to make an AI very inteligent, they must think very much to find ways how to make the AI understand that "in that tundra is PREHAPS a oil recource" without negative effects like the AI settling only in tundra and desert.
So making a very good AI is not imposible but the developers must be very clever, there must be many that programe it, and there must be much much testing to find bugs.

P.S.

In world maps through i think the AI should know the map and resources etc. because we know it too

 

[HourlyDaily]

In my humble opinion, as a person who has studied AI in university, most developers spend far too less time developing the AI when building a video game. Most of the time they rely on such workarounds as allowing the computer to 'cheat', which in some cases is REQUIRED (AI programming *is* hard after all), but it shouldn't be required on the scale we see it in Civ3, for example. I really wish developers spent more time developing AI routines, because it would make gameplay a lot more realistic and fun.

I don't even think that most games employ a decision-tree style AI, which in my opinion would do wonders for a game such as Civilization. The downfall is that it would take a lot of time to get it just right, but this is exactly my complaint - too much time is spent on graphics, and too little on gameplay.

I think the development cycle has probably been long enough. People might say "AOEIII, great graphics and fun, but I'd rather play Civ4 - more to do and that AI is really intuitive." In the long run the replay value is higher and more people will come back for expansions and Civ5.

I believe the AI will take a great step-up, but it would make Civ4 impassable as the greatest game yet made if the AI didn't cheat (Regent or Monarch whichever it is on Civ3) but plays and reacts with long term views.

 

[The Adjudicator]

In my humble opinion, as a person who has studied AI in university, most developers spend far too less time developing the AI when building a video game. The downfall is that it would take a lot of time to get it just right, but this is exactly my complaint - too much time is spent on graphics, and too little on gameplay.

Hence the genius of allowing the legions of fanatics to reprogram the AI in CIV 4! Give a few thousand geeks like us a few months and we will have her all worked out.

 

[warpus]

Hence the genius of allowing the legions of fanatics to reprogram the AI in CIV 4! Give a few thousand geeks like us a few months and we will have her all worked out.

See, scripted actions do not equal AI, although that is what people have been calling it. There is no AI in Civ3 at all, but we all call i that anyway What I would LOVE to see is a REAL AI that learns on its own, tries out random things, builds a decision-tree in the background, and learns from its mistakes. Currently we have nothing of the sort; the computer simply uses a lookup table and perhaps some randomization to 'decide' what to do.

Although it would be (probably) possible to customize Civ4 to include real AI, I doubt anybody is actually going to take the time to do it. Fake AI is often good enough for the casual player, and most games rely on it.

 

[warpstorm]

I think a decision tree is the wrong approach for this game unless it is done for only the highest level goals as it would take up way too much memory. Having said that there are other ways to model goals.

 

[oldStatesman]

See, scripted actions do not equal AI, although that is what people have been calling it. There is no AI in Civ3 at all, but we all call i that anyway What I would LOVE to see is a REAL AI that learns on its own, tries out random things, builds a decision-tree in the background, and learns from its mistakes. Currently we have nothing of the sort; the computer simply uses a lookup table and perhaps some randomization to 'decide' what to do.

Although it would be (probably) possible to customize Civ4 to include real AI, I doubt anybody is actually going to take the time to do it. Fake AI is often good enough for the casual player, and most games rely on it.

Wow. If you know where a 'real' AI is, please let me know...it has been the Holy Grail of the Computer field since at least, oh, forever!

 

[warpus]

I think a decision tree is the wrong approach for this game unless it is done for only the highest level goals as it would take up way too much memory. Having said that there are other ways to model goals.

It wouldn't take up too much memory at all, but I agree that computing this tree in real-time would be bothersome, if that is what you meant. The best solution, and one that would work quite well imo, would be for Firaxis to build a decision tree in real-time during development, refining it along the way, and allowing the AI learn how to play most effectively. The game could then be shipped with a variety of pre-compiled decision trees for various difficulties and perhaps even level types (archip, continents, etc.). You could even allow the AI to refine the decision tree slightly as the user plays the game, to better adapt to his/her playing style. This shouldn't take up too many CPU cycles and/or memory, if you do it right.

