Welcome

Please register for Total War Access to use the forums. If you're an existing user, your forum details will be merged with Total War Access if you register with the same email or username. For more information please read our FAQ’s here.

Categories

Wouldn't it be cool to see "WarhammerStar" be developed by DeepMind for TWW?

WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
Recently I stumbled upon DeepMind's "AlphaStar" project.

For those of you who are not up to speed with this, this is an AI created, using machine learning/neural networks to play Starcraft II. It is basically "handicapped" to play like a human in terms number of actions it can take per minute and to only be able to "see" what the screen is showing (like a human would). It was never programmed how to play. It just played itself millions (billions?) of times to figure out how to play.

It is REALLY good and able to play on ladder at the highest level, though still beatable by humans and with room for improvement.

I wonder how cool it would be if their next project would be to create WarhammerStar to play Quickbattles.

What kind of armies would it build? Would it find abusive builds that humans have not found yet? What novel tactics would it employ? How good would it get? Would this be harder then creating an AI that can play Starcraft II give that Starcraft has only 9 MUs while TWW has 100s?

Tagged:

Comments

  • tzurugbytzurugby Posts: 275Registered Users
    Honestly, seeing what AlphaZero accomplished in chess stunned me. It is only a matter of time before Total War gets a true AI, and it won't be long after that that it can't be beaten even by the best players.

    I once thought that computer chess engines would ruin chess, but they have only made it better. I know there are many players who dream about a time where the "AI" for TW will be competitive without receiving "artificial" bonuses. I think it is no more than 10-15 years away at the outside, but it is inevitable.....
  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
    I am not sure I agree it is inevitable that such an AI be included in games.

    I think most players would not find it fun to be crushed by AI, which they inevitably would.

    From a company's perspective, putting funds into a human like AI would not be a good investment I think.

    Imagine playing every SP battle vs Lotus_Moon, Felcon or Xiphos (or any of the other top players). Might not be as fun as it seems due to the fact that often in campaign battles are not even affairs like they are in MP.

    It's just fun to speculate.

    But to your point, all I ever wanted in strategy games since I was a kid was for AI to be challenging.

    I actually did find AI to be more or less competent in a few games. There was a game on the Sega Genesis called "Liberty or Death" from 1993. Despite being able to crush most strategy games I ever played, I was never able to win this one.

    Conquest of the New World also had decent AI that I would lose to sometimes.


  • 39821739175248623982173917524862 Posts: 817Registered Users
    Isn't this AI only developed specifically for SC2? Also Total War and SC2 are very different games, with the former relying on strategic planning while the other more on mechanical play like micro and optimizing build orders.
  • psychoakpsychoak Posts: 2,367Registered Users
    I would love to be crushed by the AI.

    My dream game is literally this game, with an AI that isn't ****.

    I should be able to play normal, and have a cheats free challenge from an AI that can match a decent player in battle.

    If I can play hard, and have to be seriously good to beat it, without cheats, what's the downside? A better AI isn't any worse in terms of difficulty, than stacking up stupidly huge bonuses to compensate for abject stupidity.

    With a 10% advantage in manpower, RTS combat shifts massively to one side. You can take 10 of a unit up against 11 of a unit, and wipe with half the one side still up. The same is true of minor stat bonuses. They result in absolutely massive power shifts. A single point is a substantial advantage in results.

    Despite this, people can win battles against very hard battle difficulty, while being outnumbered. The abject stupidity of the AI is offsetting titanic balance adjustments, with room to spare. If it were actually good enough to not cheat and still provide a substantial challenge? Holy ****.
  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
    This AI was specifically designed to play SC2.

    I was thinking more along the lines of battle AI for TWW, not campaign
  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
    psychoak said:

    I would love to be crushed by the AI.

    My dream game is literally this game, with an AI that isn't ****.

    I should be able to play normal, and have a cheats free challenge from an AI that can match a decent player in battle.

    If I can play hard, and have to be seriously good to beat it, without cheats, what's the downside? A better AI isn't any worse in terms of difficulty, than stacking up stupidly huge bonuses to compensate for abject stupidity.

    With a 10% advantage in manpower, RTS combat shifts massively to one side. You can take 10 of a unit up against 11 of a unit, and wipe with half the one side still up. The same is true of minor stat bonuses. They result in absolutely massive power shifts. A single point is a substantial advantage in results.

    Despite this, people can win battles against very hard battle difficulty, while being outnumbered. The abject stupidity of the AI is offsetting titanic balance adjustments, with room to spare. If it were actually good enough to not cheat and still provide a substantial challenge? Holy ****.

