macger

Join Date 07/2012

AI unfair settings need a revision

I played some 1 vs 1 maps several times as TEC Enclave vs AI-TEC Enclave/Advent unfair, agressive. Some maps can be played extremely close (f.e. Cynosian Rift-AI-TEC Enclave, random 1 vs 1 map) but others (Power Struggle, Return to War, Cynosian Rift>AI-Advent Wrath) are impossible to play.

The AI settings are not well balanced in my opinion: the fewer planets between the two home planets, the less playable these maps are. Basically, you are flooded by the AI very early on with an insane number of ships. “Unfair” should be very difficult as opposed to the ‘hard’ level, but at the moment the unfair settings usually lead to an unplayable soase 2.

The background to this is that the unfairAI basically receives additional permanent global resources of +5 credits/sec, +2 metal/sec and +2 crystal/sec. But that's not all: the AI receives an additional at once permanent bonus of +0.5 credits/sec, +0.3 metal/sec and +0.3 crystal/sec on each conquered planet/astroid - already without upgrading from commerce or mining.

Who is surprised that a game like this is simply abandoned? That's not "unfair" in my opinion, but nonsense for playing: owning 4 planets/astroids, the Ai gets + 3900 credits, + 1680 metal and + 1680 crystal resources within 10 minutes more than the human player, which is way too much and certainly leads to unplayable matches 1 vs 1. (btw:You can naturally forget to play the ai-level “Nightmare” 1 vs 1!)

I think, the unfair-settings need urgently a rebalancing and a revision.

sins ii feedback

Advanced Search

Watch this post

Do not email me updates for this post Email me updates for this post

Successfully updated karma reason!

Thank you for your feedback, A.I balancing and difficulty balancing is an ongoing endeavor which will be tweaked over time with multiple updates. I'd be interested to see what others opinions on these settings are within this thread

Successfully updated karma reason!

Have you considered scaling difficulty? The AI starts with no bonuses but gains bonuses (to income, research rate, build rate, etc.) over time, scaling up to a maximum after some time. This way the ai is less overwhelming at first and can present a more consistent challenge to a player.

It would also help with another problem I've noticed. The Advent economy starts fairly mediocre but ramps quickly with powerful techs and planet items. Combined with the huge starting advantages the AIs get the end result is advent AIs being an absolute terror in the early game, maxing out their fleet almost twice as quickly as Vazari and TEC AIs, but falling off quickly. By comparison I've found that Vazari AIs seem to benefit the least from their early advantages (because your large starting bank of resources dosnt seem to get modified) and the early economy being bad and ramping slowly.

This also has the effect that Vazari is the weakest against the AI currently. The Vazari have lots of starting resources and the capacity to get two more capital ships before anyone else can to make up for a weak econony. But the massive ai advantages make beating them early impossible. Vazari starbases are also available much later and their weapon and armor upgrades are available even later, so rushing to a chokepoint and fortifying it (the standard strategy for beating the ai) dosnt work, especially on smaller maps.

Scaling difficulty would fix all this. It would make the AI more manageable early on and keep the Vazati from being completely overrun. It would keep Advent AIs from being so overwhelming early on and keep all AIs from falling behind later on (it's not hard to overtake even the hardest AIs later on).

Tldr; scaling AI difficulty did great things for Stellaris single player, it should be adopted here too.

Successfully updated karma reason!

@Lewytz2640

Based on the results of my analyses I cannot agree with you on the subject of “scaling”:

The AI starts on the home planet immediately with the described global resource bonuses. If the AI has colonized another planet/astroid, it also starts immediately with the described resource bonuses, although neither mining nor commerce have been upgraded at the start. These resource bonuses are a base amount on which the mining/commerce upgrades are then based. In addition, the AI cleverly combines the construction of planetary objects (e.g. global military research center etc.), which the human player often forgets.

