So i played this Big Europe game and one of my teammates needed to go. he was the weakest player so we played the game further. We already lost, but we didn't know that then. The AI expanded in Turkey even though my teammate wanted and expanded in Sicily even though i wanted that. I'm used to the AI attacking neutrals more than the enemy, but now all of his orders near us were attack teammates. this made my teammate not take a region in turkey because he was attacked by the AI(who was supposed to be on our side). And even though we didn't go that far because we surrendered the AI did the same in Sicily. Is there something i'm missing because this just seems like an annoying design flaw. Here is the game if you're interested.
https://www.warlight.net/MultiPlayer?GameID=9448744

the ai wasn't attacking your teamate. it was attacking a neutral which your teamate also attacked.

I think the ai should be smart enough, though, to modify its orders after your last teamate commits to not attack any territories that any of its team is taking.

Edited 9/13/2015 00:45:29

I know that it was attacking neutrals, but why should have the AI have the option to attack teammates at all? It should've just transfered armies. Well maybe in a scenario in which the AI only misses a few territories for a big bonus and it needs to attack a teammate for that, but if that is too hard to program i think it would be better to leave it out completely.

The AI tries to complete logical bonuses for maximum effect, if you were in 1 territory of a huge super-bonus the AI controlled the rest of, you'd be pretty pissed if the AI didn't attack you and complete it.

Most of the time I see people complaining is when they don't understand this, and are (usually) attacking the AI to try and take its bonuses.

Looking at the game you posted that's exactly what you were doing, why didn't you let the AI take those bonuses and attack the enemy instead?

In my opinion we really need a better AI, this one is so poor by todays level of play and complexity of maps now. Then people might trust it more and not attacking it during games.

Edited 9/13/2015 01:50:24

Yes i know that, but why is the AI using the command attack teammates true on a neutral which a teammate also wants to take. And the AI had a few territories in Turkey while my teammate controlled most of it. For the region Ankara same thing

So, first of all, AIs can only see 1 space ahead, also it doesn't know "a teammate also wants to take it"

What the AI is doing is looking logically at the game and thinking "these bonuses need taking"

What you're missing here is that trying to take Turkey and Sicily were actually bad moves by you, and in fact if I was on your team I would have pointed it out as such. I'm happy for you to disagree with me, I'm just trying to answer your statement: "Is there something i'm missing because this just seems like an annoying design flaw."

first time i heard logic and AI in one sentence. How is trying to take turkey when you have most of it a bad move anyway? And red was already in the region Ankara when the AI used an attack teammates order on one of its territories. Everywhere where he was one territory away from a teammate he used attack teammates true. How is that not annoying and unnecessary?

Ok, I'm going to answer, but if you're just trolling I'm going to be annoyed... But other players might also be interested as well though, hopefully.

When I say logic I mean computer logic. Every program ever written uses algorithms to determine an outcome from an input, this program is then compiled into binary code which runs through a processor. This is esentially what a computer is. The logic is created by the programmer when he writes the code and the computer follows this logic no matter if it's "bad" or "good" logic. So therefore all AIs, no matter what, use logic in generating the outcome to any given problem. Also it's always dependent on the programmer how good that logic is.

https://en.wikipedia.org/wiki/Logic_in_computer_science
https://en.wikipedia.org/wiki/Algorithm

Now in this case the logic of the AI is pretty bad at certain things, it can't see past one square and as far as I can tell it doesn't understand superbonuses.

Most importantly, the AI can't predict anything either, so it doesn't think to itself "the player might move there, or the enemy might stack here" it just looks at the turn exactly as it is, it doesn't check incomes from players, anything.

Ok, so the AI takes over on turn 15. Now, it can't "see" that you have most of Turkey, it's a) a super-bonus and b) it can only see you in Mersin. Thanks to your surrender-monkey the AI is in also already in control of Mugla.

Ok, so the AI only looks at the state of the game as it is that turn, and only 1 territory ahead, it sees in you Mersin, so it ignores it, as you can clearly see. Then from Mugla it sees that there are 2 neutral territories that are part of Antalya that it is already in, so it starts to complete that bonus from there. (The fact you attacked the Antalya territory was your mistake, you should have seen that the computer was already in it) The AI also completes Crete.

