Author Topic: Difficulty for fully "smart" AI  (Read 6467 times)

Offline legionof1

  • Newbie Mark II
  • *
  • Posts: 13
Re: Difficulty for fully "smart" AI
« Reply #15 on: September 12, 2015, 04:28:54 am »
Bypassing the fog of war i don't feel is a real problem. The Ai does next to nothing with such knowledge. Normal wave Attacks are predicted. CPA's follow shortest route path from gather point to your borders. Exo attacks aim for specific targets(usually your home command) Spawn from both ai homeworlds and take shortest path between the those to meet up and from there go to target.

Offline motai

  • Jr. Member
  • **
  • Posts: 52
Re: Difficulty for fully "smart" AI
« Reply #16 on: September 13, 2015, 03:23:45 pm »
my 2 cents on the ai ignoring fog of war. the ai conquered the galaxy and seeded scanning platforms everywhere.

1. i dont want the ai giving me cloaked ship/scout warnings. very annoying.
2. scouts as a cloaked ship make it that much more frustrating in prevention of harvester rebuilding.
3. as i noted in opening line there is no explicit reason there is not cloaked an undetectable scanners in every corner for the ai. the only potential limit is the supply argument preventing communication but that would still mean your border wormhole would be visible.
      ( as a collarary one of the thought for hacking could be to hack the ai sensor network to allow vision without a scout)
4. if it allows the ai to make intelligent choices and improve challenge without crippling human strategy then its ok by me.
     ( an extension of this thought, the ai would think faster and react to a point of making look preplanned. our limitation in computing power mean that doing this processing ahead of time helps eliminate lag. always a bonus.)

Offline ZaneWolfe

  • Sr. Member
  • ****
  • Posts: 272
Re: Difficulty for fully "smart" AI
« Reply #17 on: September 21, 2015, 05:42:38 pm »
If I remember correctly, the tracker for "Ships are X hops away from what I can see" for the AI is based on SHIP COUNT, not SHIP STRENGTH. That being the case, its possible to cheese the AI via Swarmer ships. Large swarms of weak ships prevent the AI from attacking a fairly weak defensive point. Also, given the HUGE advantage the AI has of having won the actual Human/AI War, it does make sense they have greater knowledge than you. A well known fact is that the AI also has a HUGE tech edge over humanity, best seen in Core/Mark V units, and even in Mark IV units. The AI's technological edge isn't theory or just something we are told. It exists at all times throughout the game. This disparity could easily explain why the AI's scanners can pick up the signatures of mobile units much further than humans can. While the issue may counter one of the core concepts of the game, it fits well into the spirit of things, and the lore. Also it NEEDS to have an ability like that, otherwise it will get cheesed out every time by people pulling their ships out a single jump and then instantly jumping back in when the AI strikes. Since the AI isn't programmed to remember you did that last time, it would be an abuse of the game that AI War is supposed to prevent in the first place. (IE simple cheese tactics should not work perfectly every single time.)

Offline Elestan

  • Full Member Mark II
  • ***
  • Posts: 158
Re: Difficulty for fully "smart" AI
« Reply #18 on: September 23, 2015, 02:21:54 am »
While I still disagree on the AI 'cheating' issue, I'm going to avoid detouring this thread further away from my main point, which was that the game would benefit a lot if the AI could think beyond straight-line paths from wormhole to target.  This deficit makes some fairly degenerate defense strategies dominate, and makes it necessary to give the AI a ludicrous number of ships to throw at the players in order to give it a chance of breaking through.  A little more tactical smarts could allow the AI to be equally fearsome with a fraction of the brute force.  It's like the AI has a smart general picking the systems to hit, a smart colonel picking the in-system targets and allocating the forces to attack them, smart sergeants picking the individual targets to shoot at, but its lieutenants and captains are all idiots who only know how to do frontal charges.

Offline Captain Jack

  • Hero Member Mark II
  • *****
  • Posts: 808
  • Just lucky
Re: Difficulty for fully "smart" AI
« Reply #19 on: September 23, 2015, 11:41:36 am »
Given that it's a networked system, it's more like the AI spends a lot of time being a general, a lot of time being a colonel, less time being a sergeant, barely thinks about captaincy and just doesn't care at all about being in the infantry. Because the greatest efficiency gains are found at the top.

Lucky for you, there's an easy fix to get what you want. Under AI Options when you start a new game, go to the AI Modifiers list and select "Cross Planet Waves". Then under AI Types on the player list, you'll want the Vorpal AI and the Warp Jumper AI.

