A&A and G40: AI Critical concepts and foundational principles of gameplay

  • '19 '15 '14

    Lately I have been playing against the Machine, and the HardAI that redrum is developing. It has been forcing me to think a lot about the sorts of things that we do as players in order to beat each other. The kind of higher level strategic or conceptual stuff that we use to win. It occurs to me that it might be helpful to start with just a big list of concepts, as if you were teaching a 101 introduction to Axis and Allies global course. I feel like the people playing Global and frequenting these boards are top drawer generals, and the most likely to understand what I’m talking about here, and to provide feedback for priorities.

    What sorts of subjects would you highlight?

    Here some ideas I was ruminating on. Beyond just the basic rules and unit values, thinking more about key concepts starting out of Classic and some of the more nuanced stuff since AA50 and Global…

    Infantry Push: Understanding the primacy of the infantry unit, how to maximize their placement, how to stack them, trade them at advantage, and push them to the front. Stack advance, using the defensive power of infantry relative to its cost, to prevent the enemy from trading territory. Currently the AI does this quite well.

    Transports: how to move ground units with transports, launching and shucking, from the main production centers and “safe zones” to the front. How to mass transports for largescale amphibious assaults. Also how to defend transports, eg the relationship of warships (as largely subordinate) to transports, with the warships in the escort role.

    Fodder: the concept of using weaker units to absorb hits on defense or attack, to protect more potent heavy hitting units.

    Trading territory and Deadzoning: know how/when to risk weaker units to destroy enemy units or take territory. Also knowing how to threaten prevent the enemy from moving “too close” by threatening larger counter attacks. How to use Aircraft in this capacity, and exploit their unique advantage of “not landing” in a newly conquered territory.

    Strafing: how to destroy enemy units and then withdraw from combat so as not to leave your own units at risk. Knowing when its better to withdraw from a battle and preserve units from counter attack, rather than taking the territory for the money. Also, how to use strafes to exploit the movement advantage on retreat (when attacking from multiple territories.)

    Blocking: sacrificing units to prevent enemy movement, either with blitz units or ships.

    Turtling: when to withdraw from a territory or several territories, in order to mass your forces in a “more important” territory. When to stack rather than attack, and how to play like Fabius Maximus vs Hannibal, winning by preserving the army rather than engaging the enemy.

    Can Openers: how to exploit the turn order to open up opportunities for your teammate on attack.

    Tank Drives: using the movement advantage of armor to pressure key enemy territories/factories. JTDTM (Japanese Tank Drive to Moscow) being the prime example.

    Production: where and when to purchase new production (when not to) and how to exploit the existing production to the greatest effect when buying and placing new units. Also how to disrupt production, either by direct attack or via SBR.

    Fighter Transits and air defense: how to exploit the turn order and position aircraft to help your teammate on defense. Especially using aircraft to support your teammate in a newly conquered territory/factory.

    Double team, Triple team: exploiting the turn order to gang up on a single opponent over the course of a round, or a multiple rounds.

    Artillery: how artillery can be used to counter the infantry push mechanic in games since the unit was introduced.

    Managing the Center: appreciating the importance of “the middle” of the map to the endgame for both the Axis and the Allies, but especially the Axis.

    Managing the Periphery: how to take peripheral territories gain the most ipcs for minimum investment in units. The prime example being a region like Africa on most boards.

    Carrier logic: how to use Carriers and Aircraft in conjunction to maximize their effect on either attack or defense, but especially on defense.

    Destroyer logic: how to use this unit against submarines, for blocking, and as fodder defense.

    Massing Air, and Air attacks vs Ships: using large numbers of aircraft in conjunction to destroy units especially navies. Exploiting the fact that air hits can only be assigned to surface vessels if no destroyer is present.

    Air Blitz: Sacrificing Air units to achieve Objectives, Victory conditions, or Capital capture: Basically knowing when it is advantageous to destroy more valuable units in order for a ground unit to take the territory.

    Sacrificing naval units to preserve fleet defense: against counter attack the following turn, or how to retain a movement advantage. Like understanding when its better to kill a fighter on attack than a carrier deck, or when its better on defense to kill a carrier deck before a fighter. Things like that.

