You’ll Like This: On Shogun 2 Total War’s AI

By Alec Meer on October 19th, 2010 at 10:24 pm.

Total War games and their AI, eh? I hear people are universally content with it, most especially with how solid, reliable and believable it was in Empire: Total War. Not even the slightest grumble. Oh, wait, I’m thinking of that dream I had again, just before the bit with the anteater wearing a sweater.

Grumblers! You have been heard. “This game is not going out the door until the AI is perfect,” the Creative Assembly are saying of the upcoming Shogun 2: Total War. Crumbs. Maybe they mean it this time…

It’s all in this New Zealand-published interview with the game’s community manager, in which he quotes the heartfelt pledges of the RTS’s main minds:

“Mike Simpson, our creative director, has basically said ‘this game is not going out the door until the AI is perfect’. AI really has been a bit of an issue for us in the past. But even Napoleon was better than Empire. I mean, one of our goals for Napoleon was to fix the issues we had in Empire, and I think to an extent we did that really well. But in terms of Shogun 2 we’re not going to release it until it’s perfect.

There’s the line-of-sight stuff we’re introducing – basically, if any of our designers are playing the game and see something stupid, it’s like everyone stops. Everyone gets around the PC to have a look at it, and see what can we do to fix it. We want to get it so there’s nothing stupid happening in the game. We don’t want to see stupid AI. It is a big thorn in our side, and it’s something we’ve really had to work on. We’re pretty confident about Shogun 2.”

Fine words. Fine, pretty words. Do you believe them? Being that open about past failings suggests they’re genuinely trying to fix it, but then we’ve heard similar before. What most needs to be worked on, strategy Ronin? What must not be carried over into this bright new age of Total War?

Also, I’m not sure we’ve posted this yet – an in-game battle report, chattily narrated by a couple of the devs – one of whom is, relevantly, the battle AI programmer. They go into some of the thinking and machinations behind S2TW’s land barneys, plus it’s a good chance to get up-close and eyeball-personal with a bunch of the units.

__________________

« | »

.

124 Comments »

  1. Army of None says:

    You were correct. I did like this.

  2. Dominic White says:

    The problem is that ‘perfect’ AI is impossible. AI is still a very young field of computing. It wasn’t that long ago that they finally made a chess computer that could consistently beat a grand-master, and that’s largely through brute-force logic and probability mapping.

    To this day, no AI can play Go (Othello to the rest of you) to expert-level, despite being played with only a single kind of piece, and with very simple rules. The best minds in coding haven’t cracked that problem yet, and yet people still want convincing artificial intelligence in a game where you play upon a board of unknown size, makeup and elevation, using literally thousands of pieces of hundreds of different types, all tied into a metagame of long-term planning.

    Honestly, I don’t envy the people are CA trying to find a solution to that problem one bit. Right now, the only halfway reasonable way of doing things seems to a a flow-chart of scripted reactions and a pre-planned ‘playbook’ of setups it can use. And even that would be difficult.

    • Brumisator says:

      Well of course it won’t be “perfect”, I’m pretty sure what they mean and what most people understood is that the debugging on this will be much more through than in the last few games, and that the AI will have more capabilities, will be more advanced and “smarter” I guess.

      I don’t think your Go analogy holds up here. they can easily make an AI that will steamroll any human opponent, taking any fun out of the game, the point here is to make an AI that is balanced: smart enough to play well, but not supremely invincible.

    • Peter Radiator Full Pig says:

      What? What AI do you speak of? Extra resources and knowing stuff a normal player does not is not AI.

      If they actually had an unbeatable AI in those games, i would love them that much more.
      Loosing against an intellegent opponent is much more fun that kicking the crap out of useless placeholders.

    • squidlarkin says:

      Sorry, I can’t let this stand: Othello is very much not Go.

    • Lazer says:

      I’m sorry but implying that Go is a more simple problem than a Total War game is silly. While Go has simple rules and only one piece, the possible moves are staggering. A game of Go stars with 55 possible moves, and then geometrically rises from there as more and more pieces enter the board. Supposedly, there are more Go moves possible towards the end of a game than there are atoms in a universe.

      In an RTS? “Oh hey. He’s charging me with horses. I should move my spear guys over. But only if it’s not gonna result in my archer guys getting killed!”

      Granted it’s still a difficult problem to make an AI do such a thing without breaking, but it’s a far more manageable task than Go, and Chess even for that matter. There is cause/effect and situational things that can be relied on for the AI to react at, as opposed to pure pattern recognition.

    • mandrill says:

      Go is not Othello (or Reversi seeing as Othello is trademarked) and a computer cannot be programmed to beat a Go master because winning at Go is all about pattern recognition, which computers are incredibly bad at.

      There is no way to brute force a game of Go because the number of permutations is orders of magnitude greater than those in chess. To calculate 4 moves in advance (2 for each player) a computer only has to work out 200 possible commbinations for chess. to do the same for Go would require the computer to calculate 320 billion possible board formations and compare them. thats 138 hours on the fastest supercomputer in the world.

    • Hallgrim says:

      @Peter Radiator Full Pig:

      What’s more important: the player’s perception of the AI, or the programmer’s perception of the AI? Violating the “rules” of the game is only wrong when it is so egregious that the player’s notice it.