That should be it! Have fun!  >D

Offline Elestan

  • Full Member Mark II
  • ***
  • Posts: 158
Re: Difficulty for fully "smart" AI
« Reply #20 on: September 23, 2015, 12:57:26 pm »
Under AI Options when you start a new game, go to the AI Modifiers list and select "Cross Planet Waves". Then under AI Types on the player list, you'll want the Vorpal AI and the Warp Jumper AI.
I'm already using CPWs, which (IMHO) are a big improvement over the default wave behavior, so I'm glad those were added.

Vorpal just gives the AI units that are good at breaking defenses.  That actually only makes the problem worse, because it means the player has to focus an even more overwhelming amount of firepower on defending.

Warp Jumper (to my understanding) essentially allows the AI to teleport ships to any point on any system perimeter.  That doesn't address the core issue; the AI will still use direct-line paths to its targets.  Instead of a dumb tactical AI, it becomes a dumb tactical AI with an(other) unfair advantage.

I'm not saying the game is too easy in general; I'm still only on 7/7, so there's plenty of room to crank up the AI's raw power.  But I shouldn't be able to point to obvious tactical errors when I'm playing against the full-intelligence AI.  I also don't want to have to compensate for those errors by giving the AI other advantages.  I use Bouncer+Kiter, CPW, NWW, Schizo, because those options all involve increasing challenge without letting the AI do things that the human player can't.
« Last Edit: September 23, 2015, 01:42:51 pm by Elestan »

Offline Captain Jack

  • Hero Member Mark II
  • *****
  • Posts: 808
  • Just lucky
Re: Difficulty for fully "smart" AI
« Reply #21 on: September 23, 2015, 03:17:44 pm »
Nah, if I'm reading/remembering right, Vorpal's primary change is that it adds Cross Planet ATTACKS on top of the cross planet waves. The shieldbreakers are there to ensure you actually commit to the active defense necessary to hold off the attackers.

So your big issue is the AI doesn't do any maneuvering? I hate to say it, but that's completely intentional. AI War is as much a tower defense game as it is RTS. Of course defensive strategies are powerful, the game's AI needs to follow lanes. It's justified by the metaplot since the AI has infinite material resources but limited computing resources it's just more efficient to set the attackers to go in a straight line. It's the AI entities not taking you seriously.

There are ways to mitigate the effect, but it's never going to go away. The closest you can get are the AI Plots Hybrid Hives and Advanced Hybrids, as well as the Neinzul Preservation Warden and Roaming Enclave minor factions.

Offline Kahuna

  • Core Member
  • *****
  • Posts: 2,222
  • Kahuna Matata!
Re: Difficulty for fully "smart" AI
« Reply #22 on: September 23, 2015, 03:31:51 pm »
So your big issue is the AI doesn't do any maneuvering? I hate to say it, but that's completely intentional. AI War is as much a tower defense game as it is RTS. Of course defensive strategies are powerful, the game's AI needs to follow lanes. It's justified by the metaplot since the AI has infinite material resources but limited computing resources it's just more efficient to set the attackers to go in a straight line. It's the AI entities not taking you seriously.
Got ninjaed.
What I was going to say too^^

A couple of points about AI War's design.
https://arcengames.com/mediawiki/index.php?title=AI_War:Fast_Facts
Quote
12. Micromanagement is out, group tactics are in.
Don't try to micromanage your individual ships in most cases, there are too many. But also don't just move all your ships around in one big blob. Several smaller groups can create diversions, distractions, or flanking maneuvers. You can also do feints, keep artillery at a safe distance, have multiple waves or reserve forces that are held back from initial contact in a battle. This stuff really works, does not require any hotkeys, and definitely matters. If you want to beat the better AIs, you have to really use tactics.
The player is encouraged to use groups tactics so yeah I guess the AI could too. It's true AI's waves are kind of "big blobs" of meat that happily walk into your grinder if you want to see it that way.

Quote
19. Understand that AI War is not just an RTS. Really.
It controls like an RTS, but don't let that fool you. It's got a lot of elements from grand strategy, turn-based strategy, 4X, and tower defense in there. The flow and feel of AI War is unlike anything else out there, so if you find yourself foundering you might want to play the Intermediate Tutorial to have it help guide you through the execution of a number of unique strategies and tactics just mentioned in this brief guide. For a detailed guide on this subject, see Like Chess, A Game Of AI War Has Three Abstract "Phases".
AI War has some tower defense in it so there's that.
Also Making the AI to avoid defenses as you described could make micromanagement somewhat more important. It would be great if it could be done without making the game any more fiddly and micro heavy.