The AI does not generate any further increasing bonuses during the course of the game, neither for income, building, research nor for faster action execution. This is all exactly comparable with the possibilities of the human player. However, since the resources are the basis for research, building, etc.), only the AI's action speed plays a determining role: and this gives the impression of seemingly increasing bonuses.

Here is my explanation for this impression:

Unlike the human player, the AI makes faster progress working without losing time. Immediately after completing an action (e.g. research, building), the AI starts the next action. The human player can at the beginning achieve almost the same speed by permanently stopping the game quickly and completing the planned actions. BUT: the more extensive and non-time-compliant the actions are, the more the human player loses time. And this loss of time, or the time gained by the AI, is multiplied because the AI can carry out the actions on all planets simultaneously. In this way, the AI gains an ever-increasing lead in research, ship building, etc. over the course of the game because there are sufficient resources available due to the resource bonuses.

This is why the AI settings for resource bonuses are in my opinion the crucial point.

Successfully updated karma reason!

Suggested settings for unfair-ai bei 1vs 1 game

I have now played the 1 vs 1 map “Return to War” (tec enclave vs tec enclave unfair, aggressive) several times with different unfair-ai settings (in this map both home planets are only 1 large planet away, which makes playing very difficult, the other planets are symmetrically distributed). The following settings for unfair-ai have resulted for me as settings for a challenging, exciting and very difficult game:

global empire resource bonuses + 2.5 credits, + 1.0 metal, + 1.0 crystal.
global planet resources bonuses 0.0!
update-interval 2.

With these settings I could win this map with skillful and balanced play before KI could build a titan.With some mistakes you lose quickly.

An even greater challenge and, in my opinion, a recommended real (and not impossible) unfair settings are:

global empire resource bonuses + 2.5 credits, + 1.0 metal, + 1.0 crystal
global planet resources bonuses + 0.25 credits, +0.25 metal, + 0.25 crystal
update-interval 0.

Note: if the home planets are separated by more than 2 planets, the map is just "difficult" because then the player gains some time.

Successfully updated karma reason!

I think you have misunderstood me. My comment was in reply to @ramastardock and was a question to the stardom team as well as a suggestion. I know that the ai does not scale in difficulty with time. I am aware that the ai has many resource bonuses natively, both from the start and on colonized worlds. My suggestion is that scaling dolifficulty would allow for higher ai difficulties that allow for a more consistent challenge.

The problem with static ai difficulty is that it just turns every game into an exercise in survival. Eventually the human player overtakes the ai and wins as long as they survive the early onslaught. Ai bonuses scaling with time allows the bonuses to be much higher when fully applied without making the early game impossible on small maps. This would lead to a more consistent difficulty curve instead of an overwhelming ai on the first 40 minutes which becomes a pushover after an hour.

Successfully updated karma reason!

Ah, I see. I fully agree with your “Ai scaling” suggestion, especially for larger maps. In my opinion, there should also be a scaling depending on the maps: lower AI bonuses for small maps with few planets, higher AI bonuses for larger maps.

A few more thoughts on AI:

Basically, I see that the AI progresses quite cleverly, especially in the first 20 minutes, both in research, fleet construction and orbit building on the home planet as well as in the rapid colonization of planets. But I noticed 2 serious AI weaknesses:

1.Colonized planets are not immediately developed further via mining/commerce and logistics, thus lacking solid resources later on. This results in a significant weakening of the AI after lost battles (after approx. 40-60 minutes), which is almost impossible to catch up on.

2. The AI often continues to fight in battle even when the human player is totally superior and thus loses most of the fleet, which also leads to a weakness that is almost impossible to catch up with.

So my suggestions are:

1. improve the action code so that the AI develops colonized planets quickly.

2. if the opponent is significantly superior, the AI should retreat immediately.

3. when comparing combat strengths, the strength of the stationary defenses of the attacked orbit should also be taken into account for the battle.

4. for large maps, the strength of the AI should start small and be increased over time.

Successfully updated karma reason!