    I totally agree with you that AI is one of the most important things in a strategy game. But what DeepMind did with AlphaStar is probably too much for 99.99% of players. I also agree giving AI cheesy stat bonuses is unfun.

    What is even more unfun and dumb for me is the fact that the AI never plays "to win" like an autonomous player, has no real survival instinct and it's role seems to be to just be an opponent for the human player.

    How awesome would it be if the AI actually played true (with some randomness) to the lore and what said faction might "really" do in the Warhammer Universe?

    But, I do think 99.99% of players would not enjoy SP campaign if the battle AI was some version of Alphastar. When you watch it play on ladder, against high level SC2 players, it just murders them (mostly). And this is without any sort of cheating. No bonuses, even handicapped to not allow its actions per minute to be higher than average pros. In campaign, you are often at a disadvantage, so to have any chance of survival, the whole campaign element would have to re-adjusted, along with the campaign AI so that it does not target the player as it does now.

    For me, I would like the AI to be good in battles, but not super good if you know what I mean. I would not like to play at the level of top players.

    All I ask for is that it has rudimentary knowledge of battle tactics. Like don't get your lord killed. Protect your arty. flank enemy with Cav. look for favourable engagements. Try to kill enemy lord. Don't blob stupidly. don't walk straight into overwhelming ranged firepower etc.





  • psychoakpsychoak Posts: 2,367Registered Users
    Once you actually have an AI that is competent to beat 99% of the population without cheating, it's a fairly simple matter to make it be dumb and not behave optimally. If you start off with perfect, all you need are simple things like adding a two second delay on reactions, giving a percent chance to not react, etcetera.


    It's getting smart that's hard, not getting dumb afterward.
  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
    psychoak said:

    Once you actually have an AI that is competent to beat 99% of the population without cheating, it's a fairly simple matter to make it be dumb and not behave optimally. If you start off with perfect, all you need are simple things like adding a two second delay on reactions, giving a percent chance to not react, etcetera.


    It's getting smart that's hard, not getting dumb afterward.

    That's a good point.

    Right now though, the amount of resources needed to get it to be "smart" is so large that it is not practical at all for game developers.

    The other issue is of course that even if you can create a "smart" AI, it might need a powerful machine to power it, thus not being something that you can put in a game since that kind of power might be out of reach of most people.
  • TennisgolfbollTennisgolfboll Posts: 7,852Registered Users
    psychoak said:

    Once you actually have an AI that is competent to beat 99% of the population without cheating, it's a fairly simple matter to make it be dumb and not behave optimally. If you start off with perfect, all you need are simple things like adding a two second delay on reactions, giving a percent chance to not react, etcetera.


    It's getting smart that's hard, not getting dumb afterward.

    Smart? Such an ai would still be alot dumber than a simple rat.

    Kinda like comparing a stone (the AI) to Einstein (the rat).

    Ofc "it" could "play" vs us and we would have alot of fun.
    Read all my replies as if we are having a pint and a good old time. I will always read your reply like that.
  • Ephraim_DaltonEphraim_Dalton Senior Member Posts: 21,134Registered Users
    edited August 26
    tzurugby said:

    Honestly, seeing what AlphaZero accomplished in chess stunned me. It is only a matter of time before Total War gets a true AI, and it won't be long after that that it can't be beaten even by the best players.

    I once thought that computer chess engines would ruin chess, but they have only made it better. I know there are many players who dream about a time where the "AI" for TW will be competitive without receiving "artificial" bonuses. I think it is no more than 10-15 years away at the outside, but it is inevitable.....

    Comparisons to Chess are problematic because Chess has a much more restricted number of possible board constellations than TWWH (or any TW title actually) plus vastly different starting units. That's why TW's AI doesn't work like a Chess computer and is more openly "mechanical" since if it did approach it like a Cess AI that would probably extend turn times to weeks or months (or years actually).
  • nonentitynonentity Posts: 176Registered Users
    Honestly, the current version of AlphaStar isn't really that impressive -- it relies heavily on out-microing it's opponents as oppose to out-thinking them. As a result, ladder players have been abusing/cheesing the AI (not dissimilar to how players abuse current TWW AI) with unorthodox strategies that isn't apart of the neural network's training set. Maybe in a few years, I can see AlphaStar dominate the SC2 scene but what is the point? I much rather be playing against human players with the same limitations and advantages as myself.