(Now I don't want to split hairs here, but with the enemy already in Turkey, it was never going to be efficient to try and complete the bonus, also seeing as it would take you at least 6 turns at this point to complete even without resistance. Moving into Sicilia was also an error by the same reasoning)

Right, so turn 16. You can see that the AI doesn't attack you in Mercin again, seeing as you are there and it doesn't own any of that bonus. It completes Antalya, then starts on Ankara (which it cannot see you are in, remember.) It also starts working on the other 2 bonuses in Turkey, as it can't see that you started in them.

(Btw, your best bet here was just to head straight for Istanbul, and not try and take the bonuses)

The AI moves into Malta because it's empty, also the AI absolutely loves taking neutral territories, it's programmed to favour them (the logic is that without an enemy in it there will be less resistance)

I'm not going to go into detail about how the AI interacts with enemies, I can if you want however but I don't want to make a long post (lol at the length of this already...)

Turn 17, AI continues completing Sicilia (can't see you in there yet) then starts to complete Izmir (no reason for it to think you would want it, you're not in it and it is (it can't see super bonuses) Now, here's where it gets interesting, the AI attacks Usak in an attempt to take it (it's neutral and doesn't know you want the bonus, even though you're in it, the AI is also already in that bonus as well) but you've already moved there, so you get caught in the cross-fire. Then its last move is a transfer to get its troop up to the front lines, which transfers to you instead because you took it beforehand.

Turn 18, even though you surrendered, that doesn't actually happen until turn 18 starts to play out, so the AI acts as you were last turn. You can see that it doesn't attack you where you were in Sicilia, but only the 2 neutral territories it can see (it doesn't know you want the bonus, and you're both in it, it's not attacking you where you were, as you can clearly see.) The only thing it attacks in Turkey is the neutral spot, it doesn't even try and retake Antalya off you. In fact I can't see any example in this entire game where the AI has deliberately attacked you... And where it did happen it was an accident that you could have avoided. Just always assume AIs will take neighbouring neutrals, and don't bother trying to take so many bonuses next time, that's one of the reasons you lost, nothing much to do with the AI.

I'm not sure why the AI is set to 'attack team mates true', but I can only assume that it's so players/AIs don't accidently block the AI in. It would seem to make more sense to not attack team mates, but there might be a reason it was programmed this way. It could even be an oversight, but it's most likely to help the AI take over bonuses it needs, and to be honest there's no reason for you to be fighting over a bonus with an AI, I see it all the time with players though but it's usually very inefficient in a team game to do that except for very key territories.

The AI also has "some" randomness programmed into it, so it's not always 100% predictable, that's deliberate else it would be too easy to play against (I know, even easier lol...)

By the way, you can look at the AI's orders before you make your moves. You can just click on it's name, then "View Teamate's Orders". I realise that in a tight real-time game on a large map that it's not always practical to check, but hopefully with understand of the AIs logic you don't have to :)

In my opinion the AI desperatly needs recoding for the modern maps, as it probably not worth using in most games.

I hope that fully answers your questions, feel free to ask more, I would love to hear other players' perspectives on the matter as well :)

It might've gotten out of hand, but the whole point of this thread was why the AI was set to attack teammates true when taking neutrals close to teammates. Which still seems illogical to me

One guess:

When two AIs are on a team, they need to be able to conquer whole bonuses rather than share them (and thereby not benefiting from the production).

The AI code uses a "attack teammates" line in order to make this work: the AIs can now conquer bonuses that they share with their friends.

Probably not an ideal fix, but could be a whole lot better than the alternative.

For instance:

AI 1 has almost completed an important bonus, except for one territory.

AI 2, his teammate, cannot see AI 1's armies. He moves in to grab that one remaining territory.

Scenario 1: "Attack teammates" is false, so AI 1's armies get transferred to AI 2. (Turn 1) On the next turn, AI 2 transfers those armies back to AI 1 or moves them out. (Turn 2) Finally, AI 1 takes the territory and benefits from the bonus (Turn 3).

Scenario 2: "Attack teammates" is true. AI 1 has a chance of winning the fight and gaining the bonus. If this happens, he got the extra production three turns earlier than the poor AI in Scenario 1, above.

Incidentally, I don't know if AIs are "blind". I think, rather, that they can see through all fog (they don't attack large wastelands in Dense Fog, for example), and just aren't programmed to take advantage of that to "seek out" opponents.

