Some observations about the AI that I think should be fixed

First:

Even if the AI has 3 capital ships and an good sized fleet they refuse to attack an Pirate Base. This is normally not an issue, but sometimes random maps place an Pirate base between the AI and the rest of the map and because of this bug they AI is trapping it self into it's starting area.

Solution: Make the AI not be afraid of Pirates and have them attack them without mercy.

Second:

This one is kinda funny but is annoying, If you have an two AI game with no other players the AI will ally with each other making the game never end. >.<

Solution: Don't let the AI ally with each other if the game has less then 4 players

Third:

The AI seems confused alot of the time, randomly sending forces back and forth between to planets. It seems to also not be able to stay on one objective for long, because it will sometimes have an massive fleet ready to jump but an enemy scout jumps into the system and the AI sends it's ENTIRE 100+ ship fleet to go kill that scout. And then the fleet sits and does nothing for 10-15 minutes.

Solution: Have the AI create goals and get them to focus on achieving that goal as quickly and efficiently as possible.

Fourth:

The AI seems let it's fleets idle for ridiculous lengths of time. Because of this matches can take insane lengths of time just to end.

Solution: See number 3.
27,826 views 15 replies
Reply #1 Top
2nd: You can also win if all players are allied with each other, I think. I know I've won when allied with an AI.
Reply #2 Top
An effective way to make them move is to bomb one of their "important" planets. Like the home planet or one that generates a lot of income.
Reply #3 Top
And if you don't bomb the home planet: just played an ai where I wiped out the home planets entire infrastructure while the ai was distracted fighting some ice planet of mine elsewhere ;)

It seems that whoever programmed this AI: 1) Didn't build the AI to be adaptable or 2) Built the AI to be adaptable, but failed to throw in a module for pirate base destruction. or 3) Made a stupid, scripted AI, that isn't adaptable, isn't modulated because its scripted, and failed to throw in a script for pirate base destruction.

I think 3) is the situation in most cases (though I haven't played enough sin games to be sure it's the case here), and it torques me off. Programming an AI is challenging, fun, and potentially elegant, and any programmer who settles for a script instead of something that can learn and adapt either doesn't have the time to do so, or is just simply unimaginative.

Note: Not sure if the adaptable AI is possible to build without a good understanding of object oriented programming, and a mathematical foundation in linear algebra and probability. The oop is needed to build complex code, and to get away from scripted behavior. The foundation in linear algebra is useful to thoroughly analyze economic data as well as defensive and offensive data for pivot points. And probability is useful for building a probabilistic model or memory for adaptability (though creativity will do just as well here).

As a further note: Linear algebra could be a good method to analyze a game before it even ships for balance issues. Just find those pivot points, and you've found the crux points upon which the game hangs.
Reply #4 Top
yeah I've found these same problems with my locked teams game. Whenever I get attacked on a planet that's like 5 jumps away from my fleet, and one jump away from my ally's fleet, I usually always get to defend my planet before my ally decided to move, even after telling him a BILLION OF TIMES OMG SAVE MAH PLANET. The AI is just slow, I could make 5 jumps before he could make one.
Reply #5 Top
It seems that whoever programmed this AI: 1) Didn't build the AI to be adaptable or 2) Built the AI to be adaptable, but failed to throw in a module for pirate base destruction. or 3) Made a stupid, scripted AI, that isn't adaptable, isn't modulated because its scripted, and failed to throw in a script for pirate base destruction.
End of quote


Neither of those.

I'm not sure why you would even consider "pirate base destruction" an important point for an AI. I don't even do it until there's nothing else left in the system, since they're my pets.

