Human/AI equality

[Cyclotron]

Re: Re: Human/AI equality

A note: The only reason I am responding to this is because it's my thread.

Okay, I just sifted through the drivel, and there is only on paragraph that actually pertains to the topic.

Originally posted by Coracle
I suppose you'd think in the real world it would make sense in the 20th century for China to still hate Greece for having wiped out Troy over three thousand years earlier. It makes no sense at all, just like the Diplomatic AI.

Coracle, if you read my post, I don't care about the real world. To play a competitive game with the AI, the AI needs to want to win like I do. I desperately want the AI to remember anything and everything I do, because if I do it should too. You can't honestly tell me that you don't remember a civ attacking you 4000 yers before, so AI civs should not either. I want the AI to be on an even playing field with me, and that is not accomplished with the repairing of reputation. Reputation should never get better, and an attack I made in the stone age should still be fresh in the mind of the AI in the modern age... otherwise it's cheating in favor of the human. I don't care about sense, and I don't care about realism, I want the AI to care about mopping the floor with my @$$ through all four eras.

[Inverse Icarus]

Re: Re: Re: Human/AI equality

Originally posted by cyclotron7
A note: The only reason I am responding to this is because it's my thread.

Okay, I just sifted through the drivel, and there is only on paragraph that actually pertains to the topic.



Coracle, if you read my post, I don't care about the real world. To play a competitive game with the AI, the AI needs to want to win like I do. I desperately want the AI to remember anything and everything I do, because if I do it should too. You can't honestly tell me that you don't remember a civ attacking you 4000 yers before, so AI civs should not either. I want the AI to be on an even playing field with me, and that is not accomplished with the repairing of reputation. Reputation should never get better, and an attack I made in the stone age should still be fresh in the mind of the AI in the modern age... otherwise it's cheating in favor of the human. I don't care about sense, and I don't care about realism, I want the AI to care about mopping the floor with my @$$ through all four eras.

it's really too bad that the human can wimp out and quit when he or she is getting slapped by an AI early on... imagine if you had to play each game to victory or your annihalation, whichever came first.

that would be sick. and then maybe you'd be in the AIs position.

[Cyclotron]

Originally posted by UberKruX
it's really too bad that the human can wimp out and quit when he or she is getting slapped by an AI early on... imagine if you had to play each game to victory or your annihalation, whichever came first.

that would be sick. and then maybe you'd be in the AIs position.

I'm not sure what your point is. If the human wants to cheat, that's their perogative... but I would like it if the game would not force cheating upon me (reputation) or make certain choices in the game entail cheating automatically (privateers, UN).

[Inverse Icarus]

Originally posted by cyclotron7


I'm not sure what your point is. If the human wants to cheat, that's their perogative... but I would like it if the game would not force cheating upon me (reputation) or make certain choices in the game entail cheating automatically (privateers, UN).

yea. reading my statement i dont even know what i meant. maybe it's time i went to bed.

[Cyclotron]

Originally posted by UberKruX
yea. reading my statement i dont even know what i meant. maybe it's time i went to bed.

Alright then...

[dexters]

Originally posted by cyclotron7


I'm not sure what your point is. If the human wants to cheat, that's their perogative... but I would like it if the game would not force cheating upon me (reputation) or make certain choices in the game entail cheating automatically (privateers, UN).

Cyclotron, this issue has been beaten to death. Short of running 1 terahertz machines with some sort of neural network, AI is AI. Given the requirement for the game to run on desktop PCs, there's only so much that can be done. Could intelligent agents that learn from your actions have been implemented? I'm sure it could have. In fact, I Soren has a simple system with the City Governors and the aggression settings in the early game. Regarding the governers, the system was so re-liant upon you telling it what to do, as opposed to observing what you do, that it has had the effect of aggevating people more than helping them. Fixes in subsequent patches had made it smarter and more intuitive, but that's an example of an intelligent agent, and when used right, it can anticipate your moves (it's happened to me a lot)

But using intelligent agents to control AI behavior is still, I think, a ways off. Imagine 15 civs (less the human player) having 15 different agents of their own, learning different things at different places, then carrying out what it has learned into action. I think the end of turn calculations will be it's enough to crash a 2 Ghz Athlon with 512 of DDR.