I've seen them ignore bonuses that have wastelands in them, for example, even though they shouldn't have been able to "see" the wastelands.

I could be wrong, of course.

Mostly "what he (Fleecemaster) said," but with a couple of different points/opinions:

My *hunch* is that the AI is set to attack team-mates when it attacks neutrals is to favour aggressive expansion when there's a team of multiple AIs.

Basically, if two AI teammates are trying to complete the same bonus, they will continue to do so until all the territories in the bonus are taken. Then, whichever team mate has the majority of territories in the bonus is considered the defacto 'owner' of that bonus. But this calculation only applies when *all* of the territories are taken by allies. So, if there are any neutrals remaining, they will continue to compete for them.

So, imagine if attack team-mates was turned off: In that case, whoever takes the neutral first would win it, which means that AIs with smaller incomes would tend to win neutrals (because they would have fewer moves and thus tend to be the first ones to attack). That *might* seem like a good idea, except that smaller income also means 'takes longer to complete the bonus'. So, if instead attack team-mates is on, then the stronger, more aggressive AIs will tend to win the neutrals, completing more of the bonus faster, and essentially expanding faster and more aggressively.

It's debatable whether this is actually a better strategy, but at least it kinda seems to have some sort of reasoning behind it.

On the other hand, it might simply be an unintended oversight that never got fixed. As I've read somewhere from Fizzer (perhaps the wiki), he's reluctant to make any changes to the AI at this point because according to him it would 'invalidate' all the Single Player gold star achievements. Personally, I think that's a cop-out since AIs could be a pluggable feature, like the different rounding and luck modes that are user-configurable, but that's the official 'reason' not to 'fix' the AI, as far as I recall.

Lastly, I would strongly disagree with Fleecemaster regarding whether to compete with an AI ally over territories and bonuses. Generally speaking, unless the AI's income is *huge* compared to the enemy, they are by far more of a liability than a help, especially when playing more experienced players. Consider this: If *you're* not gobbling up the AI's territory, that just means you're leaving it all to the *enemy* to gobble up.

The trick is to gobble it up in a way that *prevents* the AI from trying to re-take the territories back from you.

Here's what you could have done differently in that game to successfully take those territories: First, predict where the AI is likely to attack. Second, either take those territories early with more armies than needed, *or* take those territories later than the AI's moves, so that the AI takes early and you take it later. In both cases, you should ensure that 'attack team mates' is on for your attacks.

Additionally, there's one other important 'trick' you need to remember when gobbling up allied AI territories/bonuses: If the AI owns a majority of territories in a bonus (where all neutrals are already taken, and no enemies are present), it will try to take allied territories in that bonus with a 2v1 attack. *However*, it will *only* do this if *none* of the territories in that bonus contain any extra armies (i.e. all the territories have a '1' in them). Therefore, if you want to prevent the AI from trying to recapture the territories you are gobbling up from it, you should simply attack them with an extra army or two so that you leave more than 1 army in those conquered territories. For example, attack the AI's 1 with 3 armies instead of 2, so that after capture there will be 2 armies in the captured territory rather than 1. The AI will not try to recapture in this case, and you can proceed like a bulldozer, eating up your AI ally's bonuses.

The only real reason to keep a weak AI ally alive, IMO, is for card pieces.

Edited 9/13/2015 07:04:47

Thanks for the input guys, those were both nice reads with lots of insights :)

I haven't done enough testing with the AI and dense fog, but what you say there Poireau, about being able to see everything, like about the wastelands in bonuses.

It might be that the AI looks at bonuses that it's neighbouring, rather than the whole map, because in a custom scenario I made I had trouble getting the AI to move towards the only positive bonus on the map, which was quite far from it. Maybe there were other factors in that though, there are certainly some odd behaviors I see, but I don't feel the AI is good enough to be "worth" testing properly.

wct, I mostly agree with what you're saying about "gobbling" up the AI, and I've certainly done that myself on occasion in games, but I think that it's only in highly specific situations that it's worth it, and generally players should be focusing on the enemy, rather than grabbing land off of what is essentially an ally. In the above example I would have said that the turkey bonus should have been secured at Istanbul before attempting to attack the AI, or even to focus on going over Turkey through Ukraine (which the player hardly did at all). There was still a lot of other land to expand into on that map, in fact with greens help I think they could have secured Ukraine, expanding into Turkey makes no sense to me.