Note: Not sure if the adaptable AI is possible to build without a good understanding of object oriented programming, and a mathematical foundation in linear algebra and probability.
End of quote


Trying too hard to sound knowledgeable tends to backfire spectacularly. These are very basic things for anyone with a CS degree or equivalent (since IC is in Canada). Something like Object oriented programming is learned in beginner classes :P

Trust me. You don't know more than they do. :P
Reply #6 Top
I'm not sure why you would even consider "pirate base destruction" an important point for an AI. I don't even do it until there's nothing else left in the system, since they're my pets.
End of quote




If you had had read my post more carefully you would have noticed I said,
This is normally not an issue, but sometimes random maps place an Pirate base between the AI and the rest of the map and because of this bug the AI is trapping it self into it's starting area.
End of quote



So how about the AI is coded that if it needs to destroy and pirate base in order to expand it will attack it?
Reply #7 Top
Sorry, I was getting the other guy off his high horse ;)

I have seen the AIs eventually take over pirate bases, but it usually doesn't happen quickly. There are of course some predictable quirks since it can't be a true AI. Very likely it waits to expand its fleet until it achieves an economic marker and knows it can support the fleet upgrade. But if it's trapped by a pirate base and can't reach that market, it gets stuck in a catch-22. It needs more ships to take out the base, but it doesn't want to build more ships because its economy can't really support it.

This is my educated guess about what happens, no real easy solution. The AI can't use tactics like human players, and pirate defenses can shred a small enough fleet easily. Maybe it can be coded such that if the only way to expand is to get past the pirate base it would do something special, but yeah.
Reply #8 Top
They need to make the AI modifiable so people like Sorian can make epically hard and fun AI's.
Reply #9 Top
It's not quite that easy, you know :P IC has done interviews where they said their major limiter to the AI is processing power, not their ability/desire. Sins is unique in that its AI needs to be more of the typical turn-based type, which is known to be very good and usually brutal, rather than a basic RTS AI that is based on build orders and cheats with knowing the map and such. But while a 4X AI gets the benefit of essentially an infinite amount of time (as far as it's concerned) between player turns to analyze everything and decide what to do, the Sins AI needs to do this constantly because everything is happening in real time.
Reply #10 Top
Sorians AI has made Forged Alliance playable in single player while at the same time being many times more difficult. I'm sure there are ways to make the Sins AI much harder without impacting performance or even improving it.
Reply #11 Top
Dunno, I can't speculate about how much more Sins' AI can be pushed to the limit, I can only say what I know from the dev team :) I would imagine it can be improved, but they likely also considered people playing against 9 AIs on a big map, and the drain on system resources that would be. Limited to fewer players, I can definitely see a more challenging AI :)

But, by comparison, Sins AI is already more complex than SupCom's/FA's was.

As an aside.. maybe I should look into his FA AI mod, it's one of the reasons I stopped playing it :)
Reply #12 Top
Since you haven't played FA in a while your probably going to get your a** handed to you in a bucket by his AI. I know I did when I first tried his AI.

Also if you want to improve your performance for FA even more I also would suggest looking at Duncanes AI fixes which fix the stock AI's.
Reply #13 Top
Annatar11, thanks for all the information, I can certainly see now how real time ai's are quite a bit more challenging than the turn based stuff I've done.

Hey I needed that reality check, but I still wonder many times if any of these mathematical foundations actually get used in game programming. I'd love to try and apply linear algebra for instance in some way so as to discover new strategies and find what the best options are at different stages of the game. It just seems to be really really complicated, with in game research changing the dps of ships and costs to the player all the time. Where do you start?
Reply #14 Top
1 thing which could also be fixed with the AI is its tendensy to retreat quickly when they become under attack because nearly every time i attack the enemy AI's planet with 100 ships and they have 100 ships, they run nearly instantly from an even fight
Also when the AI attacks without scouting the enemy planet, phase jumps in and retreats almost instantl
Reply #15 Top
1 thing which could also be fixed with the AI is its tendensy to retreat quickly when they become under attack because nearly every time i attack the enemy AI's planet with 100 ships and they have 100 ships, they run nearly instantly from an even fight
Also when the AI attacks without scouting the enemy planet, phase jumps in and retreats almost instantl
End of quote



Yeah that's really annoying, and if you come to help an AI ally, he flees anyways, even if you outnumber the opponent with your support-fleet.
Looks like as soon the AI decides to flee it will flee no matter what.