1. We have added a Gift Upgrades feature that allows you to gift an account upgrade to another member, just in time for the holiday season. You can see the gift option when going to the Account Upgrades screen, or on any user profile screen.
    Dismiss Notice

Why AI is so weak?

Discussion in 'Civ3 - General Discussions' started by akots, Aug 21, 2003.

  1. The Last Conformist

    The Last Conformist Irresistibly Attractive

    Joined:
    Aug 25, 2001
    Messages:
    27,779
    Location:
    Not on your side
    Well, in the mods I've played, the AI has carried out transoceanic attacks in the same piecemeal fashion as usual. In what circumstances would not applying concentration in this circumstance of force be preferable?

    I might add I've yet to see a modder complain about the AI launching concentrated attacks on land, yet it does that routinely.
     
  2. trumpeteer

    trumpeteer Player of Trumpet

    Joined:
    May 16, 2003
    Messages:
    449
    Location:
    Hell
    The blunt truth is that the AI sucks at sea-based attacks and all aspects of sea-centered warfare.
     
  3. Xerol

    Xerol Emperor

    Joined:
    Oct 6, 2002
    Messages:
    1,542
    Location:
    In an IDE.
    One other thing I've seen the AI not do: Pick a "strategy". Usually it just goes about it's way, doing a "Build, Conquer, Culture, Diplo" thing, spreading everything about fairly equally without concentrating on a certain kind of victory(but there is some specialisation: militaristic civs tend to attack more. other than that, I haven't noticed much concentration)

    One of the more difficult aspects of this would be giving the AI the chance to "choose its fate" by itself, i.e. analysing the current situation and deciding on which would bring about the best victory, and as an added effect, do it in the way the typical civ player would: what gets the highest score. A few things that would need to be changed before a "targeted" AI could be programmed:

    -Before Intelligence Agency, a civ would be informed if a city was investigated. By putting a spy in the civ, there would still be a 10% chance of "informing" the civ that they have been investigated. This would help the human player tell if the AI is checking up on culture or whatever.
    -

    Now, the AI should wait until it knows all the civs, and a certain date afterwards, before "deciding" on a victory target, and it should have a backup plan in place in case it doesn't work. If the AI is fairly ahead in tech, it might go after a space race victory. If it's got a whopping military, conquest or domination should be the goal. If it's got a lot of wonders, or the scientific/religious traits, Culture should be an option(20k for wonders, 100k for traits) And it should try to go for Diplo once in a while, if it's not scoring high enough in the other categories. And if it does try for diplo, it should take on the same goals that a human player going for diplo would: Keep at peace if necessary, but if war comes about, well, "Dead people can't vote".

    And, if it seems to be coming up even on things(i.e. high culture + big military), the ol' coin flip should go on: just randomly pick out of the best options.

    Now, once it's chosen a specialisation, it should work towards that goal:
    Culture 20k: Research culture-giving techs first, i.e. Education, Literature, Monotheism, Construction. Then as soon as the techs are completed, get to work building those buildings.
    Culture 100k: Wonder-giving techs are the goal here, get to 'em first and on higher levels, perhaps a prebuild.
    With both culture victories, gov't should be Demo or Republic. If a religious civ, perhaps swapping Monarchy in during wartime.
    Conquest: Military techs, military strength, strong economy for upgrades, Leo's workshop if it can get it. Workers to connect resources, targeted wars to get unavailable resources/lux's, big-time warfare other times, and investigating cities before the battles begin to scout-out the border cities.
    Domination: Same thing as conquest, except keeping more cities and building the lowest-cost culture building in captured cities to grow territory.
    For Conquest/Domination, Communism or Monarchy should be the Gov't of choice.
    Space Race: Put lotsa money into techs to get a huge lead, build Libraries/Universties/Research Labs everywhere and target Science-increasing wonders and try to get a lot of them in one city, especially if that city has the colossus(for the early game). Then, a prebuild for Apollo if it's a higher difficulty level(monarch or higher) and perhaps a few prebuilds for other spaceship parts right before the appropriate techs get researched. Small wars to gain aluminum/uranium, but NO RESOURCE CHEATING, i.e. it wouldn't go to war over the resources until they became available. Gov't would be Rep/Demo.
    And finally, Diplo. This is kind of the "catch-all" category when it doesn't think it can pull up any other victory, or if its desired victory is impossible/largely impractical. A few major changes for this:
    -THE AI WOULD GIVE GIFTS. To raise its status, of course. And not just to the human player(s), but also to other AIs.
    -The "Dead man can't vote" approach to war.
    Gov't would suit whatever current strategy it's using.

