Brad Wardell's views about technology, politics, religion, world affairs, and all sorts of politically incorrect topics.
It's never done
Published on January 14, 2007 By Draginol In GalCiv Journals

I play and I play and I play.

And every tiem I play, I still see stupid AI stuff. Which is bound to happen because, realistically, unless one hard codes it, the AI will make mistakes.  The challenge is always one of making sure that the AI is considering as many variables as possible and then tune it over and over and over and over.

Tonight's game I won easily.  Here were the results:

Technologies researched (and in order):

Player

Drengin Empire

Thalan Empire

  1. Xeno Research
  2. Planet Imps
  3. Xeno Economics
  4. Uni Trans
  5. Xeno Ent
  6. New Prop
  7. Ion Drive
  8. Impulse Drive
  9. Galactic Warfare
  10. Space Milt
  11. Space Weapons
  12. Mass Driver
  13. MiniBalls
  14. MB 2
  15. Mass Drivers
  16. MD 2
  17. ADv Computing
  18. Basic Mini
  19. Xeno Ind Theory
  20. Xeno Factory Const
  21. Research Centers
  22. Soil Enh
  23. Xeno Ent
  24. BAsic Logistics
  25. Planet Invasion
  26. Planet Bombard
  27. Space Mining
  28. Adv Hulls
  29. Medium Scale
  30. Gen Life SUp
  31. Enh Logistics
  32. Xeno Trade Centers
  33. Research Academies
  34. Stellar Marines
  35. Adv Troop Mod
  36. Shock Troops
  37. Dip Rel
  38. Int Gov
  39. Trade
  40. Impulse II
  41. Impulse III

Influence Victory!

 

  1. Xeno Research
  2. Xeno Com
  3. Planet Imps
  4. Uni Trans
  5. New Prop
  6. Soil Enh
  7. Ion Drive
  8. Diplomatic Rel
  9. Space Milt
  10. Xeno Eco
  11. Starship Def
  12. Space Weapons
  13. Impulse Drive
  14. Missile Weapon Theory
  15. Trade
  16. Shield Def
  17. Def
  18. Adv Computing
  19. Adv Trade
  20. Basic Mini
  21. Enh Mini
  22. Master Trade
  23. Stinger
  24. Xeno Ind
  25. Stinger II
  26. Xeno Fact
  27. Man Centers
  28. Impulse II
  29. Impulse III
  30. Armor Theory
  31. Warp Drive
  32. Fusion Power Plants
  33. Titanum Armor
  34. Titanium II
  35. Adv Mini
  36. Basic Logistics
  37. Planet Invasion
  38. Planet Defense
  39. Anti-Matter Power
  40. Space Mining
  41. Research Centers
  42. Xeno Ent
  43. Adv Hulls
  44. Enh Logistics
  45. Medium Scale
  46. Xeno Trade Centers
  47. Xeno Bank
  1. Xeno Research
  2. Xeno Com
  3. Planet Imps
  4. New Prop
  5. Uni Trans
  6. Ion Drive
  7. Space Weapons
  8. Xeno Factory Const
  9. Missile Weapon Theory
  10. Impulse Drive
  11. Stinger
  12. Stinger II
  13. Fusion Power Plants
  14. Basic Logistics
  15. Space Mining
  16. Soil Enh
  17. Stinger III
  18. Adv Computing
  19. Basic Mini
  20. Impulse II
  21. Dip Relations
  22. Impulse III
  23. Trade
  24. STarship defs
  25. Adv Trade
  26. Armor Theory
  27. Titanium Armor
  28. Enh Mini
  29. Titanium II
  30. Titanium III
  31. Adv Mini
  32. Duranthium
  33. Warp Drive
  34. Duranthiuu II
  35. Duranthium III
  36. General Life SUpport
  37. Xeno Biology
  38. Space Mining 2
  39. Enh Logistics
  40. INt Government
  41. Master Trade
  42. Planetary Invasion
  43. Habitat Imp
  44. Research Centers
  45. Adv Hulls
  46. Medium Scale
  47. Enh Tri
  48. Superior Hulls

 

 

Wars:

First War: Player (Terran Alliance) vs. Torians

vs.

