Runaway and Espionage AIs

[shyuhe]

Actually I requested it. I think the AI prioritizes OB differently when it's in mutual struggle. You can't help him if you can't send troops through his territory

 

[MartinHarper]

Could you explain what the iVassalPowerModifier of 800 does?

No more auto AI workers swapping improvements endlessly.

This is a bug, fixed in the Better AI mod. The AI still occasionally replaces towns with farms if it is running Representation/Nationhood, though.

 

[Shafi]

(More) intelligent AI is probably quite feasible. The question is, deep down, would we really want it?...or, in other words, however much we like to complain about how the AI sux, perhaps we actually enjoy preying upon such AI stupidities....

Hell ... i know i love my stupid AI's . No way i want such monstrous AI spoiling my evenings of AI bashing.

Some might enjoy the challenge, but for the masses (like me), civ is just a way to let off some steam and relax, and how better to do that than beat the crap out of the stupid AI

In fact, i'm so bad that just y'day i kinda fired up a game on warlord level and went on the rampage just for kicks ---- it was fun ....

So --- smart AI, not for me.

 

[r_rolo1]

Like I said in the thread phungus started in the better AI subforum about this, I have my serious doubts about this XML tweaks producing a better performing AI when all the AI behave like this .... Sure, most likely they will eat the roleplaying AI alive, but that does not make them better players, just better roleplay AI smashers

 

[TheLazyHase]

Hell ... i know i love my stupid AI's . No way i want such monstrous AI spoiling my evenings of AI bashing.

Some might enjoy the challenge, but for the masses (like me), civ is just a way to let off some steam and relax, and how better to do that than beat the crap out of the stupid AI

In fact, i'm so bad that just y'day i kinda fired up a game on warlord level and went on the rampage just for kicks ---- it was fun ....

So --- smart AI, not for me.

I would love smart AI, but his bonus need to be tuned down considerably.

The runaway AI, in noble (no special bonus) is not exactly super impressive for example. This is because it's still pretty stupid, even if brutally stupid instead of being harmlessly stupid.

 

[NihilZero]

I think this is what the Civ4 developers were talking about when they were joking on how the Civ4 AI was "designed to lose", or at least designed to not play in a ruthlessly optimal fashion that would confine the human opponent to a similar constricted style of gameplay. In other words, the Civ4 AI was designed for fun and interact-ability, not competitiveness. If they had wanted to design the Civ4 AI for pure competitiveness, I bet this is what the Civ4 AI would look like.

But can you imagine how many new, general-public players would just get (unreasonably) frustrated when their "friendly" neighbor declares on them out of nowhere with overwhelming force? I say, "unreasonably," because isn't this what the human is "allowed" to do all the time? In a way, you have programmed an AI civ that plays more like a human, without the arbitrary constraints on opportunistic behavior, and thus without the easy predictability and amenability to clever manipulation.

(More) intelligent AI is probably quite feasible. The question is, deep down, would we really want it?...or, in other words, however much we like to complain about how the AI sux, perhaps we actually enjoy preying upon such AI stupidities....

Ah, but if the AIs were tough everybody would just play on noble. There would be no need for higher levels to give the AIs so many bonuses.

Broadly speaking, it's not possible to give strategy game AIs the ability to fairly beat humans. It's different in a game like chess where brute force calculating power and databases of the openings can turn computers into world champions, but in less easily calculatable games like civ which require more abstract strategic thinking it can't be done.

 

[Elkad]

Broadly speaking, it's not possible to give strategy game AIs the ability to fairly beat humans. It's different in a game like chess where brute force calculating power and databases of the openings can turn computers into world champions, but in less easily calculatable games like civ which require more abstract strategic thinking it can't be done.

Just a few years ago, we called those chess moves "abstract thinking a computer could never handle".

Can't be done now. In 20 years? I expect we'll all be playing on Chieftain.


I wonder about adding another alliance agreement besides Defensive Pacts and Permanent Alliances. Call it Favored Nation or something. Only allowed to have 1 (maybe 2 on a map with lots of civs) per civ. With it in place, the civ would trade much more freely (even at a paper loss) with it's favored partner, freely gift excess resources, etc.

 

[TheLazyHase]

Just a few years ago, we called those chess moves "abstract thinking a computer could never handle".

Can't be done now. In 20 years? I expect we'll all be playing on Chieftain.

When we began to have good result with chess, people were saying that in ten year they will be go AI with decent level. And it does not work.

An Civ4 AI need some dramatic new concept to be able to play at an human level. Nothing assure that theses progress will come out.

 

[NihilZero]

Just a few years ago, we called those chess moves "abstract thinking a computer could never handle".

Not at all. Chess programs don't "think", they just use raw processing power to calculate enormous numbers of possible variations and then choose the best one. This makes them excellent at tactics and combinations, but their inability to do any kind of abstract thinking means they can still be beaten by a long term strategy.

Even today's computers can only calculate so many moves ahead. An experienced chess player with a strategic understanding of the game can build positions that, while are very obviously advantageous to another thinking human, can't be tactically shown as being so in concrete terms in, say, ten or twelve moves. So computers can't understand the advantages of such strategies. Analogies of this can be found in any strategic game.

Can't be done now. In 20 years? I expect we'll all be playing on Chieftain.

