a.i.

[Moulton]

The AI definitely knows things it should not. I knew once I was mounting an invasion fleet off its coast... outside of is visible range. Maybe it steals warplays automatically... buat I never had a hint of espionage going on.
Last night I started a change in strategy with China, and gave her rite of passage. She has several sections separated my my land, and was continually trying to cross. I am twice her size and strength, and do not considere her a threat, and have no available land to colonze. Two turns later I saw two stacks marching.. 16 units of ombined armor each. then 2 more next turn.... She was at war with germany, and was heading to where Germany was launching an invasion--in an area wher China had no units or cities, and should have had no clue
Other wise, AI does mount new strategies from time to time, and that does make it more of a challenge.
Now, if I could just figure out how to launch Ironclads in one turn, with pop 6 cities....

 

[57%]

AI is just a determination by a programmer(s) to value the potential outcomes so the AI can choose the path of the greatest relative benefit. Chess is dramatically easier to express in logical comparisons (not that I'm saying it's easier to play, as there's a reason Kasparov and other masters are highly respected) -- there are only 6 types of pieces, and 64 locations on the board.

The "Chess AI" places a value on every piece, and another value depending on the placement upon the board. For example, a Queen is valued at 9, and a Queen in the center has a higher value than a Queen on the side. There is no random outcome to any move. These numerical comparisons have been around before AI, and are used in chess strategy books. The "AI" just took advantage of these existing easily-expressed mathematical comparisons, and allowed the ability to pursue any number of moves with likely counter-moves to the Nth degree. The use of existing "Chess Masters" helped refine the numerical values on certain pieces and their placement, in response to the evolving match. What Big Blue accomplished was allowing a team of programmers and masters to decide in advance how to react to certain board configurations.

Extrapolate to Civ3. Infinitely larger board (hundreds by hundreds), many units with varying functions, randomized outcomes, different government types, resource allocation, trading, diplomacy ... It should go without saying that effectively determining in advance the value of every potential situation is not going to be possible for the programmers. And in this way any human will learn what traps the AI will be lured into, again and again, so the human can always take advantage of the AI's Achilles' Heel.

 

[Smirk]

Originally posted by Mikeweather
That's what Garry Kasparov thought too. Deep Blue showed him otherwise

As many said its only a matter of time, chess is a relatively simple problem, the rules and possible layouts are finite. Huge compared to tic-tac-toe but on the same order, Civ adds much more complexity to the simulation of war (thats all chess is), more land, terrain, larger variety of units.

However, I don't think Civ is less tractable than chess. Overall there is more variablity in Civ, but locally there isn't, since most units only move 1 space a turn, you can very easily predict when those enemy troops will threaten your city. So you can in turn look at each city and determine what to do there based on local conditions. The AI just does not threaten your kingdom as a whole (unless its really small), if sea is involved anywhere its even worse, you can expect the AI to drop 1 unit on your separate landmass every 100 years or so.

Now compare chess where a queen in the center can threathen anything on the board in 1 turn.

The biggest problem in Civ AI is the crutch of cheating, I think if they set down at the beginning determined not to cheat it may motivate them to produce a better AI. I know it has for me on the occasions I have done so.

Cheating also creates an artifical opponent that is completely different from a human opponent. You can't send dummy transports, cause the AI knows they are empty. You can't move defense from one city to another, because the AI then attacks the weaker defended city. Those strategies can work against humans. And you see the same general strategies in chess also, without the fog of war.

Any kind of comparison to human intelligence will always be false until the cheating is removed. As soon as the AI does something based on information it can't possibly know it instantly becomes just another programmed routine. With the loss of immersion that causes.

Another unfortunate side-effect of this is once/if-ever multiplayer is released most of us will have to relearn how to play the game.

 

[spork]

The AI programming for the game has nothing to do with machine intelligence. It's just a deeply embedded series of conditional statements which encode certain decision procedures. How a good AI has nothing to do with a smart machine, but rather a programmer who is also a good player. It's sad, though in my opinion true, that serious game players (like the people who post to this forum) would probably be able to do more to improve the AI than the authors of the game. It's because we have mastered strategies. Strategies are nothing more than routines, plus some conditons about when the routines should and shouldn't be carried out. Once you break it down like this, it's not even that hard to program (although I'm not sure of the architecture of the Civ AI).

For example, for any given city, there is a reasonably simple strategy for what to do to maximize shield production. (Funny that the actual AI is absolutely terrible at this!) Of course, your stategies must differ if in the "big picture", your civ needs a bunch of workers and settlers. But these are just extra conditionals. A if a great player can clearly articulate his/her strategy, that strategy can be written as AI. Of course, at some point all the conditions become too messy and it's time to release the game.

Honestly, sometimes I see hints of meticulous work when I watch the AI. Other times I'm very disappointed (in retrospect; when I'm playing, I'm thankful that my enemies are being dumb). I'm sure there are a lot dumb things you've caught the AI doing.

It would also make a fun survey, or maybe later a poll: "what are the dumbest habbits of AI states?"

I'm not sure what my own answer would be, but I know of several great candidates.

 

[zebomba]

The dumbest things are:
-landgrabbing, a human 'd rarely do that. I don't mean the normal expansion, but rather building cities deep inside enemy turf. Cities in the mid of the tundra or desert, lol. Ok, I make outposts, but I make it act and look like an outpost. No cathedral inside it, of course.
-building queue, as the few times I could see what was going on inside their cities, I only spotted dumb things. Size 12 cities without temples. Heh. Half of the pop was entertainers.
-the AI also never gets the picture unless you're INSIDE its borders. You can stack troops at ease near it, as it never swaps to war counter measures.
-no mass attacks unless they're much bigger than you. What's the point of sending a pair of units at a time? Well, maybe they never do that against me as I strike first.
-wonders are placed randomly, and they build it even when beaten down to 2-3 cities. And after they lose their capital, it seems to be moved randomly too.
-city placement. I like to raze cities not only bc they are far, but even the close ones are ill placed. Cmon, 1 tile to the left and you can have 3 resources at the same time. The worst thing I saw was when they founded a new city that could NOT get iron and coal inside its radius, despite it was possible. Tsc.

And the list can go on and on...