Understanding AI in games - explanations for Players

[raystuttgart]

Hi guys,

this is kind of a reaction because we had lots of players trying to create AI concepts for our new features which they considered "extremely easy".
Thus they simply did not understand then when we modders told them "No, this is not going to work".

Thus I try to explain:

Do not try to create "AI concepts" by "thinking as a Human Player".
AI logic simply does not work like a "Human brain".

• Things your Human brain may understand in seconds may be almost impossible to teach AI and thus need years of modding effort.
• Things your Human brain could never understand can be calculated in seconds by AI and modded within a few hours.

Only the most experienced modders understand good enough how AI in Civ4Col works to even try to design and implement more complex AI logic.
(Only they really understand what is possible to teach it with reasonable effort or not.)

To simplify it for a Player:

• AI will rip easily rip apart a Human Player on the tactical level (perceiving current situations and massive calculations).
• But on the strategical level (proactive planning and predicting for future gameplay scenarios) AI is chanceless against a Human Player.

-----

Considering the features we implement:

Almost every feature we implement though has "strategic impact" (for future gameplay scenarios several turns later).
They can not be implemented as "purely reactive tactics" just checking current values.
If we programmed AI as "purely reactive on current data" we would completely ruin it ...

-----

To describe it in "Modders Language":
(It is necessary to understand the fundamental difference between "complicated" and "complex".)

• AI is extremely good in solving complicated problems. (Which Human brains can do very bad.)
  
• AI is extremely bad in solving complex problems. (Which Human brains excell at.)

Thus the "pirmary key principal" / the only real conceptual solution in current AI design for games is this:

• Transform "comlex problems" (which AI can not solve) into a "comlicated problem" (which AI can solve.)
• This basically comes down again to "problem analysis" and "solution design".
  
• And to do this efficiently you can use "proven in use design patterns" that match the problems.

 

[Nightinggale]

Summary: the goal of the AI isn't to create intelligence. It is to create the illusion of intelligence in order to create interesting gameplay.

 

[raystuttgart]

And wanted to use AI specially for strategic decision making - as unbiased (rely only on data)/ over human capabilities in intelligence.

It is not the problem "to rely on data".
AI needs to do that of course.

The question is simply: "Which data?"
(And of course different combinations.)

• current data
  
• stastically collected data
• new data refined from combinations of both

You can program different concepts:
(And of course different combinations.)

1. Purely reactive decision logic--> e.g. simple if / else or switch / case algorithms (decisions just on existing current data)
2. Proactiv trend decision logic --> e.g. weighting algorithms / complex rule sets based on a wider range of collected data (deriving statistic data for decisions)
3. Or alternative 1: Fake some true AI logic by simple repeated or even random actions using the feature and cheat in the background as compensation ...
4. Or alternative 2: Simply let AI not use the feature at all and cheat in the background as compenstation ...