    Thus, instead of focusing on developing AlphaWar with DeepMind, CA should really start looking into improving its multiplayer features. It's 2019 -- most 4X strategy games have large player count multiplayer campaigns with simultaneous turns -- all competitive RTS games have great ladder experiences -- why doesn't TW? The only thing keeping the current multiplayer scene alive is the community's passion and love for the game -- but without CA investing in basic infrastructure like dedicate servers, I don't see TW ever becoming anything more than just a single-player experience.

    P.S I hope CA really invests in the Everchosen Invitationals -- they are by far the best thing to come out of the online community and with CA's support,will really help grow the scene. Consider flying out the players to the studio, like how Youtubers and journalist are invited for promotions. The STEAM breakdown last time could have easily been averted if the tournament was hosted on LAN.
  • psychoakpsychoak Posts: 2,367Registered Users
    We seem to be missing that it's playing Starcraft, not Chess.

    Much more of an apples to apples comparison.
  • Deep_echo_soundDeep_echo_sound Posts: 447Registered Users

    Recently I stumbled upon DeepMind's "AlphaStar" project.

    For those of you who are not up to speed with this, this is an AI created, using machine learning/neural networks to play Starcraft II. It is basically "handicapped" to play like a human in terms number of actions it can take per minute and to only be able to "see" what the screen is showing (like a human would). It was never programmed how to play. It just played itself millions (billions?) of times to figure out how to play.

    It is REALLY good and able to play on ladder at the highest level, though still beatable by humans and with room for improvement.

    I wonder how cool it would be if their next project would be to create WarhammerStar to play Quickbattles.

    What kind of armies would it build? Would it find abusive builds that humans have not found yet? What novel tactics would it employ? How good would it get? Would this be harder then creating an AI that can play Starcraft II give that Starcraft has only 9 MUs while TWW has 100s?

    This was exactly what I was about, when some players whine that they have no challenge in campaign even on legendary difficulty. And recommend me to play on easy if I want to overpower A.I. in the late game. So pitiful.
    In "starcraft 2" and "D.O.T.A. 2" A.I., over the last years, made tremendous improvements. Right now no human in the world can beat them, if they are not limited by reaction time and actions per minute.
    If... If only this kind of A.I. will be implemented in the "total war", no one will ever finish a single campaign.
    Most people do not understand the power of A.I. even remotely.
    It will ideally dodge all cannon balls with cavalry when it is physically possible by the engine. (In case of 1 unit of cavalry vs 1 unit of cannons.)
    The cycle-charges of A.I. will be absolutely perfect and irresistible, and aiming with range units will be near perfect.
    Even worse with magic. Right now vortexes are great in campaign. But they will be far, like far worse than even in single player now. When cast by the player on moving unpinned targets, chain lightning will kill exactly 0 units. Right now it can kill hundreds.
    And it can predict, with an accuracy of exactly 1 point, that this amount of pressure on that player's unit in that second will rout it. And that your second unit, next to it, will rout in that amount of seconds exactly, freeing his flank entirely. And so on.
    It will destroy even the best players, like they are complete zeros in this game. When you have 20 or 40 units to command simultaneously.
    I am not sure that I want that kind of A.I. in game without very heavy restrictions. Like 30 A.P.M. max, 1 second delay reaction time. And completely "blind" to some things, like areas where the player will cast a spells like vortexes. Plus, some tactics completely forbidden, like aerial charges last moment denies. When some collision damage to ground troops already done, for example by dragon, but he actually does not land, and they can not hit him back.

    And, in addition, no more "blobs" is not that fun, because abusing A.I. is like the only way to get around with his huge bonuses on very hard difficulty. With range. Like killing hundreds with catapults feels so good. When your missiles land in that juicy enemy pack.
    Because, like mentioned, in pure hand to hand A.I. is so, so much stronger because of cheats.
    Now, in campaigns, A.I. is often stronger (in terms of units in army), but you are smarter, so still can defeat that allegorical stupid giant. Feeling good about yourself outsmarting someone, like heroes that you read about or watched in fairy tales, cartoons, and films, from the childhood.

    Battle A.I. is almost great. Maybe someone wants more challenge - then great! A.I. for them to suffer.
    But what I really like to see, is smarter campaign A.I. That attacks with a strong double or triple stacks (can concentrate forces) and that ambushes and bites player into the ambushes sometimes, to be unpredictable. Like players do, when 1 strong stack is in the ambush stance and 1 weak stack luring you in to that trap. In campaign there no real time competition, so A.I. can not use it reaction, only predictive analytics. Which is great and can be very fun and challenging on top difficulty levels.
    But, even that, need to have some restrictions still, because A.I. can just declare war on player, the with all factions, from the turn 1, and destroy her/him with no chance to resist.