Such an approach would give us computer opponents who play a lot more like humans, which would increase the replay value of the game, not to mention that it would be a lot more fun. I think most seasoned CIV3 players would agree that it is easy to predict what a computer is going to do based on a given situation and circumstances. This is something that a true AI that learns would avoid, giving us a more unpredictable (and human) opponent.

Usually it is only a matter of time until you get used to the "AI" and figure out its weaknesses. My approach would remove that aspect from the game. It would involve a lot more work for the developer, of course, but in my opinion it is something a lot of games seriously could use. I have been turned off by a lot of games because of really ****ty (so-called) AI.. Civ has never been one of those games but I think you will agree that it is sort of predictable.

 

[warpus]

Wow. If you know where a 'real' AI is, please let me know...it has been the Holy Grail of the Computer field since at least, oh, forever!

Hmmm.. I am only going from memory here, but I think that 'Creatures' used true AI instead of pseudo AI.. it could have used a mix of both, though. actually, it probably did

If you have studied AI, you will know that a lot of problems are "unsolvable". Chess isn't, but it is very simple compared to something as complex as Civ. To build a true AI system for civilization would be impossible, but to include some true AI is a quite realistic expectation, imo. You will obviously have to script some things, but you could allow the computer to truly learn certain other details.

Perphaps they are doing this, who knows.

 

[warpstorm]

And what is the short list of independent and conditional attributes you'd use to generate the tree?

 

[warpus]

And what is the short list of independent and conditional attributes you'd use to generate the tree?

That question could only be answered after many days of development and testing.

 

[warpstorm]

So, you aren't willing to make a stab at it? I think there would be an awful lot of things. For example the local terrain (oh, wait that's probably more than one attribute per city), each of their opponents and how they feel about me and how I feel about them and how Ithink they feel about each other, what turn it is, what technologies I think everyone has, what religions are dominant in each nation, what turn it is, how close I think everyone is to winning by each victory condition, the relative populations, wealth, and military strentghs as I perceive them, who has which resources. This is just a sampling of the types of independent attributes that I'd start with. A lot more would turn up quickly if I spent more than 30 seconds on it. THis doesn't get into the conditionals and what rules could be extracted from them.

 

[Leprechaune]

So, you aren't willing to make a stab at it? I think there would be an awful lot of things. For example the local terrain (oh, wait that's probably more than one attribute per city), each of their opponents and how they feel about me and how I feel about them and how Ithink they feel about each other, what turn it is, what technologies I think everyone has, what religions are dominant in each nation, what turn it is, how close I think everyone is to winning by each victory condition, the relative populations, wealth, and military strentghs as I perceive them, who has which resources. This is just a sampling of the types of independent attributes that I'd start with. A lot more would turn up quickly if I spent more than 30 seconds on it. THis doesn't get into the conditionals and what rules could be extracted from them.

Hehehehe... This is somehow familliar.

 

[Gato Loco]

Remember, only a minority of players will spend enough time with a game to get good enough that an improved AI will be an issue. Of the remainder, many will only play MP. Thus AI isn't a high priority for most games. Civ is somewhat of an exception, but you still have to keep in mind that there's no reason to spend lots of time and money on a feature that most customers won't ever use.

 

[Crazy Jerome]

"See, scripted actions do not equal AI, although that is what people have been calling it. "

Warpus, are you aware that the level of control being touted by Firaxis for mods would seem to indicate the ability to do a lot more than scripting? Now exactly what can be down with that is in dispute. I seem to recall Leprechaune and I arguing quite heatedly about it. But there is very good indications at this point that the various competing theories will have a chance to be tested, if enough people are so inclined.

 

[Kyriakos]

Hm, does the AI have reseach/production bonuses in monarch level? because i am really sick of how it ends up getting important techs unexpectadly, and just when i am about to get my second unique tech + a scientific leader In the end i resolve to get involved in at least one pre-medieval age war, so as to expand more rapidly and have a chance of getting a scientific leader...
Btw is there any reason to use him so that he can generate more scientific growth? i have always used him to build wonders.