    And if it can't get ANY of those victories realistically, the ol' histograph. Try to survive to 2050, build up happy citizens and territory, future tech if it gets that far(although impact would be minimum, it does bring the score up some)

    Well, that's my thoughts for today. Gotta let the keyboard cool off for a while now.
     
  4. Smidlee

    Smidlee Deity

    Joined:
    Jul 10, 2003
    Messages:
    3,348
    sorry but so far man hasnt been able to program an AI that does all that . Strategy (even in chess ) is any AI main weakness. it may seem to add this is easy but this approach can leads the AI to play worst and leave more exploits to destroy the AI. Computers are great storing numbers and calculating numbers but computers cant think .So everytime you have the AI to do one thing well it will leave the door wide open in another area . AI's just isn't as flexable as a human players are .
     
  5. eg577

    eg577 Warlord

    Joined:
    Jun 1, 2002
    Messages:
    211
    You don't need a thinking computer to make a good civ AI. A competent AI could easily be made with purely hardcoded knowledge, especially if we are willing to give it outrageous shield/food/support advantages like there already are in diety.

    Like myself and QwertySoft have stated, if the AI would (even if done completely by random) choose a specific strategy to focus on (i.e., win by culture, win by conquest) it would be very simple to tell the computer what to do and would also dramatically improve how well the AI could compete. This is largely because if the AI had a specific goal in mind it wouldn't have to resort to calls to the random number generator when the it decides what to build next. What's more, randomly choosing a specific method to win and sticking to it is more or less the way everyone plays.
     
  6. SolarFlare

    SolarFlare Privateer at large

    Joined:
    Jul 27, 2003
    Messages:
    370
    Location:
    Philad'a
    But the AI's goal is to oppose you, not to beat you. It doesn't know what kind of victory you're going for, so it can't counter. On the flip side, if only one victory condition is set, the AI should act differently (and I'm not sure if it does or not, frankly I don't care, this is just hypothetical anyway), knowing that to be the goal.
     
  7. akots

    akots Poet

    Joined:
    Aug 5, 2003
    Messages:
    3,981
    Location:
    Moscow-Houston, Russia-USA
    Just for comparison, I have recently tried the Civ2 gotm. Surprizingly, people still play Civ2. It was mazing to me that actually, the Civ2 AI is stronger in coordinated attacks that Civ3 AI. Especially the barbs landing from the ships. Certainly, in Civ2 corruption is less of a factor and AI has more possibilities to biuld the military. However, coordination and massive attack forces is still a strong point of Civ2 AI on high levels (King and above). Well, something you gain, something you lose. Though I like the Civ2 idea of spreading corruption towards commerce not shields.
     
  8. The Last Conformist

    The Last Conformist Irresistibly Attractive

    Joined:
    Aug 25, 2001
    Messages:
    27,779
    Location:
    Not on your side
    akots: I must say that my experience - and I played and beat Civ2 on all levels - is pretty much the opposite. Not only did the Civ2 AI suck even worse at launching transoceanic attacks than the Civ3 does, it was also very poor at concentrating its forces in land combat.

    Of course, concentration of force was harder to do properly in Civ2, since SODs were suicidal. But the AI rarely seemed to even try - units advanced one and one, without any attempt to cover or support one another, and never if ever would the AI hold back attackers to amass sufficient strength for a decisive attack.
     
  9. Xerol

    Xerol Emperor

    Joined:
    Oct 6, 2002
    Messages:
    1,542
    Location:
    In an IDE.
    Well, even if you did turn off all victory conditions but one, say Conquest, there's still many ways to go about it, and there's general strategies that should be followed regardless of what victory is in mind, i.e. filling in territory gaps between recently captured cities with temples, keeping defensive units in most cities, and other things, most of which the AI seems to be doing a good job with right now. However, I think we should get our hopes up for a focused AI in Civ4, mainly because by then most people's systems will be fairly good and the in-between turn wait would be too short to make huge maps with 15 rivals fun. So, make a more complex AI that spends a lot of time to give everyone time to read, write, listen to music, eat, drink, etc., bascially everything that everyone does now during the between-turn waits. It's one of the fixtures of Civ that makes it a great game.
     

Share This Page