    So, in summary.
    1. Real A.I., even heavily restricted, may be very unfun to play against in real time battles.
    When campaign battles already, sometimes, much more fun and enjoyable that multiplayer battles. That are often somewhat nervy and too stressful. Like, whole campaign with hundreds of these battles vs your own computer? Sounds like a job.
    2. On the other hand, on campaign map, with some heavy restrictions, like "your goal is not to kill human player", so A.I. can not unite all factions against you, it can be very great on top difficulty.
    With no reaction advantage, no anti-player goal, but real play for every faction to win the game, it can be very, very fun and challenging at the same time!
    And, if you can still abuse A.I. in manual battles, to deny it's economic and strategic superiority, and feel like a hero! So still can win the game that was created with intention to be beaten by player.
  • 42konyo42konyo Posts: 775Registered Users
    Can't wait till "A.I" like in this example are mainstream and are implemented in TW titles, no longer will players have to be handicapped to increase the difficulty, instead they'd just lessen the handicap on the AI which is how RTS/grand strategy games should be played, it'll take a while maybe even a decade or 2 but once CA dares to experiment with it i'll rejoice and be the first one to preorder.

    @nonentity

    "P.S I hope CA really invests in the Everchosen Invitationals -- they are by far the best thing to come out of the online community and with CA's support,will really help grow the scene. Consider flying out the players to the studio, like how Youtubers and journalist are invited for promotions. The STEAM breakdown last time could have easily been averted if the tournament was hosted on LAN."

    How is a promotional tool for new content the "best thing to come out of the online community"?
    Correct me if things have changed but it's not a real tournament since it's invite only for streamers and social media personalities and their inner circles while completely disregarding the top ranked players (which due to negligence/abuse doesn't really count for all that much i'll admit), there's no way for community members to qualify for a spot in the tournament nor is it even a proper tournament since it's done for promotional purposes (new dlc) which means that the build they're using is unbalanced and the new faction/units are always OP until the first post release patch comes around, i'd hardly call all that something from the online community let alone the best thing to come out of it.

  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
    nonentity said:

    Honestly, the current version of AlphaStar isn't really that impressive -- it relies heavily on out-microing it's opponents as oppose to out-thinking them. As a result, ladder players have been abusing/cheesing the AI (not dissimilar to how players abuse current TWW AI) with unorthodox strategies that isn't apart of the neural network's training set. Maybe in a few years, I can see AlphaStar dominate the SC2 scene but what is the point? I much rather be playing against human players with the same limitations and advantages as myself.

    Thus, instead of focusing on developing AlphaWar with DeepMind, CA should really start looking into improving its multiplayer features. It's 2019 -- most 4X strategy games have large player count multiplayer campaigns with simultaneous turns -- all competitive RTS games have great ladder experiences -- why doesn't TW? The only thing keeping the current multiplayer scene alive is the community's passion and love for the game -- but without CA investing in basic infrastructure like dedicate servers, I don't see TW ever becoming anything more than just a single-player experience.

    P.S I hope CA really invests in the Everchosen Invitationals -- they are by far the best thing to come out of the online community and with CA's support,will really help grow the scene. Consider flying out the players to the studio, like how Youtubers and journalist are invited for promotions. The STEAM breakdown last time could have easily been averted if the tournament was hosted on LAN.

    I just have to correct this comment a little bit (no offense). While it is true that unorthodox strategies are having some success against AlphaStar (one example in recent game was when it lost with Zerg against a Terran bunker rush), it is not the case that players "abuse/cheese" the AI. This is not my opinion.
    This is fact.
    Players don't know they are playing the AI. When playing the actual game, there is really no way to tell whether you are playing against AI or a human because Alphastar plays eerily similarly to a very good player. (it sends scouts, makes mistakes and loses units when it's "not paying attention" etc.)

    Keep in mind the AI does not cheat. it is limited in APM similar to what a top Starcraft II pro would do. In fact, when you watch replays of the games, it doesn't micro units perfectly (because it doesn't have the APM to do it).

    Also, the AI is not "all seeing". It has to move the camera just like a human and cannot take actions "off camera". So it's entirely possible to sneak up on the AI etc.

    There was very early version of Alphastar with no limitation on APM and with an "all seeing" view. It beat top humans 10-1. But it was quickly limited to human level powers because of the unfairness.

