Computer teaches itself to play civilization V

[DaveMcW]

They provided the source code: http://groups.csail.mit.edu/rbg/code/civ/

 

[Camikaze]

Interestingly, civfanatics was referenced in the study (footnote 6).

 

[turboraton]

Next we know said AI joined the forums to complain about how cIV was better than CiV.

Edit: AND steam.

 

[Tubbilicious]

There was a project set up like this within CIV from the community with the AI that did the best after every game always being carried forward to the next game, thereby making the AI always stronger. The difference is with this that the AI actually read the rules of the game, interpreted those rules, and then played - it would be interesting to note if there was a playstyle the AI chose, and what factors led it to different playstyles.

 

[Narnia]

The source article never actually says it was Civilization V specifically. The image is of Civ V, but then you know what press people and bloggers are like with images...

I wonder if it wasn't the original Civilzation.

I think I read that it was civ 2

 

[CYZ]

This is all very interesting. It seems very possible to use selection to let an AI self-evolve.Possibly this is the way AI will be created in the future?

You could let an AI evolve to the point where you have an AI capable of reasonably dealing with different situations. Then let it evolve further by setting certain map/victory settings to create a different AI for every civ. For example: Let the Polynesia AI specialize on archipelago maps.

Once the Ais reach a reasonable playing level, without being too single focussed on one map typ/victory setting you could set it to stop evolving further and release it.

Once this technology is advanced further you could create AIs that adapt to a player during one (or more!) game. I wonder if this will be the future for AI coding.

 

[joncnunn]

If they don't need the source code, it would be interesting for them to run some tests on Civ V Duel.

Run 100 games each using Civ I, Civ II, Civ III, Civ IV, and Civ V's manuals.

If the Civ II manual outperforms the Civ V manual in Civ V, it could confirm issues with the Civ V manual.

 

[Thormodr]

If they don't need the source code, it would be interesting for them to run some tests on Civ V Duel.

Run 100 games each using Civ I, Civ II, Civ III, Civ IV, and Civ V's manuals.

If the Civ II manual outperforms the Civ V manual in Civ V, it could confirm issues with the Civ V manual.

Well, if the Civilization 5 AI had to rely on the Civilopedia, it might explain why it is so moronic. Lol.

Anyway, it would be interesting to see a comparison test done like that.