    • Mo says:

      Solving Go is irrelevant to game AI. They are completely different problems.

      The former is about calculating all/most possible moves (and moves ahead) on a board and executing the best one. This is an “easy” problem, but requires tonnes of CPU time and RAM. It is unlikely this will be possible in real time in the near future.

      The latter is about using a set of heuristics to pick the best move. Ideally, they would interact with each other to enable some form of emergence. Game AI is still a bloody hard problem. And you’re right, I don’t envy the people at CA either! Fingers crossed, hope they can get this, not perfect, but pretty close.

    • Ybfelix says:

      @Mandrill: So Go can be brute-forced, the algorithms are theorically ready, we just need better computers, is it?

      Can you CS guys inform me if we are using the chess supercomputers to do RTS and write accordingly, will the AI be fundamentally better?

    • Phasma Felis says:

      @Ybfelix: Well, there’s the small problem that you could turn all the matter in the universe into RAM and you still wouldn’t have enough to store all the possibilities. But other than that, yeah, it’s totally doable in theory.

    • Lazer says:

      ALSO: Even if you have a computer that is strong enough to brute force Go, that would require the computer to have a prerecorded knowledge of every move possible, which is kind of defeating the purpose of having something that can look at a problem and try to solve it like a human being to a reasonable degree.

    • PeopleLikeFrank says:

      You can’t brute force Chess any more than you can Go – the total possible game space for Chess is not as great as Go, but still greater than the number of atoms in the universe.

      Any reasonably complex problem requires heuristics – they’re basically an educated guess to try to pick a likely path to get where you want, when you can’t map out every possible path you could take.

    • Tei says:

      You don’t need to “solve” chess or go. You only need to make a competent player, that make some minor “human like” mistakes.

      About solving chess or go by brute force.
      Thats now how chess masters work. Chess masters don’t see individual movements, like a general don’t step in the battlefield to tell private ryan where to shot and wen to launch a grenade. a chess master see “patterns”, abstract the game in bigger blocks than movements. Creating these blocks is creative, abstract and smart, and make playing chess a much easier game. The problem a chess master is solving is much easier than the one you guys are talking about, because he reduce it first, to a smaller and much easy to solve problem.

      In a way, RTS games already do this. Probably about all RTS game artificial oponents already have a “danger button– go defend base” and a “squad” concept ( 2 knights, 8 peasants and 1 archery unit). So these artificual oponents are not tryiing to solve “RTS” by brute force. Brute force is stupid, solving stuff by brute force is not smart. Is just the last option, you relie, wen theres nothing better.

    • Tei says:

      typo error:
      Thats not how chess masters work.

    • DrazharLn says:

      Deep Blue (the famous chessbot) was not an AI.

      Go is not Othello.

      Flowcharts are not the best that can be done in the realm of AI development.

      That’s too brusque, my apologies.

      Deep Blue beat Kasparov by simulating playing every possible move it could make at each junction and then the countermoves and that could be made in response to each. It did not simulate every possible move that could be played, but it simulated enough and far enough to beat Kasparov.

      Modern chess playing software is actually much smarter and simulates far less moves because it can intelligently discard possibilities that are unfavourable.

      Simpson’s choice of words was poor, of course the AI will not be a perfect AI (whatever the hell that is), I imagine what he means is that the “AI” will be decent enough that players will not gawk at its stupidity, indeed, it should provide the illusion of intelligence and a reality of challenge to the game.

      As to how difficult the job would be, bear in mind that the game does not need to produce something that will play the game perfectly, just one that can provide the player with a challenge. In addition to this, the game will probably not be looking at an enormous number of possible moves: simplifications will be made and the problem abstracted.

      It’s still not an easy task to make a system capable of passing this RTS Turing test, but it is by no means insurmountable and their failure to do so in the past is reprehensible.

  3. Sceptrum says:

    It’s either empty-talk like before with previous titles or hey, NTW might not have sold as well as CA/SEGA believed.

    I will keep an eye on Shogun 2, but won’t pre-order this TW title like I did with the previous ones (since Shogun 1)

  4. Wilson says:

    If they do manage to produce an excellent AI, I hope it provides motivation to other devs to also put more resources and effort into producing good AI. I wish them luck!

  5. Novotny says:

    Well, I wish them luck. Having been repeatedly burnt in the past by CA, I could never again contemplate a pre-order. It’ll also take some very good reviews for me to consider a purchase. But fingers crossed and all that.

    • battles_atlas says:

      What will reviewers do with this though? Given the scores Empires got, you have to assume a TW game with good AI would score well over 100%, and the universe would blue screen in a cascading logic fail.

      TW is a real test case in how you actually review a game. Considered in isolation Empires is a pretty amazing achievement in many ways. However if you look at it from the perspective of someone who has paid for several previous games in the series (ie most buyers), then it has huge problems.

      That said, there is no excuse for the reviewer’s inability to mention the multitude of bugs in the game when it was released.

      As for the AI claim today, heard it all before.

    • Strategem says:

      I have played quite literally every single game with the exception of some expansions from the TW series, and I have to say that all Empire is is rows of men standing in front of each other and shooting until someone is eliminated.

      In a very, very bad and boring sense, it is as realistic as battles were at that time..and therefore terrible, unlike the more fun periods with swordslancesarchershorsesandyay.