Result: Torians destroyed

Second War: Player (Terran Alliance) vs. Altarians

vs.

Result: Altarians destroyed

Other Ships:

Thalans:

Attack: 4 Defense: 4

Attack 2: Defense 2:

Player ships: 3 attack, 0 defense.

Changes made, next game...


Comments
on Jan 14, 2007

Another game:

Player

Drengin Empire

Thalan Empire

  1. Xeno Research
  2. Planet Imps
  3. Xeno Eco
  4. Uni Trans
  5. Xeno Entertainment
  6. Xeno Bio
  7. Extreme Col
  8. Barren World TEch
  9. Research Centers
  10. Adv Barren World
  11. Galactic Warfare
  12. Space Milt
  13. Space Weapons
  14. Xeno Ind
  15. Mass Driver Theory
  16. MiniBalls
  17. MB2
  18. MD1
  19. MD2
  20. MD3
  21. Xeno Medice (for Plague)
  22. Plague Cure
  23. New Prop
  24. Ion
  25. Impulse
  26. Adv Computing
  27. Planet Invasion
  28. Planet Bombardment
  29. Stellar Marines
  30. Shock Troops
  31. Basic Logistics
  32. Enh Logistics
  33. Xeno Factories
  34. Space Mining
  35. General Life Support
  36. Singularity Drivers
  37. Sing2
  38. ENhc Mini
  39. adv Hulls
  40. Medium Scale
  41. Impulse II
  42. Impulse III
  43. Warp Drive
  1. Xeno Research
  2. New Props
  3. Space Milt
  4. Starship Def
  5. Space Weapons
  6. Xeno Biology
  7. Extreme Col
  8. Barren World Tech
  9. Adv Barren
  10. Mass Driver Theory
  11. Planet Imps
  12. MiniBalls
  13. Planet Invasion
  14. Soil Enh
  15. Xeno Medicine (for Plague)
  16. Xeno Economics
  17. Xeno Industrial Theory
  18. Plague Cure
  19. Ion Drive
  20. Fusion power
  21. MB2
  22. Research Centers
  23. Xeno Factories
  24. Armor Theory
  25. Mass Drivers
  26. Anti-Matter Power
  27. Titanium Armor
  28. MD2
  29. MD3
  30. Singularity Driver

Kill by Player

  1. Xeno Research
  2. New Props
  3. Planet Imps
  4. Ion Drive
  5. Space Weapons
  6. Mass Driver Theory
  7. Mini  Balls
  8. MB 2
  9. Mass Drivers
  10. Xeno Com
  11. Impulse
  12. Uni Trans
  13. Xeno Bio
  14. Extreme Col
  15. Barren World Tech
  16. Aquait World Tech
  17. Fusion Power
  18. Anti-Matter
  19. Research Centers
  20. Xeno Medicine
  21. Plague Cure

Killed by Plague

 

 

Notes: Player beat Drengin to high quality Barren world by 1 turn.

Wars:

Player vs. Drengin

vs.

on Jan 14, 2007

And another game where the Drengin won:

Player

Drengin Empire

Thalan Empire

  1. Xeno Research
  2. Planet Imps
  3. Xeno Eco
  4. Xeno Ind
  5. Uni Trans
  6. Xeno Ent
  7. New Prop
  8. Ion Drive
  9. Impulse Drive
  10. Research Centers
  11. Adv Computing
  12. Basic Mini
  13. Gal Warfare
  14. Space Mil
  15. Space Weapons
  16. MD Theory
  17. MiniBalls
  18. MB2
  19. MD
  20. MD2
  21. MD3
  22. Planet Invasion
  23. Basic Log
  24. Space Minig
  25. Soil Enh
  26. Enh Logistics
  27. Xeno Factory Const
  28. Diplomatic Rel
  29. Adv Hulls
  30. Medium Scale Building
  31. Sing Driv
  32. Sing 2
  33. Enh Mini
  34. ADv Logistics
  35. Man Centers
  36. Planet Bombardment
  37. Rsearch Aacad
  38. Stellar Marines
  39. Fusion Power Plants

Defeated by Drengin Empire!