A little more tactical smarts could allow the AI to be equally fearsome with a fraction of the brute force.
Vorpal just gives the AI units that are good at breaking defenses.  That actually only makes the problem worse, because it means the player has to focus an even more overwhelming amount of firepower on defending.
Combine Cross Planet Waves with Vorpal AI and you kind of have that. CPWs spawn randomly on the edge of the "map" and Vorpal is "equally fearsome with a fraction of the brute force" because of the "anti defense units". Sure the ships from CPWs still travel in a straight line but you never know where they will come from. It's not like you would rebuild all defenses every time after the CPW spawns.

Warp Jumper (to my understanding) essentially allows the AI to teleport ships to any point on any system perimeter.  That doesn't address the core issue; the AI will still use direct-line paths to its targets.  Instead of a dumb tactical AI, it becomes a dumb tactical AI with an(other) unfair advantage.
You have 3 solutions to your "problem" so it seems to me that you just refuse to see their value. I know it's not exactly what you're looking for but all of those 3 combined makes the AI to
1: not fall for "Kahuna-style defenses"
2: have an "equally fearsome with a fraction of the brute force"
3: be able to attack any planet so the player never knows where the next attack will come. Which forces the player to spread it's defenses. Which makes the AI to require a fraction of the brute force to break trough.

There are also a fourth and fifth ones built into the game already which are the AI "not attacking heavily defended planets with threat fleet" and "taking into account player's mobile firepower x hops away". But that just means the AI is map hacking so it's unfair. Once again the AI in AI War is designed to be asymmetrical. It's not meant to be like Terran vs Terran (human players both if which have the same skill level) in Starcraft.
« Last Edit: September 23, 2015, 03:35:31 pm by Kahuna »
set /A diff=10
if %diff%==max (
   set /A me=:)
) else (
   set /A me=SadPanda
)
echo Check out my AI War strategy guide and find your inner Super Cat!
echo 2592 hours of AI War and counting!
echo Kahuna matata!

Offline Elestan

  • Full Member Mark II
  • ***
  • Posts: 158
Re: Difficulty for fully "smart" AI
« Reply #23 on: September 23, 2015, 05:40:58 pm »
Nah, if I'm reading/remembering right, Vorpal's primary change is that it adds Cross Planet ATTACKS on top of the cross planet waves.
I'm already getting hit by the occasional CPA, so I'd guess Vorpal just does them more often.

Combine Cross Planet Waves with Vorpal AI and you kind of have that. CPWs spawn randomly on the edge of the "map" and Vorpal is "equally fearsome with a fraction of the brute force" because of the "anti defense units". Sure the ships from CPWs still travel in a straight line but you never know where they will come from. It's not like you would rebuild all defenses every time after the CPW spawns.
The problem is that this goes too far in the opposite direction; it tries to compensate for the AI's lack of tactics by giving the AI a superpower that makes it totally impossible to chokepoint them.  I do want to chokepoint them, but I want to want to use the map to do it (meaning fortifying the wormholes and my facilities more), rather than being able to just take advantage of the AI's predictability. 

3: be able to attack any planet so the player never knows where the next attack will come.
I definitely don't want that.  I want to be able to put my scouts out, so I can see the AI coming, not have them magically teleport behind my lines.  That's just trying to compensate for an AI weakness by letting it cheat more.  Which is all-too common in video game AIs, I know, but one of this game's selling points was that it was supposed to have a really smart AI that could challenge the player without having to cheat (reference my earlier quote from Chris' developer blog), so my expectations coming in were quite high.

So your big issue is the AI doesn't do any maneuvering? I hate to say it, but that's completely intentional. AI War is as much a tower defense game as it is RTS. Of course defensive strategies are powerful, the game's AI needs to follow lanes.
Why does the AI have to follow lanes?  I mean, I understand that any tactics need to be kept fairly simple in order to work with the number of units involved, but I see no fundamental reason that the AI couldn't have some slightly more sophisticated behaviors implemented.  They're already using some aspects of simulated flocking and swarming, and adding obstacle/hazard avoidance to swarming systems has been done many times.

It's justified by the metaplot...
To me, this just feels like a layer of tactical behavior that the developers haven't gotten around to implementing yet.  Which is fine; developers are busy people (I know; I am one).  But I prefer to advocate for getting this aspect of the AI improved, rather than justify/rationalize it away.

Offline Captain Jack

  • Hero Member Mark II
  • *****
  • Posts: 808
  • Just lucky
Re: Difficulty for fully "smart" AI
« Reply #24 on: September 23, 2015, 09:07:31 pm »
I knew that line was going to be misconstrued.  ::)  Here's what I'm trying to communicate and you seem to be missing: AI War is a very deliberately created game. It's a work in progress too, and both what it is now and what it's intended to be send the same message: forget tactics, use strategy.