      So outside of introducing research/schools/gentlemen and a bad “Pirates!” clone subgame,I really would put Empire last in terms of innovation within their series.

      It’s a good thing they are redoing Shogun. If they do it well, that could get really interesting, although quite frankly the mechanics in Medieval battles felt a lot nicer than in Medieval 2..

  6. RaveTurned says:

    Damn those Grumblers! I wants my Shogun NOW I tells ya!

  7. DarkNoghri says:

    Not having really played the newest iterations, I can only base my desires upon Rome.

    Can soldiers walk and fight on bridges, gates, and walls, at all? It got to the point in R:TW that I would almost hesitate to upgrade from wooden palisades to stone, simply because I wouldn’t have to deal with the horrendous AI on the tops of the walls.

    • Henk says:

      Units could fight on bridges/city walls in empire and napoleon, but entering and exiting bridges was dodgy and units got stuck or all suddenly died if part of the wall/bridge got destroyed. So it worked just like the rest of the game.

  8. SomeGuy says:

    Im gona wait till the price is 1-20£ on this one, dont think i can trust them to get the AI right.

  9. Peter Radiator Full Pig says:

    Ive been burned before, so many times.
    Never will I believe that they can fix the AI for diplomacy.
    If you have 1 city, an unbeatable army outside it, and all I want for peace is some coin, GIVE IN!
    And other such nonsense.

    • Peter Radiator Full Pig says:

      Oh, and an obligitoury boo hiss at CA.
      One of the goals of Empire should of been fixing the problems of Empire.
      Not deciding patch content should be a new game.

      Wheres my multiplayer campaign, CA?
      What about that goal…

    • drewski says:

      Clearly, the AI should do exactly what you want at all times and it’s failure to do so means the design is flawed.

      Wait, what?

    • battles_atlas says:

      Great drewski, now you’ve triggered a troll alert. I have to get a ladder and swap the lightbulb out for a red one now.

    • Optimaximal says:

      @drewski

      I think it’s less trolling, more common sense.

      If the AI cannot win a battle and is presented with a peace offering, it should 7/8 times out of 10, take it. I’ve not played any TW game since Shogun, but as far as I can remember, the AI just never gave in… There were no states for it other than victory or abject failure.

  10. pkt-zer0 says:

    “This game is not going out the door until the AI is perfect”

    Shogun 2 is the next Duke Nukem Forever?

    • alh_p says:

      Yep it’s an absurd claim, obviously made by a marketing/sales guy who isn’t responsible for the AI, but is for demand for the game. Cretin.

    • Guildenstern says:

      Nope, it’s the next Skynet.

    • PeopleLikeFrank says:

      Goddamit, haven’t they watched any of those movies?

  11. Bluebreaker says:

    ” AI really has been a bit of an issue for us in the past. But even Napoleon was better than Empire.”

    yeah okay, giving massive CHEATS to the SAME AI makes AI better…
    guys don’t expect nothing from this you are being deceived, again.

    • Dominic White says:

      As mentioned, nobody has done better, because it simply can’t be done. The very best minds in the fields of artificial intelligence can’t even create an AI that can consistently win at a board game that only uses a single type of piece and a tiny ruleset.

      AI is one of the most astoundingly misunderstood fields in games, and technology in general. We’re at the point where it’s considered fairly good if a character in a game can navigate down a hallway without actually having a node-map drawn out for it to walk along.

    • Peter Radiator Full Pig says:

      You keep mentioning this, Mr. White, but it is misleading.
      Yeah, Go has a simple rule set. But the sheer number of decisions to be made on a 19×19 board game is immense.
      Simple rule set, complex choices.
      Start wittling down the choices, so the computer doesnt have to spend as long, and you will have a perfect AI.
      Or, give it enough time, and a computer programmed to play Go will play the perfect game. Times the limit, not the AI.

      And perfect AI, in this sense, means for the game. To play on the persons level, not to win every time.
      There are alot of units, but generally some core ones (Heavy cavalry can be ordered the same, as can bowmen of spearmen, even with different stats)

      Make a good battle AI, that can project how well it might do, and you can make an AI that knows what it might cost to win a war against opponent, plus how many fronts he fighting.

      Make him know the cost, and how to avoid it, and he may well play a much better game than the AI from Empire (The last CA game i will ever buy) whos AI declared war when it has almost no army orr you have it beaten already, and it can see that, etc.

    • Bluebreaker says:

      You are discussing on making a supreme AI capable of wining a human that understand the game rules. I’m talking about a AI that understands some minimal basics of the game.
      Like how to make money without cheats, where to put army stacks, defend cities/forts, to not abandon guns when pointlessly turning the mega line formation, or getting units unable to move properly when a single soldier gets stuck for not charging at the required time… and so on.

  12. televizor says:

    “There’s the line-of-sight stuff we’re introducing – basically, if any of our designers are playing the game and see something stupid, it’s like everyone stops. Everyone gets around the PC to have a look at it, and see what can we do to fix it.”

    Don’t they have a testing department?

    • SanguineAngel says:

      In fairness, this makes sense. I think a testing department is likely to run into these things later in development when fixing it is either a lot harder or not practical. Getting the team to face issues as they occur in real time with the development or at least earlier in development, surely provides a greater opportunity to fix things.