  1. Xeno Research
  2. Planet Imps
  3. Xeno Com
  4. Uni Trans
  5. Space Milt
  6. Space Weapons
  7. Soil Enh
  8. Dip Rel
  9. Xeno Eco
  10. Xeno Bio
  11. Xeno INd
  12. MD Theory
  13. MiniBalls
  14. Xeno MEdicine
  15. Extreme Col
  16. Barren World
  17. Heavy Gravity
  18. MB2
  19. Fusion Power
  20. Research centers
  21. Soil Enh
  22. Habitat
  23. Planet Invasion
  24. Planet Defense
  25. Int Gov
  26. Research Acad
  27. Alliances
  28. Basic Log
  29. Int Republic
  30. Mass Drivers
  31. Space Mining
  32. MD2
  33. Enh Logistics
  34. MD3
  35. ADv Hulls
  36. Sing Driver
  37. Xeno Factory const
  38. Medium Scale Building
  39. New Prop
  40. Ion
  41. Impulse
  42. Invention Matrix
  43. Planetary Bombardment
  44. Tidal Disrupt
  45. Impulse II
  46. Adv Computing
  47. Titanium Armor
  48. Sing 3
  49. Stellar Marines
  50. Basic Mini
  51. Enh Mini
  52. Sing IV
  53. Tit2
  54. Adv Troop
  55. ADv Logistics
  56. Graviton Driver
  57. Shock Troops
  58. Adv Planet defense
  59. Graviton 2
  60. Graviton 3
  61. Quantum Driver
  62. Quantum Driver 2
  63. Advanced Mini
  64. Impulse Drive II
  65. Warp Drive
  66. Space Mining 2
  67. Quantum 3
  68. HD Spike
  69. Superhior Hulls
  70. Spike Driver 2
  71. Expert Logistics
  72. HD Spiek Driver
  73. HD 3
  74. HD 4
  1. Xeno Research
  2. New Props
  3. Xeno Com
  4. Starship Def
  5. Uni Trans
  6. Planet Imps
  7. Ion Drive
  8. Gal Warfare
  9. Fusion Power
  10. Impulse Drive
  11. Dip Relations
  12. Xeno Factory
  13. Adv Computing
  14. Basic Mini
  15. Research Centers
  16. Armor Theory
  17. Xeno Bio
  18. Titanium
  19. Tit2
  20. Basic Log
  21. Space Mining
  22. Tit3
  23. Enh Mini
  24. Gen Life Support
  25. Adv Mini
  26. Duranthium
  27. Impulse III

Bankrupted selves

 by too many starbases

Fixed

Surrended to Drengin

 

 

War: Drengin vs. Player (Terran Alliance)

vs. 

Later in war...

vs.

vs.

Terrans:
Frigate: 16 Attack 0 defense
Fighter: 8 attack 0 defense

Drengin:
Frigate:  Attack: 10 / Defense: 0
Fighter: 12 attack 4 defense (new tech)

on Jan 14, 2007
Does the Metaverse collect this kind of data?

Or on a half-wild tangent, is there any chance of extracting some value from the epic generator work to build some Metaverse data collection that cuold help devs, esp. re long game vs. short game stuff?
on Jan 15, 2007
Draginol, if it was for me, you should hard code only one change to AI behaviour: the AI *MUST* send troop transports only with escorts AND the escorts have to be strong. The strong escort could be defined as follows: one class less than your max researched ship class and a 1 to 1 ratio between combined ship attack and combined defence.

