Dano13

Join Date 03/2003

AI - Roleplay or Play to Win?

One of the HUGE advantages that a human has over an AI is that we can see their thoughts. What I mean is that we KNOW for a FACT what they currently feel towards us and towards others. Barring any random event, if we are close to someone we can pretty much leave our border's undefended. The same can not be said of an AI vs a player. The AI has NO IDEA of what we really think of them (usually Empire Food- at least to me, though we pretend to be their best buds).

Consequently we are usually never backstabbed (though it HAS happened to me once, which I was happy to see). Not sure why ... it was the Screen "A Secret Meeting occured and it was determined that ... blah blah blah" so maybe it was a random mega event. If so, that is disappointing.

So once everybody loves you there is very little tension left in the game. Of course the flip side of the coin is that if relations play so little role in the game, then Diplomacy becomes trivial - ala Rome Total War. Besides people play the game in different ways. Some play to win at any cost, and some to roleplay, and of course some roleplay by winning at any cost. If the AI played to win (i.e. would happily stab a friend in the back for gain) then those people might think that this was unrealistic.

It would be really cool I think to somehow make this a game option i.e. AI - Play to Win or Roleplay. But that would double the number of AI personalities which is probably an insane amount of work.

The next best thing would be to have an option to HIDE what the race thinks of us. I mean in real life we are only given a person's expressions and words to determine what they think, and a cynic like me knows that that is USELESS.

So if this option was selected, then when you go to the race screen you wouldn't see what they thought of you. What about what they say? Well, maybe each race has a "Deviousness" rating. If they make a random roll and fail this rating then they let their true emotions leak out (WE HATE YOU), otherwise it defaults to the Neutral or Impressed type of greeting, even though they secretly hate you.

Now with this option you DO have to worry whether the Dregin are plotting something behind their chuckles and smiles.

Sure the AI would still have an advantage in that it knows what other races think of IT, (since I imagine this is an important parameter in the AI personalities, which remain untouched), so maybe kick in a few points in gamescore or whatever if the player chooses this option.

Just my 2 BC
Dano

Locked Post

Advanced Search

Watch this post

Do not email me updates for this post Email me updates for this post

Successfully updated karma reason!

I made a similar post in another thread but to repeat.

The highest objective of an AI player at its highest level (not counting artifical bonuses) should be to WIN. Period.

Regardless of its roleplaying abilities it should always, always do its best to win and hence not to lose. Which means that if they see another player that is about to win and that player is not an ally then it should do everything, EVERYTHING in its power to defeat or weaken that player. That includes allying up with other players and together bringing said player down.

Ive seen Frogboy saying that this wouldnt be "fun" for most players and I strongly dissagree with that. I believe most players would want games to be a challenge and not in the way that their opponents get artifical bonuses but rather that they play well and do everything it can to win.

If the devs dont realise that then it wont matter how many cores and how much computer power they can take advantage of if they refuse to write algorithms that will make the AI do its very best to win.

Successfully updated karma reason!

Philocthetes

Join Date 10/2003

+153

Gee, it looks like Brad's recent line about "role play" caught more than my attention. I confess to not having finished the thread where I said basically "yay" for seeing him talk like this, and what do I see on first check for latest posts but this swell new thread.

Regardless of its roleplaying abilities it should always, always do its best to win and hence not to lose.

I have a strong inclination to accept this sort of position, except...

people play the game in different ways. Some play to win at any cost, and some to roleplay, and of course some roleplay by winning at any cost.

In other words, I don't see Being Super Smart and Staying in Character as necessarily mutually exclusive goals for AI design. IRL, no person or group can always behave in the maximizing mode that Yamota describes b/c we all have multiple goals that can be in conflict with each other. It sounds to me like Brad is trying to think of ways to write *sets* of algorithms that play nicely together and yield an end result that makes "story" sense as well as "strategic" sense.

Successfully updated karma reason!

Yeah, the AI should try to win more. Sometimes they just do stupid things because we have a close relationship, but then my warships are attacking their homeworld. They should still expect that kind of thing, and even want to do the same to me.

Successfully updated karma reason!

Im not saying that they should be mutually exclusive sometimes the two goals to win or stay in character could go in hand.

But what Im saying is that that on the HIGHEST setting, the drive to win should always trumph over the roleplaying aspect.

If players want them to be balanced then they could pick a lower difficulty setting. But there should definetely be a win at all cost setting.

Successfully updated karma reason!