    I agree CA should invest in a more fun and robust ladder/MP experience.





  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users
    @Deep_echo_sound, I am not sure where you are getting the following from:
    In "starcraft 2" and "D.O.T.A. 2" A.I., over the last years, made tremendous improvements. Right now no human in the world can beat them, if they are not limited by reaction time and actions per minute.

    I am not sure about Dota, but in Starcraft II the AI is not unbeatable. Far from it. I think against top level players on ladder it losing quite a bit. Also, it is restricted by Actions per minute and by the fact that it cannot "see" everything at once, only what it's camera is showing. So it cannot micro units in 3 different areas of the map simultaneously. The only thing you are right about is that it is not limited by reaction time, though it still has to "think" about what it wants to do and sometimes it takes a reeeeeeally long time to make a decision.

    You do make good points about a super AI that might be able to perfectly dodge ranged fire and spells, cast it's own spells, calculate exact timing of routs etc.

    But, if a project like Alphastar were to be conducted in Total war games, Ai would be limited to same APM as humans as well as being able to only control one unit at a time (or control groups) like humans. So it would not be able to dodge everything at the same time. BTW, good human players are extremely good at dodging, especially with something like a lord on pegasus or eagle.

    I know you were just giving an example of how an AI could be limited, but you are grossly overexaggerating. You wrote:

    Like 30 A.P.M. max, 1 second delay reaction time. And completely "blind" to some things, like areas where the player will cast a spells like vortexes.

    I hope you realize that humans can have more like 200-300 APM and way lower reaction time than 1 second. And good players are rarely blind to vortex spells.

    So something more reasonable would be: 100 APM, with a reaction time of 0.5s +/- 0.4s. So sometimes it can react in 0.1s, other times in 0.9s.

    Interestingly enough, you could put different restriction on different AI armies/factions.
  • BoombastekBoombastek Posts: 2,025Registered Users
    psychoak said:

    I would love to be crushed by the AI.

    My dream game is literally this game, with an AI that isn't ****.

    I should be able to play normal, and have a cheats free challenge from an AI that can match a decent player in battle.

    If I can play hard, and have to be seriously good to beat it, without cheats, what's the downside? A better AI isn't any worse in terms of difficulty, than stacking up stupidly huge bonuses to compensate for abject stupidity.

    With a 10% advantage in manpower, RTS combat shifts massively to one side. You can take 10 of a unit up against 11 of a unit, and wipe with half the one side still up. The same is true of minor stat bonuses. They result in absolutely massive power shifts. A single point is a substantial advantage in results.

    Despite this, people can win battles against very hard battle difficulty, while being outnumbered. The abject stupidity of the AI is offsetting titanic balance adjustments, with room to spare. If it were actually good enough to not cheat and still provide a substantial challenge? Holy ****.

    That would be great.

    I too don’t like buff stats on legendary difficult, but SFO 3K is removed buff stats on batle difficult and it somehow improved battle AI.
  • nonentitynonentity Posts: 176Registered Users
    @42konyo
    42konyo said:


    How is a promotional tool for new content the "best thing to come out of the online community"?
    Correct me if things have changed but it's not a real tournament since it's invite only for streamers and social media personalities and their inner circles while completely disregarding the top ranked players (which due to negligence/abuse doesn't really count for all that much i'll admit), there's no way for community members to qualify for a spot in the tournament nor is it even a proper tournament since it's done for promotional purposes (new dlc) which means that the build they're using is unbalanced and the new faction/units are always OP until the first post release patch comes around, i'd hardly call all that something from the online community let alone the best thing to come out of it.

    I know the event is completely promotional and not the best representation of the competitive scene, but it is the only one CA has invested in and by far the most popular. Despite all the hard work of tournament organizers over the years, like ShetlandApachee, My_Son_HW, and Turin, the community is really small and most events fly under the radar. While I agree the invitational isn't fair like no qualifiers and new DLC units being broken, to say the event completely disregards the the top ranked players is fallacious. The last Everchosen had top players namely: Tlaxtlan Soothsayer, Felckon, Aerocrastic, RomulanDawg, Rightist. Do I wish more top players are represented instead of regular streamers, do I wish there was an online qualifier, do I wish CA took the competition more seriously instead of making it a market scheme --- Yes, but I can't complain when CA finally takes the opportunity to highlight some of the talent of the community.
  • SiWISiWI Senior Member Posts: 10,371Registered Users
    Personally I think people underestimate the fact that one of the reasons for bad AI is simple that making a challenging AI is quite easy but making it challenging and fun is not.

