Cheating AI better than bad AI?

New idea for how it should cheat.

from Elemental Forums

I have an idea for how an AI could easily be made more intelligent as a placeholder until its done properly:

The AI can cheat, but not by getting free resources and seeing the whole map.

I think that the AI should be allowed to make up for bad planning by trading in buildings, units, and spells and getting new ones out of the mix. This solves the whole Starcraft-like tech tree problem, where the AI has to plan for every possibility.

 

So lets imagine an AI decides to focus on capturing resources and expanding early on. Then it sees you massing an army and moving towards its cities. It can then trash some of its buildings and technology, replace it with what it would have gotten if it knew you were going to attack early. In order to prevent it from always knowing what you are doing, you can tie this behavior to a %, like it has a 25% chance of not cheating, a 50% chance of cheating enough that it will survive but with casualties, and a 25% chance of really having no chance of being caught off guard.

I think that this would make for more interesting and less predictable games, as the AI would be better able to react to your strategies and not suffer for poor ability to plan ahead.

The way this would work is that once you, the player, sees that the AI has a certain unit or building, that is locked into place. So the AI can only trade in stuff that has not been seen by the player, or stuff that wouldn't make a huge impact (like trading one spell for another of the same type and level).  Obviously, this AI would get weaker the more you scout it, so you would be changing the outcome of the game by deciding how much stuff the AI can swap out in a crisis.

 

Anyways, would you prefer an AI that does this, or the current AI?

elemental
12,034 views 15 replies
Reply #1 from Elemental Forums Top

Neither.

Curtains - Customize Windows with additional styles
Reply #3 from Elemental Forums Top

I think the AI should stay on its current path. I would be OK with computer factions having the ability to get bonuses from things the player does.

Ex: Player defeats most of an enemy faction. One of the remaining factions would then get, for lack of a cooler term, chance card. The chance card will be in the form of a randomly picked major event that gives them some sort of equal and opposite power to blalance the game. This way one cannot steamroll the AI.

 

Reply #4 from Elemental Forums Top

I wish the AI would stop declaring war just because it's bigger. Maybe allying with smaller countries just to get that edge in power over their next rival. I'm tired of having close relations one turn and then a few turns later seeing it's hostile because of the (-5) I'm bigger than you thing. Something more like Sins where each action had a good and bad consequence. If you go to war with a faction and beat the tar out of them that's one thing, but if you have a border skirmish it's completely different. I have to pound AI's into the dust before they go "you've impressed me, let's make peace" at which point there's no point but to flatten them completely.

Reply #5 from Stardock Forums Top

Quoting Tydorius, reply 4
I wish the AI would stop declaring war just because it's bigger. Maybe allying with smaller countries just to get that edge in power over their next rival. I'm tired of having close relations one turn and then a few turns later seeing it's hostile because of the (-5) I'm bigger than you thing. Something more like Sins where each action had a good and bad consequence. If you go to war with a faction and beat the tar out of them that's one thing, but if you have a border skirmish it's completely different. I have to pound AI's into the dust before they go "you've impressed me, let's make peace" at which point there's no point but to flatten them completely.
End of Tydorius's quote

Yeah, this issue is exacerbated by the points system being a very poor indicator of how strong a faction actually is.

Reply #6 from Elemental Forums Top

Maybe a simple y=x+b is needed.

Computer Faction Military Strength = Human Faction Military Strength -200.

:ninja:  

Reply #7 from Stardock Forums Top

Quoting seanw3, reply 6
Maybe a simple y=x+b is needed.

Computer Faction Military Strength = Human Faction Military Strength -200.

 
End of seanw3's quote

Nah. It's the points that are broken. Specifically, they consider lots of shit units > a few strong units.

They are just added attack + defense values, which is just not going to work.

Reply #8 from Elemental Forums Top

Quoting Sythion, reply 7

Nah. It's the points that are broken. Specifically, they consider lots of shit units > a few strong units.

They are just added attack + defense values, which is just not going to work.
End of Sythion's quote

 

This is exactly the kind of thing my proposal would fix... the AI would be able to swap out 10 weak and obsolete units for 1 new strong unit, which would make it more competitive. Only for units that hadn't been in battle yet.

I realize this would be on top of the programmers finding a better way to value units... but this would help the AI from having to choose between building weak units now or saving up for strong ones.

Reply #9 from Elemental Forums Top

Every AI should cheat on higher difficult levels.

 

Without there would be no real challenge for experts.