Feints, retreats, kiting, flanking, turtling are hallmarks of the RTS genre. I'm not one of Arcen's devs but I'm fairly confident that few of these will be taught to the AI, because AI War isn't an RTS. Just as much of it is tower defense or grand strategy, plus Chris likened AI War to a puzzle game in a few places.

AI ships don't break off attack runs because the survival of an individual ship or even an entire wave is irrelevant, as is their performance. What matters is if your command station survives. The entire game is designed to get you looking at the big picture. The AI doesn't care about efficiency, and by the same token the only thing you should care about is whether you can survive its next attack. Forget the micro, on both sides of the equation. Heck, they've taken a lot of micromanagement out of AI War over the years to focus on the big picture. Accuracy, the old armor rules, entire resources and the old power mechanics, all gone. Gone to reduce the tactical layer and to make more room for your strategy.

In the game's current incarnation, you'll have to enable some of those AI tweaks or cheats to get the tactics game you want. I re-suggest all the AI plots and minor factions I've suggested before, and also add AI Level 8 to see. In the future I expect the discrete tactical behavior of the AI will not change much (because it is working as intended), and that the strategic layer will only continue to expand.

Lucky for you, there is still a chance to change that. All you have to do is beat a game of 10/10, and prove that it's because the AI couldn't tactics its way out of a wet paper bag.

Maybe a bit too sarcastic. Hm.

P.S  If AI did do tactical adjustments to go around defenses then what other option would the player but to have to put all turrets around the incoming warp holes?  Wouldn't that favor putting research in ships more than turrets?
The pendulum would swing back there, yes. It wouldn't change the defensive nature of the game, but you'd need to be a much more active part of the defense. More micro. Bleh.

Offline Elestan

  • Full Member Mark II
  • ***
  • Posts: 158
Re: Difficulty for fully "smart" AI
« Reply #25 on: September 23, 2015, 09:39:00 pm »
There is no such thing as a smart AI. That only exist in science fiction.
How are you defining 'smart'?  If you mean that they think as broadly and generally as a human, then they don't exist yet.  But for specific tasks - even complex ones - a modern "AI" can be quite capable.  They are now able to do some forms of medical diagnosis on par with human doctors, for example.

Strategy games that has the AI playing the same rules as the player is won by mid-game. The endgame is nothing but mopping up for the victory.  At least in AI War it's possible to lose in the endgame (as in chess).
It's interesting that you mention chess, because in chess, AIs have been defeating even the top human grandmasters for nearly a decade.  Also, the reason that the endgame in AI War is hard has nothing to do with the type of cheating that I'm objecting to (cheating on knowledge or ship movement).  The AI War endgame is hard because the AI's effective production rate increases instead of decreases as it loses systems. And I'm fine with that, because it fits the game's premise.

P.S  If AI did do tactical adjustments to go around defenses then what other option would  the player have but to put all turrets around the incoming warp holes?  Wouldn't that favor putting research in ships more than turrets?
The player would have to weigh how they would divide their stationary defenses between the wormholes and the structures they want to defend.  I certainly wouldn't put everything on the wormholes, because then anything that broke through would have a clear run at my command center.  But that's a far more interesting problem than the status quo, where I know the exact flight paths that the enemy ships will take, so I can engineer The One Perfect Defensive Formation for my turrets.

I would expect that a smarter tactical AI would require some rebalancing; any major change would.  Some possibilities I can think of might include:
  • Lowering the AI's "Am I strong enough to attack" calculation threshold.
  • Lowering the size of attack waves.
  • Buffing turrets a bit, or raising their caps.
  • Adding a 'defensive garrison' ship that can't use wormholes.
All of which seem fine to me, and I'm sure there are others.
« Last Edit: September 28, 2015, 09:47:51 am by Elestan »

Offline Kahuna

  • Core Member
  • *****
  • Posts: 2,222
  • Kahuna Matata!
Re: Difficulty for fully "smart" AI
« Reply #26 on: September 24, 2015, 01:43:23 am »
I just realized I accidentally mixed Cross Planet Waves and Corss Planet Attacks.^^ My bad.

I can't think if any game I've play  except chess where the AI can beat me on the tactical level. The AI becomes predictable just as much as if it went in a straight line.
In Starcraft 2 the AI tries to mimic human player by executing "real builds" designed by the professional players. The AI also has to scout just like a human player. I don't remember if the AI micros too. Sure the AI is hard for a beginner but in the end it's still predictable and exploitable.
« Last Edit: September 24, 2015, 02:05:13 am by Kahuna »
set /A diff=10
if %diff%==max (
   set /A me=:)
) else (
   set /A me=SadPanda
)
echo Check out my AI War strategy guide and find your inner Super Cat!
echo 2592 hours of AI War and counting!
echo Kahuna matata!