    Also I see most problems AI wise with the Campaign map less with battles.
    Ratling_Guns.gif?t=1554385892
  • Deep_echo_soundDeep_echo_sound Posts: 447Registered Users
    Sorry to say, but you are wrong about some things.

    https://youtube.com/watch?v=cUTMhmVh1qs

    I wrote.
    "Right now no human in the world can beat them, if they are not limited by reaction time and actions per minute."
    And it is true. Even in these matches A.I. was heavily restricted. By improving over time and with no restrictions at all.
    Please, people, read what words I wrote exactly.
    And A.I. had reaction time delay. And was able to micro 3 armies at once in one moment as we could see.
    With no limits A.I. currently can outplay any player in the world.

    And I know that "good players" can dodge, but that is not fun. That is annoying.
    This is strategy game, but fast clickers win. Even more than in "starcraft" really.

    And that was like random example, but it is very close to real restrictions needed even on legendary difficulty.
    Top "starcraft" players have sustained A.P.M. of like 400. Do I need this "crazy pianist" style players playing against me? No, let them take sedatives and stay away from my game.
    And that is the point. At lower than 1 second reaction time they will perfectly dodge cannon balls at some distance, and like almost all spells. Not even talking about arrows and perfect mass abusing.
    Even with these: "30 A.P.M. max, 1 second delay reaction time. And completely "blind" to some things, like areas where the player will cast a spells like vortexes" restrictions real A.I. still will be unpleasant to play against.
    As I said. No crowd around our biting hero? - No joy to see hundreds kills from players' catapults and spells.
    Smart flanking in the forest out of the range of player's artillery? - Wasted slots in army.
    Invisible units' timing smart attack from all sides? - Our game plan destroyed.
    Mortis engine blocked from players' real damage, when she/he only has melee infantry troops?
    Archers, gunners and alike that do not flee, in always skirmish mode, when player fright few of them with her/his single cavalry unit? - So player can only brute force A.I. not outsmart it.
    And be not the smartest on the battlefield is not that great feeling. You are the hero here! Not some invisible pesky code.

    We can all watch some games, played by some clickers-neurotics vs usual total war gamers, on the channel of the last everchosen tournaments narrator. These are mostly not fun games, to play and even to watch.
    And I do not want anything reminding of them in my game. I think, many other players too.
    They already try to ruin the single player game on the balance forum, discussing units only in the vision of "how this unit performs in this match up, or in that match up".
    When the main thing about balance must be: "how this unit feels like for campaign player". How fair and good.
    So, we do not need bots that are like these few toxic clickers multiplayer abusers, that think they can determine the whole game.
    A.I.'s job is to be beaten by average player, after some practice. If the game wants to sell well and receive good feedback from buyers.
  • nonentitynonentity Posts: 176Registered Users
    It would be so much easier to implement something like an 8 players Mortal Empires campaign than developing a neural network to train AlphaWar. Seriously, the game would be so much more fun if you can share the experience with multiple friends of your own skill level than abusing or be abused by AI.

    Playing against an unbeatable AI like OpenAI in DOTA 2 or Smashbot in Super Smash Bro Melee is not fun, and that is not something I want TW to be. Seriously imagine constant cycle charging, perfect kiting, dodging every missiles attack -- I want an AI that can outsmart me, not out-micro me with ridiculous APM.

    Just implement better multiplayer features, seriously CIV, AOW, HOI, EU4, etc ... all have more than 2 player grand campaigns, why can't TWW?
  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users

    Sorry to say, but you are wrong about some things.

    https://youtube.com/watch?v=cUTMhmVh1qs

    I wrote.
    "Right now no human in the world can beat them, if they are not limited by reaction time and actions per minute."
    And it is true. Even in these matches A.I. was heavily restricted. By improving over time and with no restrictions at all.
    Please, people, read what words I wrote exactly.
    And A.I. had reaction time delay. And was able to micro 3 armies at once in one moment as we could see.
    With no limits A.I. currently can outplay any player in the world.

    And I know that "good players" can dodge, but that is not fun. That is annoying.
    This is strategy game, but fast clickers win. Even more than in "starcraft" really.