Philocthetes

Join Date 10/2003

+153

This is almost self-contradictory; "not...mutually exclusive" doesn't work with "drive to win should always triumph." I think you mean something more like "As you increase the diff level, the AIs should give steadily greater weight to winning and less to role playing."

But rather than quibble over language, I'm suddenly moved to ask what folks think about a built-in Big Brother that could write to your prefs file (or perhaps a more durable file) and give the AIs a crude analogy to the diplo reports we get on them.

If you have a pattern over all your games, say 30% of the time you form an alliance you end up attacking your ally, then they AIs could look at you differently regardless of the species you're playing for a given game.

Successfully updated karma reason!

Sorry english is not my first language but what I meant is that the AI should try to stay in character aslong as it doesnt conflict with its objective to win the game. If it does then the objective to win should trumph.

But to move to the other issue. The best way to predict future behaviour is past behaviour (dont hate me but I heard that from dr Phil

But for the AI to judge you on this game for actions you have committed in other games seems a bit harsch and would probably limit your options to go from a warmongering empire in one game to a peaceful one going for a diplomatic win in another game.

But still it does make alot of sense so if the AI could somehow analyze the way you as a player is acting, even if its in the current game, then it would go a long way to make it a better player. For example if you are picking of one empire one by one then the remaining players, that are not allied with you, should see that and expect that you are next and take preemptive actions to prevent that. By for example helping other empires you are currently at war with or even better, attacking you while you are busy with another empire thus forcing you to engage in a two front war. Thats a tactic I commonly use against the AI.

Successfully updated karma reason!

This piggybacks on something I was thinking about earlier:
If the AI were to suddenly start playing for me, it _would_ have certain feelings about the other empires. It would be fairly trivial for the computer to, um, computer this. Let's call that my citizen's feelings on other players. What I'd like to see is that when I decide to attack another player (and thus a War Vote is taken), my citizen's opinions about that civilization will be factored in. Currently, the War Vote is pretty much how my citizens feel about _me_. It ought to be how they feel about me+the other civ. For instance
War Vote
My popularity +
-40% if relations are "close"
-30% if Friendly
-10% if Warm

And perhaps there are bonuses if your citizens feel cool to or are wary of other civs...

Successfully updated karma reason!

I like the idea of it being harder to go to war with empires with whom you have good relations. In general a system that constrains the player's actions a bit in terms of how and when they can go to war (or stay out of one) would be pretty interesting.

Successfully updated karma reason!

Mistralok1

Join Date 05/2006

+22

The AI can, and will, decide to go to war with a civ with which it has friendly relations. I spent the majority of my last game having warm, even friendly relations with the Korath. If we didn't have the common border, they may have even reached close. After many turns at warm, our relations began to suddenly drop. I hadn't done anything that I can think of to cause this- believe me, I really didn't want to antagonize THEM. The first turn, warm to neutral. The next turn, neutral to cool. And so on. Each turn, one lower, until they reached hostile and declared war. It was obvious that they had made their decision to attack while we had a prolonged period of warm relations, and that the restraints that the game puts on them is the only thing that saved my game. I had a few turns to marshal my forces, and prepare for the attack that I sensed was coming, and managed to beat it off and survive. If the AI were allowed to 'backstab' me, then the first notice that I would have had would have been the armadas entering my systems; I would have been wiped out in a single turn, and that wouldn't have been any fun AT ALL. If the AI is allowed to behave in this manor, then the galaxy would be forced into a constant state of cold war paranoia, with huge amounts of resources devoted to the military. Economies would be drained, trade would be useless (why bother when you are likely to lose them anyway), diplomacy would be totally useless, influence and research victories would be almost impossible. In short, it would trash the game. Utterly.

Successfully updated karma reason!

Y'know what?
I think you should have to spy on an enemy to learn how they feel about you. The hints from conversations, i.e: "I have rusty daggers sharper than you." Or, "You complete me." But in depth crap like...
Drengin Relations
Altarian: Cool
Terrans: Neutral
Thalans: Warm
Korx: Hostile
Ect. Should be learnt through spying.

Successfully updated karma reason!

Roleplaying all the way, no question.
Which doesn't necessarily exclude backstabbing... if it's in character for a civilization to do that, why not? Some of them *are* evil, after all.