      My programming experience is utterly minimal though so I could be way off base but it strikes me that if they aren’t careful maybe they’ll just end up going round and round in bug fixing circles.

    • Saiko Kila says:

      I hope the policy saves them time they would have otherwise spent in gym. They must be very fit.

  13. Mad Doc MacRae says:

    No one ever asked for perfect AI. Just something passable. MTW’s AI could move at you in a general line and even tried to send cavalry behind you without tarding out. Rome and Med2 couldn’t and as far as I know Empire couldn’t and nobody who’s judgment I trust bought Nappy.

    Maybe they’ll sell us DLC AI packs to complement the “slightly different uniforms” packs. That’d be great, way better than making the game moddable. <_<

    • Langman says:

      Indeed. It’s a bit tragic that the last decent AI (or at least the closest to ‘decent’ we can get) was in Medieval1. Rome, Med2 & Empire were shocking.

      Somewhere along the way they dropped the baton and never found it again.

    • Ateius says:

      That’s odd. I distinctly recall enemy armies in both R:TW and M2:TW advancing at me in line formation. And skirmishing with ranged units. And flanking me with cavalry (or, if none was available, fast-moving infantry). And acting to protect its archers when threatened and moving units to counter my own flanking attempts if I didn’t tie their units up first. All without “tarding out”.

      The battle AI wasn’t bad. What it was, was predictable. It had a certain number of tactics available to it and those tactics never changed. Once the human player has identified those tactics and what triggers them, it becomes easy to exploit. What it needs is the ability to adapt on the fly and display emergent tactical thinking, which is a pretty tall order.

      Campaign map AI, on the other hand, is ridiculously dumb. In all Total War titles, rose-tinted goggle fans, not just the most recent ones. Don’t make me start digging out my M:TW campaign stories.

    • drewski says:

      I’ve not played M2TW, but the AI definitely flanked, got in behind you and maneuvered OK in Rome. Not always, and it was a bit stupid in trees at times, and it was pretty predictable, but it wasn’t quite as dumb as you’re suggesting.

    • Fumarole says:

      I distinctly recall enemy armies in both R:TW and M2:TW advancing at me in line formation. And skirmishing with ranged units. And flanking me with cavalry[...]

      Same with Empire. Enemy cavalry was always in my rear if I left myself exposed; I lost many a cannon this way because I overextended my front.

    • Luke says:

      Thing is with saying the battle ai was predictable-I agree, but it’s not just a mechanical thing. It’s because historically there were only so many tactical options regarding ancient/medieval warfare in regards to the fighting part.

      What made a general stand out was his mastery of the difficult stuff-chain of command, logistics, choosing his ground, essentially things which would offset the limited and predictable nature of people, the majority of whom weren’t really that up for fighting, smashing into each other in a big, pointy, gory mess. It was the generals who knew how to break this format or avoid it who were the best and this could only really be done before battle lines were drawn, not in the middle of battle itself.

      The only bit which you control in Total War is the people hitting each other bit which didn’t last for that long. In that sense it’s actually reasonably accurate I think in limiting what you can do.

    • Fitt3 says:

      Yeah. The AI is poor on purpose, for realism! Yay!

    • Luke says:

      Not saying that’s why the ai is poor. I think, if you look at my other comments below, it’s terrible. But in the very limited simulation of medieval warfare which is wholly unrepresentative of the historical phenomena that they chose to concentrate on it’s realistic. I don’t imagine this is intentional and I don’t imagine CA would agree that the simulations are limited either, so it’s a coincidence is all.

      The rest of the game, and CA’s other offerings other than the first two total wars, and the way they handle the limited things they do, display rubbish ai which is not even adequate for their simple aspects.

  14. RadioactiveMan says:

    Creative Assembly: PLEASE fix the way battles in towns and cities work. That has been my biggest grumble since the beginning of the Total War franchise. Battling the environment (walls, city streets, siege equipment, unit behavior, etc.) has been a problem during city battles for every game in the series. The way individual soldiers abandon their regular within-unit spacing and formation, but retain some kind of group behavior and movement plan is terribly implemented. Also: abandon the concept of a rally point at the center of town. The rally point has never worked properly. Infinite morale combined with crap pathfinding makes a mockery of the battlefield system. Its relatively easy to “game” the AI on defense, and quite thorny to defeat the AI on offense.

    Thats my main complaint. Empire had some other quite serious troubles, but the city battles were always the biggest problem for me. You MUST fight a large number of them to win the game, and they’ve always been broken and the battles play out in a game-y fashion.

    • battles_atlas says:

      Well said Sir. Its pretty damning of Empires that a whole comment thread can slag the game off without mentioning how epically awful the sieges are. Medieval II’s had problems, but at least they looked pretty. Empire’s forts looked like placeholders, and pathfinding was utterly broken until I think the 1.5 patch which took a year to appear. Was it the same patch that allowed rifle and canon fire from the walls to actually hit the attackers? Worst of all was the AI of the attackers, which it seems consists of one single instruction: get to the central square. A super stack attacking army could be defeated with just one offset canon unit repeatedly spraying the flag with canister.

      Anyone defending CA by arguing the difficulty of coding convincing AI just needs to play one of these battles, and promptly shut the fuck up.