I accidentally made a new post because it's been many a moon since I've been in a forum, but I'll post it here and mold my ideas into a proper response too so it's all in one place.

I personally love where the AI is at right now in terms of outright difficulty. Nightmares are hard but beatable if you keep your wits about you, and impossibles truly feel impossible, but can still be beat. There wasn't an announced patch on 12/30, but my friends and I feel there was some tuning done to make them a step more aggressive and split fleets more which is even better!

I completely agree with @Lewytz2640 suggestion of a scaling AI, but I feel that it should be tied to the "defensive" sub setting. That way, if you're fighting multiple AI and you leave that on random, an AI set to aggressive have this wildly oppressive fleet early that we currently see, and another could be scaling/defensive, and have tons of star bases full of items, a huge fleet, capitols with items, etc.

Both types of AI though, need to scale into late game better than what they currently do. That's where I'd love to see the AI improve. As of right now it feels like once you're over the hump if you will, the AI just give up. I've even seen them retreat comparable fleets from their home world as you're bombing it!? Early to mid game you'll be entrenched fighting and fighting and fighting and you win the battle, and that's it, you just walk them over. I'd love the AI to be able to fall back (not dive deeper into your space which they tend to do often), rebuild, and regroup (enclave even have research specifically for this contingency) and defend their home world as you push into it. I'd also love to see the AI start prioritizing specific targets more based on their fleet comp and yours, if that's even possible, capitols feel far too safe at the moment. I feel that AI don't prioritize building items on starbases, capitols, or titans, which leads to my group of friends not fearing those fleet components even against nightmares and impossibles because you just push them over.

At this point, we're playing on bigger maps with 4-5 AI vs 2-3 of us just so we can enjoy those massive battles that are what I feel make sins amazing.

I'm sure, I'll also get some people saying, well just go play against players, I tried a couple games, the first was a 2v2v2 where we got teamed (it was blatant and obvious not just a coincidence), and the last the enemy just built 28 devastators and a titan and wiped fleets. If that's how you want to play, good on ya, but I'd rather just enjoy the typical style of gameplay with a slightly smarter AI.

Successfully updated karma reason!

I assume that your statements about the Nightmare/Impossible difficulty level were only made on the basis of large maps with several AI opponents. According to my findings, there are 2 important factors that influence/define the level of difficulty:

the number of planets/astroids to go to your opponent,
the number of planets per player.

For a noticeable effectiveness of the difficulty levels from “unfair” upwards, in my opinion, the following are required:

max. 3-4 planets/astroids to go to your opponent,
max. 3-4 planets/astroids/player.

More planets/player and more planets between the home planets reduce the difficulty of a level, as the player has enough time to build his empire. Because 4-5 planets and more are enough to build a strong core of an empire. The rest is then just more a question of playing time and winning big battles, and that's just too easy, no matter which difficulty level was selected. That's why I think scaling (agressive/defending) is especially important for large maps.

I completely agree with the following points (and this applies to all maps):

it is not good programming that the AI practically gives up if you are “over the hump”.
AI only builds up little defense, only when the game runs for a long time, then it gets a little better by building starbases, but these are not expanded.
I also think it's a pity that the aggressive AI is actually a stupid AI, as it is mainly focused on building a lot of ships and attacking relentlessly, ignoring both the qualitative improvement of the ships and a solid defense.

I admit that I like to play small maps because - in my experience - the small distance between the home planets (max. 3 ways to go to your opponent) and the few planets/players (max. 3-4 planets) mean that the challenge is greatest due to the respective difficulty level.

Successfully updated karma reason!

Incidentally, after numerous matches, I rate the AI action programming as simply - to put it nicely - inadequate.

Here my explanation: AI only has considerable quantitative advantages (increased research, increased shipbuilding), that's all, hardly any reasonable defense, constantly repeating similar attack routines, no qualitative reaction to what is happening. In my opinion, this is too simple and quickly coded.

Successfully updated karma reason!