1.03 AI is better than you hear complained about 8-)

I have also found the AI to be much better in 1.03. I think that many people are unrealistically expecting "competitive" behavior from the AI. Currently, the AI seems to be geared toward having a fun, long-term single player game. People would complain even louder if the AI always used the same winning strategies over and over.

Maybe Ironclad can add additional AI behavior profiles, such as a "competition level" one that mimics the timing and build order that people use in competitive play.

In a related question, has anyone worked out how 'surrendering' works for the AI? I've seen AI surrender quite early (ie, as soon as I pop their homeworld, but when they have many other systems), and others hold on to the bitter end (ie, make me destroy their last asteroid). Has anyone ever had the last AI in the game surrender to give you the overall win in a single-player game?

I had the AI surrender to me just last night. I played a leisurely 5v5 single player game with locked teams (on the biggest random map), and the last enemy AI actually surrendered with a few planets left to give me the win.

Ok... so I beat 3 random AI's on the toughest level on a medium sized map by myself. I locked the 3 of them on the same team. The AI is just bad. I'm sorry to say but it is true. I don't blame the devs it is hard to write AI for a game this complicated.

I am wondering if anyone knows which AI is the toughest to beat? Aggressive, Defensive, etc.?

The AI is better in 1.3 for sure. But it makes one huge mistake over and over: It still is very bad at knowing when to stay and fight. It does not "seem" to matter if it's last viable planet or not, or take into consideration economics, or if your atacking with your only fleet.

I just finished my 5th game vs 1.3 AI, and it had one planet left (and a couple asteroides). The last planet was it's homeworld; maxed out with every building and defence. I figured I might lose the battle, but would have a second fleet come in to finish the job. When I jumped in, he retreated his entire fleet to the asteroid??? So I detroy all buildings/defences, and THEN, when I start to bomb his planet, he comes back with the exact same fleet, and THEN fights my fleet. I've seen it do that many times; it only seems to figure out how much it's planets are worth after you've destroyed all the buildings around it and start to bomb it, then and only then does it seem to say "HEY, I think I want that planet protected after all".

Yeah, it seems like they didn't spend much time on behavior modeling (decision trees, etc) for the AI. Currently, the AI seems to be purely reactive. It will attack/colonize whatever planet is the weakest/closest, and that is about it.

The only time the AI almost beat me was when I was playing a game long enough for the Vasari AI to get RA. Has anyone else seen this happen?

Anyway, I still have fun with the game despite the flaws. I lost track of time the other night, and had I had a "Civ moment" where I heard birds chirping, and I turned around to see that it was light out...

On a similar topic, not only does it have to have decision trees, the decision trees need to be able to cope with the actions of other players.

Suppose the AI is investing in research, but the bounty on it skyrockets. Does it abandon the research and up the bounty on some other player, or abandon the research and beef up defenses, or just keep investing? It has to factor # of planets, sizes of fleets, location of enemies, phase shift points, etc.

Truly insane.

Oh my gosh. This is just what happened to me the first week I got Sins. I'm thinking "why in the heck is my neighbors floodlights on?". On closer look, it was the sun comming up. . .

Come to think of it shanks, I don't remember the AI ever attacking any of my planets, any of my games ever, except with small fleets of Siege Frigates on unguarded planets.

monkey and scalding, I actually contacted yarlen and blair over 7 months ago about using machine learning algorithms for sins a.i (part of my thesis is ML in RTS games), but they didnt want to release the a.i code for obvious reasons.

There have been far too many threads that have been unfairly and hysterically critical of the AI in 1.03. I just finished an 9 hour game, large random map, with 8 hard opponents (FFA). I did not throw in the towel when my home planet was attacked simultaneously by three Ai's. I survived it, then really worked on those missions. I managed to get a cease fire with all but two of the AI's. Unfortunately, not the most powerful ones. I knew it was not possible to "win" by conquering the galaxy...but I managed not to be the first civ to die.

I'm a paradox game fan, and I like to set my own personal victory conditions. In this game, I merely didn't want to be the first eliminated.

This AI plays better than most I've seen in the last 20 years- not counting some Chess programs.

If you can't find a way to challenge yourself with this game, you're not trying hard enough. Showing off about cheesy victories is pointless, and boring to most readers.