Anyways, I've strayed too far. My point here is, AI will never beat human p layers. It cheats in more subltely than it had cheated in Civ2, and from my understanding, REGENT is the even playing field. And it's a challenge. Certainly very beatable, but it keeps me on the edge of my seat, and I do engage in unfair things like re-loading and cycling a turn to get a better item in a goodie hut

[Dienstag]

Hey, Cyclotron, good point about the human remembering the whole game, so limiting the AI's memory would be unfair. But I think you go too far with

Reputation should never get better...

What if a civ that fought against me in 3000 B.C. also fights on my side in against a 3rd, stronger civ in 1000 B.C. It takes the two of you (former enemies) to prevail against another threat. To me, at least, I'd look more favorably on the first civ after that. By the current model, that civ would pretty much still loath me. Do I deserve that? Do 2000 years of peace and a mutual enemy mean nothing? Just a thought, and I realize most of my example is based on a balance-of-power concept that the AI is sorely lacking.

[Cyclotron]

Dexters, that is a very well thought out post. I agree that the AI will never be the equal of a human... I accept that. I simply want to make sure the AI is as good as it can be, and the three features I pointed out seem to me to be working against that.

Originally posted by Dienstag
What if a civ that fought against me in 3000 B.C. also fights on my side in against a 3rd, stronger civ in 1000 B.C. It takes the two of you (former enemies) to prevail against another threat. To me, at least, I'd look more favorably on the first civ after that. By the current model, that civ would pretty much still loath me. Do I deserve that? Do 2000 years of peace and a mutual enemy mean nothing? Just a thought, and I realize most of my example is based on a balance-of-power concept that the AI is sorely lacking.

While I maintain that reputation should never get better, I also belive that reputation should be handled differently. There is a difference between reputation that says, "if they have ever attacked you, but non-cooperative forever" and reputation that says "if they have attacked you, never trust them again and be on your guard... but if they offer something advantageous, don't ignore it just because we hate them."

If an AI you attacked becomes your enemies' enemy after a few thousand years, it should ally with you because that is the best course of action. However, it should not agree to RoPs, etc. just because yu are an ally, and it should hold back enough of its forces to keep its guard up on its border with you. And, if the programmers were really good, maybe it could stab you in the back as soon as the war is done... if you were suitably weakened.

[dexters]

cyclotron, I suppose the individual civ's traits affects how they remember your past actions. And I'm sure we could tweak it to perform human like recollections. But the problem is, we're merely emulating, not creating human behavior. The AI remembers everything, and I'm sure it has no idea what they all mean. What it remembers is assigned some value that is added up into some neat equation or equations whose result will trigger some pre-programmed routine on their behavior.

That's how the AI works. The only way to make them forget is to tamper with the variables (say, lower aggression value). Unlike humans, they can't actually remember a bad thing you did to them but at the same time, "get over it" and be good friends because of good relations in recent years. That's a complex series of comparisons that we humans make in our mind, but is something computer AI cannot make, not without substantial coding and processing power and even then, you're probably running a complex emulator. Perhaps with advanced versions of AI, AL or neural networks, intelligent agents, but the technology that will make Civilization's AI truly threatening (unaided), as in BIG BLUE threatening, is still in the future.

I have to say though, the AI in this game is fabulous.

[reefer addict]

i think your reputation should follow your actions. back in the day england tried to shut down the drive for freedom here in america. granted at the time most americans were furious about this , but in time ( and a couple more wars or so) america and england got over it and became gracious to one another and the closest of allies. now in civ3, there would be little chance of england becoming allies, let alone let the americans base troops there. i think that your reputation should improve over time, with trading of goods, transfer of technology and military cooperation. but since thats not how the game works , its either kick a$$ or kiss it as for the U.N. , ive won and lost with it so i cant say its for human benifit only, but i still made it require robotics so i could play out the modern age , win or lose.

[Akka]

I think that I just plain hate the Cyclotron's vision about AI. That's a powergamer vision, which is geared toward "challenge", and win-or-loose basis. History is not about win or lose, it's about surviving. I prefer a less challenging but more immersive AI.

