AI Intelligence

So many AI threads!

I'm guessing that means no

Hey darca the reason there are so many air threads is that it. Is the most important part of a single player game.

I know I am helping with the galciv2 AI changes made by the new community patch. So much AI talk...so much...can't...focusssss!!!!

As to the OP, IDK, feels about the same to me too.

No as of now the ai is still in its basic form the only difference that the different levels have is how much resources the ai has to play with. 

And from the dev streams it seems that when the ai is done it will always play at its smartest the only difference will be the amount of resources it gets. 

This means that even a novice ai will pose a challange you will just be able to overwhelm it with more ships

I thought in gc2 the AI got more or less bonuses. It was always equally dumb, just got the unfair advantage of more manufacturing, econ, and research bonuses to make up for it. Also seems they naturally can see everything and don't require scouting to colonize worlds they just go straight towards it.

I hope they will stop giving the an bonuses at harder levels but make it smarter instead. Instead of giving it economic bonuses make it smart enough to have a better economy.

yeah, id like to see smarter AI ie more efficient building placement, scouting and research priorities. For example the beginner could flop around scouting having to double back to explore a missed star system or build manufacturing and econ buildings next to each other and progressively get more efficient. I am skeptical the AI could ever really challenge a human with current programming limitations so with that in mind I would not mind the AI getting bonuses at the top end. I do want the game to be challenging.

the problem with making the ai more or less intelligent at differing levels is that now you have to write multiple ai's for different people and you have problems with trying to find bugs is the ai doing this because its supposed to be stupid or is it doing this because of a bug

having the ai always intelligent with just more or less resources is simpler for coding / bugfinding and should always  present a challenge. 

dame issue as cheating ai. I would still prefer it. your right this ai eould require you to document your design as you write it.

Starcraft was able to program different ai's for different strategies i.e. turtle, all in rushes, air unit push, etc., this made playing the computer more interesting because you never knew what you were going to get if it was set to random. For the different difficulties levels I'm not sure if they get bonuses or have more pauses between actions.

Pauses between actions would be an easy way to have different difficulty levels for the AI. However rather than pauses AI would have a set "optimized"goal for research planet set up etc., but at set intervals (depending on difficulty level) research incorrect at random. Planets could have a set "optimization" build for upgrades but every set interval (again depending on level) have a incorrect building thrown in a random place. It wouldn't have to be different AI's (unless they had different tactics which would be awesome and very doable maybe in a later expac) just one set goal with decreasing random deviations from the "optimized" path as difficulty level increased.

My last post was confusing so I'm going to rewrite it. Ais that have bonuses for harder and minuses for easier levels are cheating. I would rather have an ai  that plays harder instead of getting bonuses. I understand that without avoid flowchart or pseudo code on the air it would be harder to check multiple ais for bugs. I would prefer playing a harder an instead of giving it bonuses. I like the idea of different ais playing differently as long as they were all relatively good and no one was nerfed. On advancedalgorithms at ts hardest level then the ai could compare its playing to your style. It could even have a mode where you teach it how to play. This play style could be used per faction. You could identify the ships by modules weapons or defenses and scouts. A scout is basic all a undefended ship with no modules. What you do with ships you build. Ships that seem to go nowhere but move the computer can assume you are exploring. Ships that go nowhere but are orbiting a planet or by somethng are guarding. Techs are already categorized. Do you research a kind of tech when certain variables are reached. It can compare all the game mechancs.  What units you build on certain classes of planets or with certain random events or when you build certain wonders. Do you change how you build by researching certain techs. units are categorized. The reason categories are used is because of certain techs. As far as warfare is concerned it should be looking at the hull hit points weapons defenses and fleet modules. Its used like this than just techsbecause of faction differenses. What planets you go after first. Matter of fact a list of priorities could be made  by how often you go after them. Same things about ships in combat combining planet based on class and what's on it. Star base resourse borders. The air could try random stuff based on the situation to learn new stuff. If these things work better than the previous item then replace it in memory. Have two copies of ai new and default. Have a button where you can reset to default for when you want to start over. Have a button to shut this off. For those who want to not play this way.

Problem is learning computers don't exist, when they do your new robot masters will inform you Currently we can not program computers, or humans, to anticipate intentions. So with these limits to keep the game challenging in the higher difficulties they have to let it cheat

Shall we play a game? 

This was from the old GC2 manual.  There was a better breakdown from Brad a looong while back somewhere IIRC, but I'm having trouble finding it...  In short the AI performs more complicated tasks at higher difficulty settings.  Also, worth of note: it does not cheat, (apart from economic bonuses?), it has to scout blindly for planets/resources, has to scout other AI and the human player's assets, etc.  It does not "know" anything that a human player wouldn't know - which isn't to say that it isn't smarter than most players.  heh

Fool

The AI's economy is run at 10% of normal, no higher level algorithms are enabled

Beginner

The AI's economy is run at 25% of normal, no higher level algorithms are enabled

Sub-Normal

The AI's economy is run at 50% of normal, no higher level algorithms are enabled

Normal

The AI's economy is run at 75% of normal, AI evaluates common human tactics

Bright

The AI's economy is run at 100% of normal, AI evaluates most known human tactics

Intelligent

The AI's economy is run at 100% of normal, AI expertly picks abilities and all known

human tactics are searched and countered

Genius

The AI's economy is run at 125% of normal, All AI algorithms in place (same as

above)

Incredible

The AI's economy is run at 200% of normal, All AI algorithms are in place

-tid242

-edit: oh, and an aside - the current AI in GC3 isn't as such, rather it's just a placeholder that wanders around and interacts with the player - the big focus right now is making sure all of the systems work properly before trying to write an AI that operates within said systems.

I'm not talking about anticipation but catogorising as much stuff as possible and averaging how you played over all the previous. This won't predict anything new it would do what you have done on a regular basis on your previous games. This would look at game mechanics. Instead of actual stradegies. This is more like surveying instead of predicting. It would only implement when it's tactics fail against yours. The only thing new would be random stuff it would try. Only implementing when it is better. Good to see that we are on the same page about economic bonuses being cheating. 

How about global thermal nucular war?

Is that in the game. Sometimes a average will work. Other times it would use what you do the most because what you do the most is not on the average game. It would test this against its own tactics. It would only implement this if it is superior. 

Well, Brad's also said that the analytics via Steam should be good enough that the AI can pick up on winning strategies by players - how this will relate to AI picking up on individual behavior remains to be seen.

Personally, I'm less of a proponent of the AI learning how I prefer to play and then consistently cheesing the shit out of me, than it just becoming better overall and employing generally superior strategies.  If you think about it - you probably prefer to play a specific style (defensive, offensive, early game, late game, etc), I don't think I'd find a game very fun that would just know that I tend to like to play paced long-term defensive games, and would therefore just have all of the AI races do offensive AI rushes every single game just so that I'd always lose if I played the type of game that I wanted.  Conversely, I'd like to have the AI "learn" not to get cheesed by players (myself included). 

there's a fine line there - I hope you see it, but it seems fairly self-evident to me...

-tid242

I agree that this needs to be ressetable and shutoffable. I might reset this but probably not shut his off. I this should be the hardest non cheating level. Some people like to beat the game and move on. I play a game for years.

^^^^ This is exactly what I want. Brad was amazing at making the above reality and I want his oversight and fingers in the 'Ai Pie' here at GC III. This is how an AI should be and I really hope they exceed my expectations...

I still would rather see a smarter an at higher levels.