The AI is much improved over 1.02. Fleet mix up, defense, research, and planet upgrades are all vastly improved. However, the when to stay and fight logic is still messed up as others of stated. It doesnt put enough value on "property" and doesnt seem to take into account the ability to resupply his fleet during the middle of a firefight. It is also poor about building ADDITIONAL frigate factories at its frontline colonies. It will often have just the one at its homeworld even on large maps.

Does anyone who plays online any great deal find the AI a challenge? If you have ever played a team game where someone drops and an AI takes over you know there is almost no advantage sufficient to stop it snatching defeat from the jaws of victory. Now that I think of it, it actually does a pretty good job of emulating a bad, kind of dickish* human player who will absolutely lead to your team losing the game.

1 Its macro play is wasteful and it does not understand how to manage supply lane length, speed of production, planet logistics synergies, which planets are more valuable in a given system, which planet another empire would be most badly placed after losing, etc.

2 It is easily manipulated into attacking or withdrawing from a fight and it never reinforces itself, so it is trivially easy to set it up for a fall (and it never learns that a place is dangerous even after continuously losing). It needs some better and wider heuristic method to understand what makes a planet or opposing fleet dangerous to another, rather than what looks and acts like a raw sum of HP/Attack/Defence.

3 It does not understand hitting rear areas or sending raiders through to hinder reinforcements in transport to the front. It does not understand blockading trade lanes. It does not understand which planets have the maximum discrepancy between value to another empire and their defensibility. It pays no attention to economic warfare.

4 It never seems to assassinate capital ships when it has the opportunity. It does not press or chase one when it has the advantage and does not seem to understand which unit combinations pose it the most danger in an opposing fleet.

5 Really the only dangerous thing it does is on 'unfair' where sometimes it will send along 20 cobalts with a siege frigate followup as soon as it can crank them out and send them through, but it is so bad at rushing that you will usually have repair yards/lrms researched by then anyway.

* did you know if you are on a locked team with an AI, and it loses a planet with buildings remaining, and you take the planet, it will never scuttle them and you can never destroy them? Seriously, it's a dick player emulator.

It seems like the 1.03 AI is greatly improved, but still needs tweaking in a few areas. I hope Brad and the IC guys are reading this thread, as you guys have really listed some constructive feedback regarding AI behavior improvements. Make sure you voice some of these on a 1.04 thread when it comes up.

I just had a problem with the AI. I modded a Sova Carrier to have the Phase Jump Disruption ability, but modded the ability so that it would completely stop phase jump at the cost of being unable to defend itself i.e. no weapons and a 1000/2000 hull/shield split. I also modded most of the caps ships and beefed them up. My problem came when I sent in a Carrier and two Colonys into a planet well. The enemy force was sixty 10-damage ships and one cap ship with a 45-50-75 damage per weapon and 2000/2000 hull/shield stats. I maneuvered the carrier so the disruption range would prevent the enemy force from phase jumping to a safe planet. My Colony ships do 45 damage per weapon(3), with a 1500/2000 hull/shield stat. I was about to retreat because I figured if the AI focused on one, I would easily be taken down before I could make a dent. Instead, the AI started to retreat, but obviously couldn't because of my Carrier. And they just sat there for literally 5 minutes as my Caps slowly but surely eliminated them. They didn't even return fire. This wasn't the first instance. There have been other times when my fleet was much stronger and theirs weaker, but still.

I think the AI should at least be able to determine for itself that it can't leave, therefore it should either a) attack the object responsible for its inability to phase jump or go down fighting.

I just hope the devs will fix this somehow. I know for a fact that 60 ships and a cap can take down at least one ship. If they focus fired they could have destroyed my carrier in a couple of seconds (about 13 volleys), allowing them to retreat. As a side note, it would have taken 14 volleys from Colony ships to take down just their Cap and another volley for each other ship. The AI could have easily won that battle.

I actually caught the AI doing something clever.

I was Advent, AI was TEC. I was chasing one of his fleets which was in headlong retreat. After chasing them through two systems, they finally stood their ground and fought.

In a magnetic cloud.

I'd never fought a battle in a magnetic cloud and didn't what the consequences of it were, but I found out quick: special abilities don't work there. And that's the Advent's bread and butter.

I still won the battle because I outnumbered him handily, but I took far more losses than I would have if the battle had taken place anywhere else.

I'm asking myself against how many teamed unfair AIs you can win solo?
Today I won against a 2 unfair AI Team (Advent/Vasari) as Vasari on a small 3 slot map.
Against 2 was hard but I think to win against 4 isn't possible...maybe on a big map.
Has someone tried to fight 3 or 4 teamed unfair AI's?