 

This is not chess and a good human can always optimize game systems far better than an AI can.

Reply #10 from Elemental Forums Top

If the AI is going to cheat, it should use the toilet plop cheating method.

Meaning, that the AI is not subject to requirements and restrictions from techs and city levels. If it sees that a particular unit, building, or resource is needed, it just trains the unit, or constructs the building, or 'discovers' the resource tile in its ZoC and farms it, even if it doesn't have the pre-requisites. Things should still take time to construct, train and mine/farm, and the appropriate resources should still be required.

This is a good way to meet human player challenges, as the AI doesn't have to set up its cities or strategies in a particular way, and can just react at any point as though it had already been down the required tech paths for any specific strategy, and just plop out whatever is good for the current situation, given the appropriate resources and usual time factors.

Reply #11 from Elemental Forums Top

Quoting UmbralAngel, reply 8

Quoting Sythion, reply 7
Nah. It's the points that are broken. Specifically, they consider lots of shit units > a few strong units.

They are just added attack + defense values, which is just not going to work.
 

This is exactly the kind of thing my proposal would fix... the AI would be able to swap out 10 weak and obsolete units for 1 new strong unit, which would make it more competitive. Only for units that hadn't been in battle yet.

I realize this would be on top of the programmers finding a better way to value units... but this would help the AI from having to choose between building weak units now or saving up for strong ones.
End of UmbralAngel's quote

The problem is, does the AI even know out that 1 new strong unit is better than 10 weak ones? I don't believe it does, and that's the core problem; if the AI could figure that out, then it could be focusing on training strong units legitimately, but if it can't, then it doesn't matter what magic unit-swapping cheats you give it, it won't know which units should be swapped for which. That's why, honestly, I don't think this is a good idea - you're talking about giving the AI a second chance to make a better choice, but until the AI gets better, it'll still make the wrong choice no matter how many chances you give it.

Reply #12 from Elemental Forums Top

I don't actually mind if the AI cheats as long as it's not obvious that it is cheating, and particularly if it doesn't FEEL like the AI is playing under a completely different set of rules than me.

That being said, I'd still much prefer an AI good enough to provide a challenge without cheating at all.

 

Reply #13 from Elemental Forums Top

First, swapping out units for new units is trading resources/gaining free resources. It's just skipping over the middleman of conversion to resources and then to units.

 

Second, I'd prefer a cheating AI with disclaimers. A bad AI won't be enjoyable; it's not really fun (in context of a 4x game) to play against someone (AI or human) that zergs you with pioneers and doesn't really attempt to you to test your strategy. A bad AI will simply be bad at the game no matter what you do. If there's no real threat, you may as well  scrap the 4x/TBS aspect and make a city-builder game. As far as cheating...

 

 1) Balanced cheating. That is, it should cheat in ways that humans are good at but computers are bad at. AI, currently, sucks at seeing into the future. It has to devote a lot of time and effort on a constant basis to do what we can do unconsciously. Simple example would be communication - do you really think about sentence structure, wording, and all that? Nope, you have something you want to say and you simply say it with your subconscious doing the work for you so you can worry about other things. When your spouse leaves the room to get something from the kitchen, you don't have to actively think about them to keep track of them.

 

2) Just enough cheating to be competitive. Too much cheating becomes a bandaid. It also hurts the gameplay since you end up kludging methods to even things out rather than truly outwitting something. It turns the game into a puzzle you solve once.

 

3) Honest cheating. If the AI is going to play a slightly different game, just be honest about it. AI War does this for instance and so it's cheating is perfectly acceptable. As with balanced cheating, we know AI isn't great so we should stop pretending like AI is human-level competent.

 

Reply #14 from Elemental Forums Top

Both are bad thing - the game is not enjoyable with bad AI. Nor it is enjoyable with bad AND cheating AI.

Reply #15 from Elemental Forums Top

4) Cheats that make AI challenging rather than frustrating. Whatever cheats an AI gets should be done so that the AI remains challenging and beatable rather than unwinnable. It's not really fun (in context) to go up against an opponent that you literally can't hurt outside of some massive time investment for a single problem ie you run into a dragon (or sovereign for the matter) in the first 5 turns with 15 Dodge, 15 Armor, and 20 Attack. Even rushing Warfare, you're still going to be devoting a heck of a lot of resources simply to take out that one unit and likely so much so that any other strategy will be well behind the curve. An opponent made up of 15 1/1/1 creatures on the other hand, is a little less frustrating since you can at least hurt them to some degree.