If memory serves me right the following is true:

• AI always plays with light fog.
  
• AI will never attack a teammate if there is more than 1 armie there, BEFORE deployment.
  
• AI, when facing a game with super bonuses, has no logic that one bonus is part of another, so the bonuses are treated independently, resulting is "strange" behavior. It may try to complete a super bonus, and not the small one inside, for instance. Basically the AI makes more sense when the map don't include superbounuses. (Note: The AI may for instance try to complete a 0, or even negative, bounus.)
  
  Take a look at the wiki, if you haven't already.
  https://www.warlight.net/wiki/AI

Edited 9/14/2015 00:57:03

Yes, I believe you make very good points there, Fleecemaster, especially about that specific example. I was probably too emphatic, but I really only meant that it shouldn't be considered a hard and fast 'rule' to not bother competing with your AI allies for bonuses.

TBest's comments also are accurate as far as I recall, except I will clarify one point:

In terms of fog, the AI seems to be in a fog-class of its own. It *does* see how many armies are in adjacent territories, and it also *does* see whether they are neutral, allied, or enemy. However, it pretty much does *not* see anything else except what is immediately adjacent to it. So, to say it plays with 'light fog', while perhaps technically true, doesn't quite highlight enough the fact that it is extremely short-sighted.

As a result, even in heavy fog, the AI will *always* attack (if it does attack) with twice as many armies as there were in the defending territory before deployments.

The reason it might seem to be 'ignoring' wastelands is simply because it does not have enough income to attack with double the wasteland size. E.g. if its territory A is next to B, a neutral with 2, and C, a wasteland neutral with 10, and it has less than 24 armies to allocate to territory A, then it will not attack the wasteland. If it has at least 4, then it *will* attack the neutral 2.

Another thing to note. If the AI has *more* than twice as many armies to attack with then it will attack with *everything*. The pattern is that almost all attacks from an AI territory will be 'doubling attacks', and one attack will be a 'remainder attack'.

E.g. if AI territory A is adjacent to 3 enemies/neutrals of size 2, 5, and 10, and the AI can allocate 18 armies to deploy to A, then it will end up issuing 2 attacks. The attacks will be either: One 'doubling attack' with 4v2, and another 'remainder attack' with 14v5; or one doubling attack with 10v5 and another remainder attack with 8v2. I'm not sure if there is a rule governing which of these combos will happen. I have a hunch that it actually chooses randomly where to lump in the remainder armies. Note that under no circumstances will the AI attack the 10 stack, since it can't deploy 20, only 18.

One useful consequence of this pattern of all-out-attack, where the remainder is always sent out with one of the attacks, is that the AI always leaves its attacking territories empty and vulnerable if you can use enough delay moves to attack them after the AI's moves. This pattern is so predictable that you can use it to defeat pretty much any AI, no matter its income or existing giant stacks. All you need is a little bit of income and a lot of patience with delay moves. It's also very useful to know when gobbling up allied AI territories, again to prevent the AI from retaking. This pattern is also exactly observed when playing in dense fog, so you can even use it to defeat a strong AI in dense fog. It can see your armies, so it will attack with 'doubling', but because you know how many armies were in your territory at the start of the turn, you can exactly predict how big the attacks will be, turning its fogless sight against itself.

Oh, yes, there is one *slight* exception to the 'doubling' rule, which is when an AI territory is sending out multiple attacks, for some reason, *sometimes* the AI will shift a single army from one attack sortie to another. So, instead of attacking four neutrals with 4, 4, 4, 4, occasionally you will see it attack them with something like 4, 5, 3, 4.

This is especially annoying when it is your own allied AI, and it happens to be attacking territories with 1 army in them. So, instead of attacking 2, 2, 2, 2, sometimes it will attack like 3, 2, 2, 1, and that 1v1 attack is almost always totally useless, especially in 0% SR games. Not only does it waste an army, but more importantly it wastes a *turn*, especially if trying to finish a bonus. That's one of the big reasons I find the AI allies to be almost counter-productive since they are so inefficient with expansion and/or killing off the enemy.