If the AI can't come up with strong escorts it shouldn't bother sending the troops because in 85% of the cases those troops are intercepted by some tiny fighter of mine which I always keep in reach of my planets. The other benefit for the AI of this hard rule is that those 500M or 1B people keep working and producing revenue on their home planet instead of being wasted.

IMHO this would improve overall AI performance due to less troop losses and consequently higher population. IIRC most games by human payers are ultimately won because of higher population.
on Jan 15, 2007
These kinds of dev journals are always very interesting to me. I appreciate you taking the time to post them up, Brad.
on Jan 15, 2007
I see that the AI loves those Superior Duranthium defenses... One data which could be interesting in those report is if the winner of each game was the one with the best research after the colonisation phase. As far as I see it, maybe the Drengin won only through superior research.
Thinking about it, almost all of those AI ships have better tech than your.
on Jan 15, 2007
Dang, those ships look sweet.

Maybe I should reconsider my play style -- I never build a ship with 1 movement unless it's for pure defense. They just don't get to the battle otherwise, and they can't form into fleets. This goes way way back to Civilization where chariots were always better than catapults because pulling a one-movement unit up to a city was far too slow.

I wonder if there is some kind of AI theorist you could hire. The process of "Play -- note problem -- fix -- note other problem --" reminds me of spending a night on a plastic extrusion line going "raise roll -- note problem -- lower roll -- note problem -- raise roll", and later on learning a bit of plastic theory that says if you don't have enough plastic coming from upstream, no amount of raising and lowering the roll can help. Maybe there is an AI theory that lets you produce it and test it faster than playing a whole game. Or -- wish this were possible -- that you could give the AI parameters and give 1000 beta testers different AI parameters and let them report which ones gave the best game. I guess you already do this when you let the AI play against each other.
on Jan 16, 2007
And along those lines, could you do smoke testing? For instance, load up the same Universe with the same "human" player that is scripted and see how the AI reacts to it over time and based on various tweaks? You'd have to be careful since you don't want the AI to only be good at defeating players in one map and scenario, but it would show the level of randomness and luck for various outcomes and also show (over time) the effect of your AI modifications.

Also, is is really fair to say the AI isn't scripted? For instance, if you prohibit the AI from being "stupid" and disallow them from doing something, isn't the just the inverse of telling them to always do X, then Y, then Z, just not chronological? Don't get me wrong, I think having an AI that weighs options, thinks toward the future, and then re-calculates decisions is awesome, but preventing mistakes seems a bit scripted. Just a thought.

I'm impressed with your ability to play game after game after game. You must truly love it or feel intensely obligated to try it out. Hopefully more the former than the later. That is my only major gripe with Gal Civ 2- it's sooooooo long to play. I doubt the expansion will fix that, but maybe some future thought toward this would be good.
on Jan 16, 2007
Would the GC2 engine support AI testing via Monte Carlo analysis?

I suppose the AI code would have to take it's constants and initial variable values as input. ...and the engine would have to allow for play without a human player.

After several thousand automated games, you would get some good tuning numbers.

The benefit would be AIs tuned to different galaxy sizes, rule sets, number of players, intelligence levels, etc. In a sense, the AI could select a set of constants at the beginning of the game based on the environment.

The downside is that only human play testing can tell you what variables are missing and should be added to the AI.

Just a thought. I realize these kinds of design elements are probably not in there now and it's far too late in the dev cycle to make them. Maybe this idea could reduce your testing labor in GC3!
on Jan 16, 2007
It would be interesting to have the AI benchmarked. Like "It used to defeat this scripted human player on turn 400, now it takes till turn 300 but it loses 35% of the time."
on Jan 16, 2007

We do that too actually.

What we do is have the AIs all play each other. Since each one has its own personality, we can see what % of the time personality X beats personality Y.

on Jan 17, 2007
Thus prompting your next post I see. Sweet. Looks like the Drengin AI is pretty tough, despite mistakes.