The main difference between roleplaying and playing to win is the perception of limits. In both cases, you (or the AI) in the beginning try to lead/manage your civ as best as you can with the goal of survival, expansion, prosperity,.... But when playing for the win, you (and the AI) are aware that e.g. the game will end once certain conditions are met. So the strategy has to change accordingly and you are no longer only trying to do the best for your civ but instead are trying reach that goal even if it means that you have to do things that would be detrimental to your civ's well-being if the game went on after the end (like declaring war on all your friends to prevent them from getting an influence victory). And that's simply no fun (neither when I do it nor when my "friends" do it to me). If I want that, I go and play multiplayer games. In this game here I want to pretend to manage my civ, not compete against other gamers, be they human or AI.

Successfully updated karma reason!

If the AI were allowed to 'backstab' me, then the first notice that I would have had would have been the armadas entering my systems;

I don't think I agree with this. If you knew the AI had the potential for backstabbing you, you would start dropping spy ships near or inside it's borders. If the AI started to marshal fleets, you would respond by gathering your own. The moment you see those fleets head for your border, yours would be heading there as well. I'm guessing there would be more "accidental" wars based upon mis-interpretations of intentions, but this would be realistic.

It's how the AI should watch me. When it sees me send a couple of ships straight for every galactic resource it owns, with constructors following in their wake, it should treat it as tantamount to a declaration of war...

Successfully updated karma reason!

I'm more of the idea that it should be within character. Like the Yor, it says right in their race description that any alliance or friendship with them should be taken with a grain of salt. The Yor only make alliances of convenience. The Yor, Drengin, Korath, and maybe even the Korx should be total backstabbers, ready to drop you the moment it becomes convenient.

But makeing every AI this way... even on the most difficult level would alittle to extreme. This is a game, and games should follow their own rules and story. The Terrans wouldn't attack and destroy the Thalan merely because they wanted Hatch III and the Thalan Empire got to it first, or because they were the only ones between them and controling the entire galaxy.

Successfully updated karma reason!

Mistralok1

Join Date 05/2006

+22

Y'know what?

I don't, but my grandfather did. He played baseball. Second base, I believe, during the 1940s.

I think you should have to spy on an enemy to learn how they feel about you.

Humans are lying, cheating, backstabbing weasels. We are the only lifeform on this planet that does these things. I would expect that in the GalCiv world, things would be the same. Insects don't lie, reptiles don't lie, machines don't lie- and I'm pretty sure that if I ever met a Drengin he wouldn't lie either. "You have such nice skin. It would crisp up to a beautiful brown over my firepit" would probably be a compliment from him. And I think that he would expect an equally disgusting 'compliment' in return, and be perfectly fine with it. My point is simple. Every creature has it's strength. Some have thick scales, or sharp teeth, or strong claws, or deadly venom. I don't have any of these. What I do have, is: Come here sweetie. I'm not going to hurt you...

Successfully updated karma reason!

Mistralok1

Join Date 05/2006

+22

If you knew the AI had the potential for backstabbing you, you would start dropping spy ships near or inside it's borders. If the AI started to marshal fleets, you would respond by gathering your own.

Absolutely right, Purge. This is the way we, or any other competent player would react. This is my point. If I had to play like this, I wouldn't play at all. I have no desire to live in constant fear of attack.

Successfully updated karma reason!

I'd miss the variety. In strategy games I like a whole spectrum of AI's that lead to the most random game possible.

From completely pacifistic to none stop anti-life aliens from hell that never stop fighting until someone stops them. All through the grades, races who spy, plot and backstab for their own ends, to races who really want the universe to be a political union. Yes also those who will win at any cost.. But I think that’s what makes for a fun game myself and each time a different one.

If all the AI’s were all set to win by any means necessary, you’d have very limited ways of differentiating between them and that to me would not be fun. I’d rather have them play in their own ways but just have their intelligence tweaked, as people have pointed out here.

If a good player is going around abusing races one by one, then your absolutely right, it would be nice for their friends to either threaten them as one voice and mean it or simply cut their ties and isolate them etc.

Successfully updated karma reason!

Philocthetes

Join Date 10/2003

+153

Humans are lying, cheating, backstabbing weasels. We are the only lifeform on this planet that does these things.

That's not exactly true (especially the "weasels" part

). It's risky to describe animal behavior in human terms, but I've read some primatology stuff about baboons who deliberately misuse signal calls, e.g. giving the warning cry for "snake in the grass" when there's no snake and then slapping the ground in delight while the rest of the troop scatters.

It's been about twenty years since I did this reading, but back then a few folks looking into the evolution of language considered that it was born of lies and/or jokes.

