|Civilization Five and Two have|
some... fundamental difference
Computer AI in strategy games has been proven to be... well stupid. It usually is way too much of an undertaking to create anything that really classifies as artificial intelligence. Instead the game developers give the computer an unfair advantage by letting it cheat. Maybe supplying it with more resources for example. Racing games like Need For Speed are infamous for incorporating this into their games. Called "rubber band" AI: the computer simply speeds up to catch up with you, and offer more of a challenge.
In this study the researchers incorporated the CIV rule book into a search framework, and in this case, also had the AI learn strategies through playing. The Monte Carlo search framework has been used in board games before, and considering Civilization is essentially a high functioning board game, it makes a good choice for this study. It is important to note that the instruction books for Civilization games are notoriously known for being massive.
|The Guidebook for CIV II, not the box.|
Due to the massive size of the civilization books I wouldn't expect this sort of AI to show in in every video game just yet. But as a big fan of strategy video games I hope I start encountering smarter AI soon!
I really am interested in the content of this academic article. If there is anyone out there who knows about this stuff I would love to talk to you! For everyone else who is interested, you can read, the academic text here. Just in case I also have cited the article at the bottom of this post. The project website can be found here.
Branavan, S.R.K., David Silver, and Regina Barzilay. "Learning to Win by Reading Manuals in a Monte-Carlo Framework." (2011). MIT.edu. MIT. Web. 5 Aug. 2011. <http://people.csail.mit.edu/branavan/papers/acl2011.pdf>.