  15. Sam C. says:

    Seems like some people are more than a little upset. Was the AI really that bad? I haven’t noticed anything too egregious, but then I’ve only played Rome. And I’m not all that skilled a player.

  16. Mr Chug says:

    Empire’s AI really was a bit rubbish in retrospect. The problem is that it’s just not creative after a while- the different terrain does a bit to mitigate this, but it got to the point where I was impressed when a force I was tearing apart with cannon actually switched to loose formation.

    Just give me good strategic map AI and I’ll be happy. Explaining to my friend that even though the English loved killing me more than they loved breathing at all that they were trapped on their musket-bristling island put him off the game completely.

  17. skinlo says:

    I don’t understand why people get so worked up about it. Ok, you didn’t like Empire. You don’t have to pre-order Shogun, but surely if all the reviews are great, why cut off your nose to spite your face?

    • Mad Doc MacRae says:

      That’s the icing on the disappointing AI cake. The reviews for Empire were glowing.

    • drewski says:

      The gist of his point is valid though – rather than saying ZOMG CREATIVE ARE SATAN AI IS AWFUL I’M BURNING DOWN THEIR OFFICES just wait until the game is released, if reports are that the AI is at least vaguely competent then you can buy it.

      You don’t have to trust the reviews; I’m sure the teeming millions will figure out the AI pretty quickly. They always do. But to throw the baby out before the bath has even been run seems especially futile, even by the standards of angry internet men.

    • battles_atlas says:

      Its familiarity breeding contempt drewski. The complaints with Empires were the same with Medi II (and to a lesser degree with Rome) and the promises made of Empires were the same as those now being made of Shogun II. People only get so pissed because TW is an awesome series. Its just CA appear hellbent on killing it by making the same mistakes every time, even though they’ve been acknowledging those same mistakes for years now.

    • Optimaximal says:

      There is the little niggle of CA using Steamworks – all the reviews missed or glossed over the bugs (bought reviews?) and customers who believed said reviews couldn’t return their horribly broken games that were destined to never be fixed, only replaced.

      Some retail purchasers got some mileage by returning their games under the ‘fit for sale’ clause and requested the game be removed from their Steam accounts, but it’s pretty bad form that Valve didn’t allow refunds for Empire, especially when they did for other cluster-fuck releases such as GTA4.

  18. Jimbo says:

    All he does here when he stops dicking about and decides he wants to win is flank the battle line with cavalry and then attack them from behind. This has worked in every TW game as far as I recall. The AI has never been able to handle this cheap Braveheart tactic and apparently still can’t. They need to account for this by making cavalry extremely expensive or anti-cav extremely cheap (but crap against infantry), or by making the AI a lot more adept at screening & countering cavalry with its own. There’s really no excuse for the AI ‘not noticing’ a squadron of heavy horse run right past its right flank, and then take up position behind its line without any interference at all.

    • Fumarole says:

      [...]cheap Braveheart tactic[...]

      You do realize that exploitation is the main strength of highly mobile forces, right?

    • Jimbo says:

      The main strength of highly mobile forces is that the enemy will just assume they are running away and therefore take no steps at all to defend against them? I respectfully disagree.

  19. trooperdx3117 says:

    I dont know the AI in Empire wasnt too bad and they did do a reasonably good job patching it up. Still though I think that to an extent its a good thing that they’re scaling back their ambitions for Shogun 2, that way they can nail the fundamentals before moving on to more ambitious stuff

  20. Harley Turan says:

    The simpler the rules, the more possibilities there are.

  21. M says:

    Oh god, why would they promise this. Why.

    • Gap Gen says:

      I’m thinking the AI might be unable to attack over land, or forgets to build soldiers, or something. Maybe if you offer it an alliance it will go into a blind rage and declare total war on you, down to the last peasant.

    • Ryan says:

      Oh god, why would they promise this. Why.

      Have we already forgotten the lessons of ’84

    • Saiko Kila says:

      Yeah, I remember that lesson. No matter how fucked up is the AI, there is always a good cure for it – a shotgun. Question is – to shoot AI or to shoot its creators?

  22. Daniel Klein says:

    Note that when they needed a quote about the AI, they went to the “creative director”. Since everyone in the gaming industry makes up job titles all day long I have no idea exactly what a creative director does, but I can wager a guess that he’s not very much down in the trenches, doing much heavy lifting in the code. Chances are, there is no programmer dedicated to the AI alone. This would be normal, from my understanding. You don’t usually have an AI programmer on your team. And that tells us a lot about how the AI is still being treated in most games. Something you kind of have to have, at a basic level, but that you don’t want to spend too many resources on.

    The better you get at games, especially at strategy games, the easier it becomes to spot how obviously dysfunctional the AI is in most of them. Civ5 still burns because it’s so recent, but yeah, the AI in Civ5 is so piss-poor that it spoiled my single player fun completely. I’m not talking “oh it didn’t optimize its expansion vs culture strategy optimally, thus allowing me to outpace it”; I’m talking “oh, so this is the eleventh archer you’re moving into striking range of my melee units this war. Really? Oh, you think you should suicide that horseman into my spearman instead of leaving him to defend your capital? Okay, then. Next game.”