Last night I had an AI player send a reasonably large, well designed fleet deep into my territory (bypassing several fronteir planets) and fight long and hard to almost destroy one of my planets.

Of cource, once the battle swung obviously in my favour it ran like a little schoolgirl, but I had to have fairly overwhelming force on hand before it took off.

I seriously don't think you can call this a 'problem' with the AI.
You modded the game to alter the rules the game runs by, but the AI is programmed to use a different set of rules.

I am actually curious how you achieved a complete disruption of phase lane?
I was considering the idea myself a little while ago while wondering how they would make an indertictor (name? been a few years) cruiser for a Star Wars mod.

I mean, is it just that the phase engine charge time is set to a really really high number, or is it set to something else which just never occurs?
From the way I figured it would be implemented, the charge time would be a numeric figure, and a ship sits their in 'charging mode' until it hits the numeric value, at which point it jumps.

I've never actually looked at modding sins either, so don't know offhand where to look for this information. My thoughts just stem from being a Software Engineer.

On the Original Posters topic however:
I've found the AI to be improved from my first games.
I am currently playing through a Huge map (5 stars, about 50 planets or so?) with 9 Hard AI's on unlocked teams, no Pirates.
I have found that defenses are actually present now on planets, and AI fleet composition is quite varied. I know a few times I had to change my own fleet composition to take into account what was being sent against me.

I don't think i'm the best player in the world either though given I never really played RTS games, but I know pre-patch 1.03, I didn't face the same challenges I have faced in this current map.

I added the buff from the Ion Blast that disables phase jump to the Hyperspace Disruption buff and then gave my carrier the Hyperspace Disruption ability. The range on it is relatively low and they could have easily moved outside of the radius also to retreat. Or like I said they could have just destroyed the ship and then retreated. I don't know how the AI thinks, or how it's supposed to think, but shouldn't it at least have some "fight or flight" instinct and when it can't 'fly' it should fight.

Hmmm... I do see your point in that the AI should know how to deal with different situations, but I guess it will be expanded in time to cover alternative scenarios...

The behaviour I have witnessed in-game from ships charging to jump in the presence of a phase jump inhibitor is that they will just sit there waiting to jump until they finally charge up and get out of there.
Similarly with a ship hit with an Ion blast, it will sit there waiting to jump when it wears off (well, it has no choice really, the craft is completely disabled until it does).

I guess the thing is the AI is currently programmed that if a ship is trying to jump out of a gravity well it will sit there trying to do it until it succeeds, because anything that is effecting it's jump ability was designed to be temporary, or simply extends the time to charge for jump.

Technically, a jumping ship in effect of a phase jump inhibitor should be able to move to a point outside of the effected region (if it can find one) but it currently does not do this from what I have witnessed (probably because it has already starting charging for jump, and will not break from that charging to find a better jump point if it exists).

I would imagine that a ship only knows it can't jump when it goes to jump, and it seems that if it can't jump it just waits till when it can.

As I said at the start, you are right in that it sounds like an area that could do with improvement, but i'm not surprised it works the way it does. Normally the disabling of phase jump (total disabling via Ion Blast) is temporary, and also comes with complete disabling of the craft. Also, this effect can normally only hit one craft at a time (instead of an Area of Effect thing).

That's the thing about human modification and emergent use of an application I guess... Someone will do something that is outside of the design characteristics and highlight an area for improvement.

As for the comment about "fight or flight" instance... I guess the thing is you already triggered a fight or flight instinct, which made it try 'flight' in the first place. The permanent inability to 'flight' (AOE) seems to be something that the AI cannot cope with at the moment.

The other point you make about how the craft could have destroyed the ship then retreated... I can see no manner in which the craft that were trying to jump would have been able to identify the carrier you modified as a higher priority target than any others, even if they did want to attack. Adding an ability doesn't mean the AI suddenly will be able to determine where it is coming from, or that it is not going to wear off...

I guess it is another modding question a little, but is it possible to update the AI yourself? I would suggest it would be necessary to update AI to allow it to 'see' the hazard your modded carrier poses, so that even if a 'fight' response was able to be evoked when they cannot flee, it would know what to hit first, to allow the 'flight' response again...

I can see your points. I'm still working on balancing the whole idea of a phase jump inhibitor. I wish it were possible to modify the AI, but I don't think it is. Even if it were, I would be the least qualified to touch it.