    And that was like random example, but it is very close to real restrictions needed even on legendary difficulty.
    Top "starcraft" players have sustained A.P.M. of like 400. Do I need this "crazy pianist" style players playing against me? No, let them take sedatives and stay away from my game.
    And that is the point. At lower than 1 second reaction time they will perfectly dodge cannon balls at some distance, and like almost all spells. Not even talking about arrows and perfect mass abusing.
    Even with these: "30 A.P.M. max, 1 second delay reaction time. And completely "blind" to some things, like areas where the player will cast a spells like vortexes" restrictions real A.I. still will be unpleasant to play against.
    As I said. No crowd around our biting hero? - No joy to see hundreds kills from players' catapults and spells.
    Smart flanking in the forest out of the range of player's artillery? - Wasted slots in army.
    Invisible units' timing smart attack from all sides? - Our game plan destroyed.
    Mortis engine blocked from players' real damage, when she/he only has melee infantry troops?
    Archers, gunners and alike that do not flee, in always skirmish mode, when player fright few of them with her/his single cavalry unit? - So player can only brute force A.I. not outsmart it.
    And be not the smartest on the battlefield is not that great feeling. You are the hero here! Not some invisible pesky code.

    We can all watch some games, played by some clickers-neurotics vs usual total war gamers, on the channel of the last everchosen tournaments narrator. These are mostly not fun games, to play and even to watch.
    And I do not want anything reminding of them in my game. I think, many other players too.
    They already try to ruin the single player game on the balance forum, discussing units only in the vision of "how this unit performs in this match up, or in that match up".
    When the main thing about balance must be: "how this unit feels like for campaign player". How fair and good.
    So, we do not need bots that are like these few toxic clickers multiplayer abusers, that think they can determine the whole game.
    A.I.'s job is to be beaten by average player, after some practice. If the game wants to sell well and receive good feedback from buyers.

    My apologies, I missed the part about "if they are not limited by reaction time and actions per minute". So you are right that if the AI is unrestricted, it is indeed unbeatable. However, what you linked is the initial 11 game series Alphastar played against a few pros where not only was the APM not limited, but also, it had "god vision" where it could see everything that was happening on the map, which is not only unhuman like, it is literally cheating.

    They quickly changed the rules to restrict it to human APM + vision and let it play on ladder. I watched at least several dozen games, and it lost about 4 or 5. And this wasn't playing against best players in the world, maybe "just" top 5%.

    But I can see we have different views on what is "fun" and so forth. You don't seem to like the fact that there are some very good players who would like the game balanced around MP, where everything is on a level playing field but isolated to human vs human fights.

    The point of my OP is that it would be cool to see Alphastar play the top players (with the usual restriction for alphastar) to see how well it could do, what strategies it would employ etc.

    BTW, you said AI should have 30 APM max. Do you have any idea how slow that is? I bet you that you can easily do 60 APM without trying. 30 APM is literally:Click. Wait 2 seconds. Click. Wait 2 seconds. Click. That is sooooo slow. I don't think anybody actually plays games that slowly.

    And if someone does, they are definitely never going anywhere MP, as they would get utterly wrecked by 99% of players.



  • angry_rat_loverangry_rat_lover Posts: 1,059Registered Users
    Imagine the charlemagnes
    Soon
  • TennisgolfbollTennisgolfboll Posts: 7,852Registered Users
    I think the AI is pretty good/fun in warhsmmer.

    They should implement the additions of 3k for sure (not shooting on heroes etc) on all difficulties (possibly not on easy)

    And ofc fix the AFK ai in sieges.

    But field battles arent bad once you would fix some of the issues
    Read all my replies as if we are having a pint and a good old time. I will always read your reply like that.
  • TennisgolfbollTennisgolfboll Posts: 7,852Registered Users
    Would still love a great ai ofc
    Read all my replies as if we are having a pint and a good old time. I will always read your reply like that.
  • Deep_echo_soundDeep_echo_sound Posts: 447Registered Users
    edited August 27
    I perfectly know what I wrote.
    In game like "total war", with 20 units, or even 40, player often do not need many A.P.M. Yes. A.I. can click attack order of that infantry unit on that infantry unit and do not touch it again in 40 seconds and more. Like real players do. They mostly micro lords and heroes, and fast units like cavalry. And cavalry abusing is very bad when done by player vs player. 1 click in 2 seconds, for real A.I., is needed, so it can not micro 12 cavalry units to perfectly cycle-charge and dodge. And even in this state, A.I. still can prioritize tasks, so to dodge magic missiles. Which is bad. I hate when in multiplayer someone constantly dodges my fireballs and other magic missiles. I cast magic and it's not working. So nice... And it wasn't even top players.
    In real battles even good players (not top clickers) forgot to manage their cavalry unit in the corner for minutes, that returned from the routing. A.I. does not waste anything.
    And "all-watching" is what all "A.I.s" do in strategy games. In every strategy game for decades. They just "see" coordinates. They, by definition, "cheating".
    But, I insist, that dodging machine-gun clickers are not fun. Some real players. Few of them, trolling and arrogant on that balance forum.
    A.I. will be even worse than them, because it will be actually really smarter than them too. Not only super fast clicker, like these game destroying abusers.