    AI seems to be the promise du jour (possibly coloured by my own being discontent about AI especially these days), and I will not believe that someone actually invests sufficient resources into their AI until I can see it in action.

    Promises are much cheaper than programmers.

    • Bhazor says:

      Creative Director means Project Director means the Lead Developer means Head Honcho.
      So… yeah I think he might know what the studio is doing.

    • alh_p says:

      Mike Simpson may actually have had a hand in Empire’s AI. I won’t forget his developer post, 6 months or so after Empire’s release, saying that he was no finaly happy enough to let some of his friends play the game. While the release may have been forced by Sega, it was more than a little galling to know he was finaly ok with the game…

  23. derf says:

    One open-return ticket for the i’ll-believe-it-when-i-see-it train, please.

  24. Pidesco says:

    I’ll believe it when I see it, although I could wager that they’re full of it.

  25. Matzerath says:

    Perfect A.I.? It’s going to be self-aware?
    I’m scared.

    • Torqual says:

      In the year 2012 the software developer Creative Assembly created SKYNET. The perfect battle AI. No total war battle was ever won after by a human player. Die human scum.

  26. Taillefer says:

    I think the problem is they test it on large scale battles against basic moves. As though it’s assumed the player will charge forward with infantry for the classic meatgrinder in the middle, probably back them up with archers and send in some cavalry. It doesn’t seem as bad in massive battles because it’ll likely to do some damage even if it just marches everything towards you.

    It struggles massively on smaller battles (still in the hundreds) where it doesn’t seem to know where to position units, whether it should attack or defend, whether it should move a unit that’s under fire, thinks it’s a good idea to have half the army following a single unit of cavalry all over the map in a futile chase, leaves vulnerable units completely undefended, won’t leave a garrisoned building even if it’s collapsing with them in it, and on and on. I’ve won battles where my casualties are single or double digits and the AI lost hundreds even though my units were no stronger. Which, while I’d like to believe is because I’m awesome, is mostly because of indecisive AI unable to react.

    It may help to abstract it a bit more, make your orders broader, and actually give you less control. Tell a unit of cavalry to charge down the left-flank at your chosen target and off it goes, if you halt them mid-order it takes a morale hit. It sounds like a bad direction to go in, but the AI will never be able to compete with micro-managed units. It’s worth experimenting with something.

    • Luke says:

      I’d love to believe that they give it anything approaching this level of interest to be frank. AI has clearly, been last on the agenda at CA for some time, long behind kick ass graphics, and many features, and units, and 3D maps, and clouds on the strategic map

  27. EBass says:

    Well its good to see they have actually noted that the AI is a concern and want to fix it, but promises are easy to make and AI is extremely hard to make especially on stratergy games. I won’t consider a purchase until this drops price unless it gets best game ever reviews across the spectrum WITH the reviewers noting that this is truly an AI revolution.

    Still it shouldn’t be too hard to make a significant improvement, surely all it needs is a few lines of code to tell the AI not to make armies only of crossbowmen or not to sit 11 full stacks next to your capital while your empires edges are being slaughtered.

  28. Navagon says:

    The AI not doing anything dumb is one thing. The AI displaying a wide array of tactics (both flawed and not) is another.

    I welcome the effort they’re putting in. It should result in a better game. But AI is one area that has barely been scratched on in any game. I’m not yet convinced that Total War is going to go from trailing behind to leading the pack.

  29. Howl says:

    I’m not even fussed about enemy AI. I’d be happy with issuing my own troops a move order they can cope with, instead of them waiting until my back is turned and suddenly deciding to do a giant wheel manoeuvre that turns their flank to the enemy.

    The Total War slogan should be, “Where the f*** do you all think you are going?!?”

  30. Clayton Hughes says:

    This is supposed to be a reply to Dominic White but I’m not sure it’s going to end up in the right place:

    A few things

    1) “Perfect” AI in games (that is, AI that plays flawlessly) is often quite possible. However, it’s not very exciting to always play against an optimal opponent and only be able to win because of luck, or due to some inferiority in their style of play (even though it’s optimal). It’s much more interesting to determine a personality in an opponent, to see how they play and to try to twist around them. To find that bit of weakness and exploit it.

    1) Reversi is Othello. If you play Go like you play that game you will lose extremely badly.

  31. dsch says:

    “Solving Go is … about calculating all/most possible moves (and moves ahead) on a board and executing the best one. This is an “easy” problem, but requires tonnes of CPU time and RAM.”

    Absolutely not true. Not only are Go AIs nowhere near the top percentile of human players, no one has a clue how to make it approach the top human players.

    • Phasma Felis says:

      @dsch: No, he’s completely correct. You just don’t understand what he means by “easy”.

      Go is deterministic (no random elements) and as such the strategy for “solving” it is dead simple in theory: make a big flowchart of every possible move, and on each turn pick a flowchart branch that leads toward a victory condition (or at least a tie). Look at Tic-Tac-Toe, for example: you could write the entire decision tree for Tic-Tac-Toe on a sheet of letter paper if you wrote small.

      The problem is that writing out the decision tree for Go would take a piece of paper bigger than the entire universe, even if you represented each possible move with a single atom. So it’s easy to brute-force in theory but impossible in practice.

      Instead, we have to resort to less intensive strategies. So far, these strategies suck. That’ll change eventually.