[Master Marcus]

We're far away from human/AI equality in decision making. Still, Civ III's AI is quite impressive and hard to beat at higher difficulty levels. You can dominate the charts most of the time and if not be careful with your military, be surprisingly in danger in a modern war. This is quite the contrary to many other TBS, classics such as Civ II and SMAC where you can easily conquer the map with few squadrons. You need at least 10 times the number of units in Civ III in order to thrive.

[Cyclotron]

Akka: Yes, history is not about win or lose, but a game is. I understand the need for immersion, but it just strikes me as grossly unfair that the AI suffers from long term memory loss simply in the name of realism. Immersion is important, but I honestly don't think that making the AI more competitive and more human-like in behavior will reduce that.

Reefer: I understand that reputation is very important, but I think you'll ahve to agree with me that the human player doesn't use reputation. I regard a civ that attacked me in 4000 BC the same way as I do a civ that attacked me in 1000 AD, and so do you (most likely). All I am saying is that, if we want the AI to be the best it can be given our current technology, we need to remove all factors that cheat for the human. Reputation is one of them. It removes the human's obligation to treat other players seriously during the game (as per MP) because eventually the AI will just forget everything anyways.

I just want to make it clear that I realize the AI will never be equal to a human; I just want to close the gap a little, and this seems like a very esay way to do it without even re-writing the core of the AI.

[reefer addict]

its true that you never forget who attacked you, but in this game where 15 civs all attack you at once that you should be able to somehow regain some reputation. ive had games where they all sign alliances early in the game because they see im sitting on some unseen resourse. i think it should be based on your actions. if you start the war,you lose points. if you raze cities, you lose points. if you back out of deals, you lose points but if your the one always having war shoved on you then why is it your rep. that gets hammered. self-defense should not be penalized in my opinion

[vondrack]

Re: Human/AI equality

Originally posted by cyclotron7
The UN vote. You will never, ever catch me voting for anybody but myself. I want to win, period. All human players want to win. If the AI even wants to be competitive, it needs to want to win as well. While a diplomatic victory may be interesting to have, no human would ever cause another player to win in their right mind. I always turn off diplo victory because of this... it seems like an obvious advantage for the human player.

This is very true. As the UN works now, I would never vote for anyone else but me. If the AI leaders behave differently (and they do), they inevitably behave absurd. The ultimate goal is to win - a vote for anyone else but me might help him/her win, so I will always vote for me. And they should vote for them, too, effectively rendering UN useless.

The key problem is that a lost/won UN vote ends the game IMMEDIATELY. What is then the reason for voting for anybody else but me? Let me suggest a solution... sorry, not a polished solution, just an idea:

What about changing the UN vote mechanics so that not just one vote wins you the game, but two, or even better, three of them? Consecutive, I mean. THEN, I would SERIOUSLY consider voting for other leaders, because sometimes my vote might prevent the current UN Secretary General from keeping his post (even if it was not me to replace him). THAT would add quite some diplomacy. It would be a great reason for wiping out a civ that is blocking your successful UN vote, for example. I guess it would be necessary to add something like "UN vote support" to the bargaining/trade screen, as it would actually become a trading article, just like treaties/alliances.

The UN victories would be pretty rare then, but I believe the game would be much more fun (especially as the UN wonder gets built in the later stages of the game that many people claim to be tedious... this might bring some excitement for them... ).

As this would tone the UN wonder importance down a bit, if not seriously, I would suggest balancing it with some extra bonus - Stuie gave it an extra trade bonus... that would be fine, I think...

Thinking of this, there is one piece of information I am seriously missing in the foreign advisor screen. I do not know if that Xerxes guy likes Elizabeth, hates it, is going to marry her, have her killed by a hired assassin or what... Having an embassy in Persepolis (or London) should allow me to know this (ok, not about the assassin thing... ). This is extremely important when deciding whether to go to war with somebody... and it would also be very important when estimating one's odds in the UN vote.

As it works now, the UN wonder is sort of too sudden death and forces you to either build it (perhaps preventing the vote from taking place) or bribe everybody to vote for you. This sort of sucks. Should there be a compulsory vote every, say, 5-10-15-whatever turns (irrespective of whether the UN owner wants it or not), it would, IMHO, add a lot of fun to the gameplay.

Just writing as it flows through my head, so be merciful, if I missed some crucial point...