AI Cheating -> Result bored of game already

[Saurus]

Originally posted by Falconne
This is the same problem I have with Civ.

I've got to try Galacatic Civilizations sometime, which people say has an excellent AI. But I'm put off by the space/sci fi theme; picking off randomly scattered planets to settle is not as fun as looking for stretegic terrain on a single planet.

Do that. It's an exellent game + the A.I really is fairly competent.

[Solver]

The CIv4 AI code will be available as a part of the SDK... but that will mean that a good number of people can work on it but not that revolutionary methods might be used to create a self-aware AI .

[Son of David]

If you've played for a couple weeks and mastered the medium difficulty levels, I envy you!

I'm too busy having fun with this game to take climbing the difficulty levels seriously. Prince is enough for me. *If* I get bored with that I'll move up.

I won't move up in difficulty just because people on Apolyton are boasting about their Emperor- or Deity-beating strategies! (Actually I really hate Monarch difficulty, it's NASTY.)

[Enigma_Nova]

As soon as anyone makes a self-aware AI, Firaxis will implement it in a way that requires someone to make a Memory fix to make it playable. Isabella will then be the best leader, but you will lose when your iron spawns down on the ice on a non-mirror map.

[RGBVideo]

Edit: scrap that, I hadn't read this thread. A big for Firaxis.

[QuantumPion]

To you programmer-types, how hard would it be to create an algorithm that just watches how a human plays, and then after playing hundreds and hundreds of games, figure out what works and what doesn't?

As for the AI cheating making the game unplayable, I say "meh". When I see a nearly defeated country spamming free, but obsolete units in futile attacks against my stack-of-doom, I say "meh". When I see a stack of 10 axemen coming straight for my capitol 5 turns into the game, I start over.

[MasterDave]

Originally posted by QuantumPion
To you programmer-types, how hard would it be to create an algorithm that just watches how a human plays, and then after playing hundreds and hundreds of games, figure out what works and what doesn't?

On moving from Prince level to Monarch level I have noticed that now the AI tends to cover up its first worker with a warrior or archer (usually archer) very early in the game. One of my favorite techniques for getting a leg up early on in Prince games was to steal workers from my neighbors with my first warrior as he wanders around scouting the map. If you do snag an AI's worker upon first contact, it gives you a huuuge edge, and even if that worker gets eaten by a lion or a barbarian on the way back to my civ, my AI victim is still crippled. On Prince level, I could usually weasel a peace agreement from my early victims before they mounted any kind of a counterattack.

On Monarch level it seems almost as if the AI had "learned" from all the times I stole those workers and covers them up 90% of the time. I know that is not the case, but it sure seems that way.

[Solver]

To you programmer-types, how hard would it be to create an algorithm that just watches how a human plays, and then after playing hundreds and hundreds of games, figure out what works and what doesn't?

Extremely, incredibly hard.

Computers are not sentient. You can NOT have an algorithm "figure out" anything. There are so-called learning algorithms, but in reality, they can just observe the user. It would be possible to write an AI for Civ that remembers the choices the human makes in the games he wins and then later mimics those choices. The result would be a very weak AI, for it would mimic those things without understanding what it's doing at all.

Imagine a straightforward example. You start close to an aggressive civ, like Montezuma, so you build up heavy defenses. An algorithm that watches you would conclude that it needs to build heavy defenses and would do so even alone on an island - because it would never have a clue as to WHY you built those defenses. Computer don't even have the concepts of "why" and "reasons". They do not think.

[QuantumPion]

Originally posted by MasterDave


On moving from Prince level to Monarch level I have noticed that now the AI tends to cover up its first worker with a warrior or archer (usually archer) very early in the game. One of my favorite techniques for getting a leg up early on in Prince games was to steal workers from my neighbors with my first warrior as he wanders around scouting the map. If you do snag an AI's worker upon first contact, it gives you a huuuge edge, and even if that worker gets eaten by a lion or a barbarian on the way back to my civ, my AI victim is still crippled. On Prince level, I could usually weasel a peace agreement from my early victims before they mounted any kind of a counterattack.

On Monarch level it seems almost as if the AI had "learned" from all the times I stole those workers and covers them up 90% of the time. I know that is not the case, but it sure seems that way.

Oh yeah that's my one major beef about the cheating AI, spawning free units at the most inopportune time.

During my last game, whenever one of my automated workers would go within 2 squares of my own border, the enemy would magically spawn a chariot or knight on its own border to leap across, capture and destroy the worker, and then be killed by my units the next turn. Upon loading one turn back and scouting the area, there were no units within range that could have reached my workers...

[Solver]

The Civ4 AI does not spawn units out of thin air. It does get handicaps but it doesn't get ridicilous cheats such as spawning free units to defend its cities or attack you.

[Bluefusion]

Originally posted by xxFlukexx


At least Civ4 AI isn't like WC3 AI, which was double the gather rate, call it the hardest difficulty and call it a day, phew hard work.

Actually starcraft took the cake IIRC. If you saved replays and watched it, you would see that the AIs received astronomical amounts of crystals when a worker harvested.

But... I might be confusing Starcraft with Command and Conquer...

Originally posted by Urban Ranger


Maybe they can add "custom difficulty level" in the next patch. The player can set a number of sliders, including starting units, production and research speed, etc.

This sounds like fun.

Taking your first recommendation, I would just like to be able to control starting units regardless of difficulty. It would take the MM (such as chop-rush) out of the first few dozen turns if everyone started with, let's say, two settlers, one or two workers, and a few warriors.

You can already set the starting era, why not the starting units?

[QuantumPion]

Originally posted by Solver
The Civ4 AI does not spawn units out of thin air. It does get handicaps but it doesn't get ridicilous cheats such as spawning free units to defend its cities or attack you.

actually, it does. There was another topic where a guy ran into an enemy AI on like the first turn, bee-lined straight for his capital, and right before geting there, on turn 3, the AI had an extra warrior.

[Kuciwalker]

Re: AI Cheating -> Result bored of game already

The umpteenth newbie posting about "omg the AI suxxors it cheats" -> result bored of actually reading OP's

[Kuciwalker]

Originally posted by Solver
The CIv4 AI code will be available as a part of the SDK... but that will mean that a good number of people can work on it but not that revolutionary methods might be used to create a self-aware AI .

Aren't parts of the game code becoming available too? That's even more powerful than the ability to improve the AI.

[Falconne]

Originally posted by Kuciwalker
Aren't parts of the game code becoming available too? That's even more powerful than the ability to improve the AI.

Pretty sure it's just the SDK - haven't seen any mention of any code being released.