  32. Gorgeras says:

    I will never buy another CA game ever again after Empire. Napoleon certainly didn’t fix the problems with Empire: it just apparently had less problems than Empire. How does that fix Empire which remains a broken and bugged mess?

    CA’s surviving fan-base consists of masochists and self-deluding liars. Reviews will continue to never factor in or even mention the problems with Total War games.

  33. pupsikaso says:

    AI doesn’t have to be “perfect”. It doesn’t even have to be good at all. It only has to appear good. The real hard part is to hide the smoke and mirrors. Once you’ve figured out what’s going on, the effect will be gone.

    Also, since when have we started believing developer’s promises? Especially from those that have been making the same game with pertier graphics for 10 years in a row?

  34. frags says:

    I’ll reserve judgement until I have bought the game. As for delays, I’m not sure what SEGA thinks of it. Remember Empire got three delays itself.

    • Novotny says:

      You are the perfect consumer. ‘I will buy it, and then find out if it is shit’. I’d really rather you didn’t, and CA had to consider putting some effort into the title in order to make sales. Buying stuff whether or not it’s shit doesn’t help.

  35. Ragabhava says:

    On every TW game so far I soon reached the conclusion that I woul have been ecstatical about it at the age of 8 – the time when I played with Playmobil and Lego whilst yelling ‘Kaboom’ and ‘PewPewPew’ and it was all about athmosphere – but for a grown man it’s too dumb an experience to not feel like some pure waste of time.

    Today, If I want a “cinematic experience” of medieval warfare in Japan I watch me some Ran. To hell with infantile interactivity.

  36. Mark-P says:

    So true. Every TW game so far has allowed you to win crushing victories with the basic ‘Pointy stick line in front, archers at the back, send the ponies round the side’ manoeuvre. Failure only becomes an option when the odds are stacked 3:1 against you. Combine that with a campaign AI that has been practically braindead in previous incarnations, and the games pretty quickly stop being fun and start to feel more like repeatedly punching a toddler. What’s the point in playing a strategy game against an opponent that can’t strategise? :(

    It’s exciting to hear that they’re at last prepared to admit that the AI was bad in previous games. I hope they can deliver a serviceable AI this time around. I skipped Empire after MTW2, and it sounds like they’ve burnt up more goodwill with the last installment.

    • JB says:

      Really, there was absolutely no need to play down the fun one can have punching a toddler. Uncalled for.

    • drewski says:

      By that logic, there’s no point playing any single player strategy game.

    • Fitt3 says:

      Yes, because there’s no way a single-player strategy game can have better AI than Total War.

    • Luke says:

      I love me some strategy gamin’ but I’m increasingly coming to the conclusion that bar a revolution the best challenge you’ll get is from a simple format/mechanics. The three rts games I enjoy most for single playe r’skirmishing’ just now are World in Conflict, Men of War and Dawn of War 2. The computer has very limited objectives and so will push you for them. You can still win but you at least know you’re in a fight and the strategic thought required and the simulation of battle creates enjoyment, for me at any rate.

      I used to love Shogun and Medival, and probably still would if they worked, for the same reason. When they went fancy with Rome onwards the ai was clearly confused and all over the place, and it took modders who understand this to fix them, sort of. This wasn’t possible with Empire as it was closed to modders and as a result it’s all over the place, still. Civ 5 is similarly having problems it seems in that the ai dosent understand the combat system. it’s difficult in this instance to even consider the computer an opponent. it’s just a stupid computer, and you’re wasting your time because a win is inevitable and the only thinking preventing it is hours of thoughtless action, removed of interest because of the absence of any sense of challenge.

      Simple objectives=no ai stupidity=no annoyance=immersion and thought=fun. Polish this up, as in World of conflict, with good presentation, and clear, passionate investment and belief in the format , and I’ll likely have a blast. This also describes the original Total War’s, age of empires, and COH and really all the classic and best strategy games, to a tee, I think.

      I like the idea of an epic, War and Peace esque Stratgey game but to be honest I think it’s impossible to do well just now, or anytime soon.

  37. bluebottle says:

    You don’t turn around a bad reputation with grandiose claims, especially when you’ve made those claims many, many times before. At this stage, what CA need to do to turn around that reputation, is release a game with at least passable AI. Maybe two, in fact.
    No mass fleeing the instant a battle starts, no armies of peasants and trebuchets, no generals charging into the front line of your phalanx, and no mysteriously missing naval invasions.

    Then we can start talking about perfect AI.

  38. Torqual says:

    The problem of the total war series is mentioned in your article. I do not know how many people are working an a total war title. But THE battle AI programmer implies that only one guy is working on the battle AI. So the article is hot air. The quality of the AI in total war titles did not raise since Total war Shogun. And with only one guy working on the battle AI it will never improve, because Creative Assemble does not want to spent money on that part. Their potential buyers are not serious strategy gamers, but people, who like nice flashy grafics and little tin men doing their battle thing on the screen.

    For me the total war series is only a nice looking battle screen saver with mediocre grand strategy parts. Its not for serious strategy gamers. I bought every title of total war since shogun. So i am not a serious strategy gamer, too? But i like titles like EU3 or HoI3 nether the less. But i am not really good at them.