    The only place where I will be happy to see real A.I., in strategic battles, in "total war warhammer" games, is defeating this few egoistic nasty players, that dictating the game balance to C.A., for weird some reason.
    And it seems like no one, including C.A. workers, wants to hear campaign players suggestions. Their main fan and customers base. Like changing white lions of Chrace and myriad others.
    "No, let's make the game exclusively following desires of few clickers-abusers, and their, not more than part of 5% of all "warhammer" players, support team." - This is very sad balance behavior for real "total war" fans.
  • Deep_echo_soundDeep_echo_sound Posts: 447Registered Users
    And, to add, in "starcraft 2", in platinum league, in both 1 vs 1, and team vs team, I have only 80 A.P.M. average. And it is sufficient. Besides, in "starcraft" you theoretically can control all your 200 limit and all buildings. In "warhammer total war" you only need to control 20 (40) cards and some spells/abilities buttons.
  • SagrandaSagranda Posts: 1,610Registered Users
    @psychoak basically speaks from my very soul.

    The campaign and battle AI, both, are very very lacking in TWWH/2. In both aspects it's just not smart enough to offer any challenge without massive stat boosts and every advantage it can get.
    I won't deny that it's hard and costly to make an actual great AI, but when it's done it becomes a big game changer and I really would love for that. Not only for TWWH/2, but for games in general.
    "Many players cannot help approaching a game as an optimization puzzle. What gives the most reward for the least risk? What strategy provides the highest chance – or even a guaranteed chance – of success? Given the opportunity, players will optimize the fun out of a game."

    - Soren Johnson
  • WojmirVonCarsteinWojmirVonCarstein Posts: 726Registered Users

    I perfectly know what I wrote.
    In game like "total war", with 20 units, or even 40, player often do not need many A.P.M. Yes. A.I. can click attack order of that infantry unit on that infantry unit and do not touch it again in 40 seconds and more. Like real players do. They mostly micro lords and heroes, and fast units like cavalry. And cavalry abusing is very bad when done by player vs player. 1 click in 2 seconds, for real A.I., is needed, so it can not micro 12 cavalry units to perfectly cycle-charge and dodge. And even in this state, A.I. still can prioritize tasks, so to dodge magic missiles. Which is bad. I hate when in multiplayer someone constantly dodges my fireballs and other magic missiles. I cast magic and it's not working. So nice... And it wasn't even top players.
    In real battles even good players (not top clickers) forgot to manage their cavalry unit in the corner for minutes, that returned from the routing. A.I. does not waste anything.
    And "all-watching" is what all "A.I.s" do in strategy games. In every strategy game for decades. They just "see" coordinates. They, by definition, "cheating".
    But, I insist, that dodging machine-gun clickers are not fun. Some real players. Few of them, trolling and arrogant on that balance forum.
    A.I. will be even worse than them, because it will be actually really smarter than them too. Not only super fast clicker, like these game destroying abusers.

    The only place where I will be happy to see real A.I., in strategic battles, in "total war warhammer" games, is defeating this few egoistic nasty players, that dictating the game balance to C.A., for weird some reason.
    And it seems like no one, including C.A. workers, wants to hear campaign players suggestions. Their main fan and customers base. Like changing white lions of Chrace and myriad others.
    "No, let's make the game exclusively following desires of few clickers-abusers, and their, not more than part of 5% of all "warhammer" players, support team." - This is very sad balance behavior for real "total war" fans.

    Having just watched like 25 AlphaStar games, I can confidently say that the AI, as good as it's micro is, actually loses to pressure. It has a really hard time microing in multiple areas. A good example of such a games is against a grandmaster level player named "Awers". You can find that game on YouTube.

    As to your complaints about "constantly dodging fireballs". Well fireballs are one of the easiest spells too dodge. It's a yellow ball that can be seen as soon as its cast and it has to be the slowest missile spell in the game. If you are on a pegasus or horseback there is really no reason why you should get hit by it. Unless it was cast from a really close distance.

    Let's just agree to disagree.

    Getting back to my original OP, I was just wondering how good an AI could be, if it has the usual constraints placed on it (same restriction as humans etc.)
Sign In or Register to comment.