Re "cheating," well, what about birds who lay their eggs in others' nests or the female birds in many species who have one or two eggs in a clutch that are fertilized by a male other than their nestmate?

Backstabbing might still be "ours," but we've lost many of those "unique" traits to chimps--tool use, "war," infanticide, matricide, patricide, cannibalism...

Successfully updated karma reason!

Maybe alien civs could choose to lie about how they felt about us, if their relations are declining, but through espionage you could find out what they 'really' felt like.

Successfully updated karma reason!

From completely pacifistic to none stop anti-life aliens from hell that never stop fighting until someone stops them. All through the grades, races who spy, plot and backstab for their own ends, to races who really want the universe to be a political union. Yes also those who will win at any cost.. But I think that’s what makes for a fun game myself and each time a different one.

I have to agree. I liked that a lot about GalCiv2, but I'm finding that in DA those differences have actually decreased significantly and the AI races seem a lot more homogenous.

I'd like to see these things be and stay more varied as they used to be.

I think the OP has a great concept visualized and see it as a vast improvement to the game.

Also, This:

I'm suddenly moved to ask what folks think about a built-in Big Brother that could write to your prefs file (or perhaps a more durable file) and give the AIs a crude analogy to the diplo reports we get on them.

If you have a pattern over all your games, say 30% of the time you form an alliance you end up attacking your ally, then they AIs could look at you differently regardless of the species you're playing for a given game.

Would be cool, though I'm not sure how practical it might be from a programming standpoint.

Successfully updated karma reason!

Philocthetes

Join Date 10/2003

+153

DA those differences have actually decreased significantly and the AI races seem a lot more homogenous.

I sympathize here. Oddly, the enviro tech that I like so much seems part of "the problem" here. Hopefully the devs will be able to make some mildly serious additions that will link different species to different baseline environments and make the research paths different accordingly (e.g. easier for radioactive natives to handle toxic than it is for them to handle "terrestrial").

Successfully updated karma reason!

I just have one question pertaining to this. Why on Earth did Brad make the comment that the AI wasn't designed to say, send a few fighters after your unprotected colony ship or constructor or whatever? I believe he was justifying it in the sense that he wouldn't want that happening to him so he didn't make it a part of the game. Regardless of how his wording was, why would this be? I would think that any civ upon noticing a rogue vessel in their space would send out a ship or two, if not to outright blast them into oblivion, at least to say "Hi" in a way. Does anyone else remember reading this and if so, what did you make of it? Does it make sense? Do you agree with the reasoning?

Successfully updated karma reason!

hmmm...

The philosophy behind the re-development of the AI for DA, was based on the feedback we all gave.

It was made clear in the early stages that the AI would only take hostile actions when it was beneficial for it to do so, or at least, where there was no serious risk to it's own goals.

Hence, attacking an unprotected colony ship, purely on the premise of opportunity, would be considered in as much depth as attacking a planet.

All decisions will now be analysed by the AI.

It will 'weigh up' the consequence of oppotunism, just as it would 'weigh-up' any other hostile act.

This adds a layer of perception for the human player about decisions made by the AI.

If it turned out that it did in fact destroy a colony ship with a fighter, you can now be pretty certain that the AI has considered the consequenses, and is prepared to deal with even the most serious escalation.

Of course, this could be a bluff (and where the AI racial traits come in) but...

For those of us that like to RP our games, some welcome 'critical decision gates' that up to now, at least for me, have been rather predictable are introduced.

Well, that's my take on it.

Successfully updated karma reason!

Regardless of how his wording was,

Post #6 in the "Dark Avatar > Long time DA beta testers review/opinions on the expansion" thread.

Doesn't look like RP to me.

Successfully updated karma reason!

Philocthetes

Join Date 10/2003

+153

Doesn't look like RP to me.

In that other thread, I read Brad to be talking about "fun" and not how a given hardcore player might interpret the character of a given AI.

ToS, given your apparent distaste for "RP," I'd think you would sympathize with Evil S here b/c his take seems interested in tactically sensible behavior and not an attempt to build strong personalities into the AIs.

Successfully updated karma reason!

I tend to RP myself in strat games of this sort, so I like that the AI does as well. That said, I'd say it makes sense for the highest AI levels should be playing purely to win the game, to provide maximum challenge. Ideally, I suppose there'd be a switch somewhere to flip between the two AI behaviors, but programming two completely different sets of AI behavior is asking far more of SD than is reasonable, I think.

Successfully updated karma reason!