    Have a nice day

    • drewski says:

      “Serious strategy gamers” play against other “serious strategy gamers”, or perhaps play hex grid based games.

      The Total War games don’t pretend to be anything other than entertaining ways for normal people to exercise their megalomaniacal instincts. They’re not, in any way, shape or form, meant for serious strategy gamers. In the same way that Settlers of Catan isn’t.

    • Fitt3 says:

      That, however, does not excuse the game’s AI being as shameful as it is. Even we non-wargamers want a challenge – that’s why we play video games in the first place.

  39. foop says:

    Hahahahaha. Yeah, right.

    Of course, I’ll still buy the game and be massively disappointed with the first three enormous patches before giving up in a huff.

  40. jti says:

    Didn’t they promise exactly the same thing with Empire? If someone really does something to AI I’ll buy, but I’ll wait for reports before I even consider it.

    • Fitt3 says:

      Yes, and they promised it with Rome. And Medieval 2. And Napoleon.

      So… yeah.

    • Junior says:

      Yeah, I too have no hope of them being able to follow through with their promise, again.

      I love the series, but it seems like the more features and gameplay elements they add, the more the Ai struggles to even put up a facade of intelligence, it’s always been fairly predictable, but in Rome it was still able to put up a bit of a fight.

      Empire is just laughable on the strategy map and the battlefield, the same stupid tactics work time and time again, I’ve lost count of the number of times I’ve take high ground from the enemy by flanking left until they pivot themselves off the hill.

      I’m at the point now of being a fanboy but not wanting to buy the game until it’s bundled with an expansionpack and heavily patched.

  41. Ballisticsfood says:

    My greatest fear is that they’ll do their job too well, and the robotic revolution will begin.

    Oh, wait. That would require the terminators to be able to walk through doors, wouldn’t it?

    • battles_atlas says:

      No need – Arnie’s model was very good at punching through walls, which makes doors rather a moot point. And the T-X had a built in cigarette lighter, so she could melt your pvc door frame down over the course of several hours.

  42. ThePinkNinja says:

    I loved RTW enough I will give them the benefit of the doubt until RPS get their hands on the game.

  43. Luke says:

    If you could port the ai of Civ into the strategic map, or really any 4×4 strategy game, into the diplomatic map, then Total War games would be a staggering piece of work. I don’t understand why they don’t poach a programmer form firaxis or paradox. That said:

    I was a good bit younger when I played it so maybe I was just less critical and easily overawed, but Medieval: Total War had fine AI. It wasn’t fiendish or anything but it was vastly more rewarding and challenging than any of the later iterations (although the Stainless Steel Mod for Medieval 2 did good work with a limited structure. ) It worked-being at war was a challenge and not an anoyyance, some were to big to handle and you had to be canny, others you won because you did the right thing, and the diplomacy, though basic, was suitably evocative and alive so as to be a consideration. Battles were challenging enough and would usually require thought and attention. If you didn’t do this, the computer would win. A computer will never be as intelligent as a human so this is a reasonable ratio for me. THIS WILL DO FINE.

    Empire was atrocious though: No naval invasions in a multi-theatre map? How was this missed? It wasn’t even challenging, it couldn’t even cheat properly. It was a horrible, boring, frustrating, ostentatious mess that bit off considerably more than it could chew.

    Medieval’s mechanics with Mediveal 2′s battles and it will be a country mile the best TW game. Any more complex than this and it won’t be feasible to give it depth and challenge, I daresay.

  44. jalf says:

    Another TW game, another promise that *this time*, the AI will be awesome.

  45. Anglocon says:

    Agreed bluebottle. Lets have an AI that can mount seabourne invasions, assult a fort without standing outside the walls and getting slaughtered … then lets talk about other improvements to the battle AI.

    This is a standard tactic from the CA marketing playbook. Give an interview saying how we saw the AI problems with the last game, this will all be resolved with the next game with our new procedures/whatever.

    The same play is used for the Diplomacy … its not been done yet for STW2 but its coming ….

  46. Humppakummitus says:

    Some ancient wisdom bears repeating:
    “Never trust CA’s claims about AI.” – Sun Tzu

  47. pipman3000 says:

    oh no :O they’ve canceled shogun 2?!

  48. ts061282 says:

    Go AI is master level on a 9×9. You don’t know what your talking about. Game AI can and should be better.

  49. Strategem says:

    Maybe the iconization an gridding of units is the answer. If we have sufficiently functional AI in PC boardgames / Hextile games etc, it would just take rasterizing a battlefield and assigning an average terrain property value to it as a modifier and then applying the usual rock-paper-shotgun unit type weighing and maneuvering.

    In short, if you just abstract properly and then unwind it again out of the abstraction layer, you’ve got a fairly simple / doable AI concept you just need to translate/retranslate, no?

    Just saying, since everyone keeps banging on about how “true” strategy games got it all nailed down and such. You can(and must, for display on a screen) square any circle, it’s just a question of detail..

    • Strategem says:

      To clarify: I meant as in doing all this WHILE keeping the graphics engine and representation intact.
      Just on a purely logical / AI level.

      Captcha looked like “Risen” just then.

  50. Kurt Lennon says:

    I wonder if they’ll pull a Wardell?

    “But… but… I thought it was ready! Honest!”