Arguments why it is nearly impossible to program an "almost human" AI

[Ralf]

Its to long to be replicated here. Why not read it an comment. Applies to SMAC as well:
http://apolyton.net/forums/Forum6/HTML/001531.html

[Ned]

Ralf, I read the linked threads with interest. I think both you and Stuff2 have suggested viable approaches to improving the AI. You suggest adding templates to city growth letting the AI manage wars, diplomacy and intercity development. Stuff2 recommends a three level AI, the first plotting strategy, the second plotting "logistics," and the third plotting details.

I would simply like to add an observation I have mentioned here before: The human player normally takes a considerable amount of time to play his or her turn. The computer should be actively doing something during this period of time, such a plotting overall strategy, faction by faction. This could be Stuff2's Stategic AI. The Strategic AI's instructions should be followed when it is the AI's turn unless certain predefined conditions have suddenly altered the situation during the interim - such as war being declared or the polar ice caps melting.

Ditto, the second level AI should be given some of the dead time to process its orders to better carry the current strategy. For example, such an AI could organize invasions that wait until a critical mass is achieved - or actually build the necessary transports to get the invasion troups to the enemy.

So, unless something has changed, when it is the AI's turn to move, all it has to do is carry out the micromanagement end of things. This should be simple and fast.

This said, I would also like the "difficulty level" control to be adjustable to allocate a minimum "time" to each AI level. The game player would know that if he declares war, the AI may take two minutes to adjust its strategy before moving. The time it takes would vary accoding to difficulty level. This time could be used by the human player to take a well needed break.

Anyway, the thread was very interesting. I hope Firaxis gets the message that the AI needs to "think" better.