That's very possible, especially if quantum computers become a reality any time soon.

 

[Dom Pedro II]

When we began to have good result with chess, people were saying that in ten year they will be go AI with decent level. And it does not work.

An Civ4 AI need some dramatic new concept to be able to play at an human level. Nothing assure that theses progress will come out.

Heck.. if we were willing to have individual AI player turns last as long as a normal human player's, we could probably get them performing considerably better. Of course, we want the AI to move fast and be done with their turns in a few seconds.

 

[TheMeInTeam]

Like I said in the thread phungus started in the better AI subforum about this, I have my serious doubts about this XML tweaks producing a better performing AI when all the AI behave like this .... Sure, most likely they will eat the roleplaying AI alive, but that does not make them better players, just better roleplay AI smashers

The AI only declares when it is extremely likely to win via mass power lead (or backstab, since opponent power is factored into one's own). This will result in a huge stalemate, a human dogpile target, or consecutive, nasty backstabs.

 

[michmbk]

TMIT - you mentioned in your first post in this thread back in Feb you were working on another PYL series...did that get posted ever, or did it fall by the wayside.

I absolutely loved that series - shame it doesn't have more games to it.

 

[PieceOfMind]

Not at all. Chess programs don't "think", they just use raw processing power to calculate enormous numbers of possible variations and then choose the best one. This makes them excellent at tactics and combinations, but their inability to do any kind of abstract thinking means they can still be beaten by a long term strategy.

Even today's computers can only calculate so many moves ahead. An experienced chess player with a strategic understanding of the game can build positions that, while are very obviously advantageous to another thinking human, can't be tactically shown as being so in concrete terms in, say, ten or twelve moves. So computers can't understand the advantages of such strategies. Analogies of this can be found in any strategic game.



That's very possible, especially if quantum computers become a reality any time soon.

While quantum computers will potentially mini-revolutionize computing, I doubt they're going to be able to improve things like AI for a long time. The problem is, making an algorithm that can take advantage of the nature of a quantum computer is very difficult compared with traditional programming. More likely its first applications will be things like code breaking (an encryption overhaul is going to happen eventually since bank encryption will become vulnerable).

I would expect that improvements to a game AI like this are more likely to come from more modern but still non-quantum computing.

 

[Shafi]

whats "quantum computing" ?

 

[PieceOfMind]

whats "quantum computing" ?

Off topic: A quantum computer is a device which uses quantum phenomena rather than more classical physics to perform computations. They're still very new, very theoretical, and not very well researched. They require several disciplines, including theoretical physics, engineering and computer science, even mathematics. Quantum computers, if they become practical, will be a huge deal more powerful than current computers in particular areas of computing like factoring very large composite numbers (important in cryptography). Only the most brilliant computer scientists and physicists have written good quantum algorithms to date. Building the devices is the most difficult bit and so the technology is lagging behind the theory.

 

[TheMeInTeam]

TMIT - you mentioned in your first post in this thread back in Feb you were working on another PYL series...did that get posted ever, or did it fall by the wayside.

I absolutely loved that series - shame it doesn't have more games to it.

It fell by the wayside. I decided writing XML scripts based on forum players impractical: only a few approaches are markedly different from current AIs, and the PYL element would need very popular, consistent forum-players to draw appeal, and then it would have to not offend anyone. I didn't want to deal with all that.

Now, just making another PYL is doable, but it is easily the most annoying type of map to create saves for (one has to swap archery off for each of the civs etc).

Possibly, I'll do another one and just provide the noble save, outsourcing the work of adding archery and other techs for high levels to people who care, and simply cleaning BEGINUNIT and ENDUNIT for all civs. At least the # of units will be right, just high level guys will have warriors instead of archers (still getting the workers/settlers and other bonuses apart from techs).

Under those constraints, I'd be willing to make it a more common running series.

 

[hunterai]

What is a PYL?

 

[RRRaskolnikov]

Play Your Leader game... iirc The host provides a map, and you can play any civilization which is in the game. People make reports to compare openings/strategies...

Cheers

 

[daishi5]

While quantum computers will potentially mini-revolutionize computing, I doubt they're going to be able to improve things like AI for a long time. The problem is, making an algorithm that can take advantage of the nature of a quantum computer is very difficult compared with traditional programming. More likely its first applications will be things like code breaking (an encryption overhaul is going to happen eventually since bank encryption will become vulnerable).

I would expect that improvements to a game AI like this are more likely to come from more modern but still non-quantum computing.

Ok, my knowledge of computer science is limited, so forgive my ignorance and explaining where I am wrong would be helpful.

I don't see why the computer could not be much much more efficient at city building, and civ management. At least with all the internal functions that do not require interactions with other civs. Things such as, city placement, city improvements, teching, worker use, and production are all basically an attempt to find a maximum output from a given set of choices all with a known level of production. Those things that are not interactions with the player are mostly just math questions. Of course, now that I think about it, it seems like pathfinding would be similar, and I seem to remember that pathfinding is really just guessing according to a few good estimates rather than an actual optimized solution.

 

[PieceOfMind]

Any solution that doesn't adequately weigh the resources cost (e.g. running time) is likely to be impractical. A very intelligent brute-force AI is theoretically possible but physically not practical if you don't like the AI turn times taking longer than your own.