    Scrambling: When to risk fighters on scramble defense, and how to position fighters or airbases to exploit the move on defense.

    Repair for harbors and factories: where to buy naval bases, and when to retreat from attacks in order exploit the repair. Or when to how to minimize the repair costs at factories from SBR, relative to the units you need to maintain your ability to “fight on.”

    Capital capture, and post Capital capture games: understanding the primacy of the capital for income collection, and knowing how much to invest in defending a capital before the cost outweighs the advantage. Basically knowing when to bounce! But also how to trade Capitals, or when its better to avoid liberating a capital but instead to let your teammate take your land.

    What else?

    I know there are a hundred different things, but just for the sake of a list. I feel like once we have enough stuff down, it will be easier to parse and see what sorts of behavior makes for good gameplay in A&A. Then provide input to the machine in those areas, using the broadstrokes at least. The end hope would be an AI that can beat humans consistently, or at least be entertaining. Right now it’s understanding of the infantry push is very adept. It also has the benefit of calculating on the fly, which, if you’re a human like me who doesn’t consult a calculator, the AI will always have the edge in this area. Any feedback you guys have would be great. I know the forums are replete with discussions of these concepts. Let me know what things should get a highlight from your perspective.

    Also, along these lines, it would be nice to create a mini lexicon of common A&A terms, acronyms and abbreviations. Has that been done before? If so it should be stickied. If not we should start compiling one, so new people can know what the hell we’re talking about when we start speaking in all these technical A&A krytopograms  😄

    Thanks!
    Elk

  • '14 Customizer

    Sacrificing units to achieve Objectives, Victory conditions, or Capital capture: Basically knowing when it is advantageous to destroy more valuable units like (aircraft) in order for a ground unit to take the territory.

    I think this should be called Air Blitz

  • '19 '15 '14

    Agreed. That’s a better way to put it. The sort of thing you see a lot during the climax.

    I edited the above to read Air Blitz. Good looking out


  • Abuse of Victory Conditions. Knowing when a disadvantagous attack on a VC still means winning the game. Disadvantagous as in: battle itself can be won but will be crippling in the long run. Still, very high chance of winning the game by occupying the VC without the other side being able to do something about it the next turn. Closely related to can opening and blocking concepts.

    Example:
    USA went KJF, so Europe is a German dreamscenario. 7 VCs are already under Axis control, including Moscow. The IJN, IJAF and IJA (in ~10 transports) are on the run for the USN and fled into the ME area. Now they kamikaze-attack Cairo where all invading Japanese units are destroyed. So very disadvantagous for Japan, but if Germany (or even Italy after a similar German kamikaze) can grab Cairo and hold it for just 1 turn, they win the game.

  • '14 Customizer

    @ItIsILeClerc:

    Abuse of Victory Conditions. Knowing when a disadvantagous attack on a VC still means winning the game. Disadvantagous as in: battle itself can be won but will be crippling in the long run. Still, very high chance of winning the game by occupying the VC without the other side being able to do something about it the next turn. Closely related to can opening and blocking concepts.

    Example:
    USA went KJF, so Europe is a German dreamscenario. 7 VCs are already under Axis control, including Moscow. The IJN, IJAF and IJA (in ~10 transports) are on the run for the USN and fled into the ME area. Now they kamikaze-attack Cairo where all invading Japanese units are destroyed. So very disadvantagous for Japan, but if Germany (or even Italy after a similar German kamikaze) can grab Cairo and hold it for just 1 turn, they win the game.

    I did that exact maneuver in a league game and won. I called it the “throw Japan under the bus” maneuver. I even took planes as casualties to maximize my occupying force of land units.


  • It will be verry hard to program an AI that takes care of all these things.

    Because it is not always the value of the units but also the location and replacabitiy of the units that makes their value.
    Also the force projection of the units should be taken into account, sure if i scramble 3 fighters i can take 4 of him with me ( so i win on value ) they are on a factory so i can replace them easy but i cannot attack his 10 defenceless transports this turn for example.

    Also this counts for when you attack as well attack now because i can get his heavy hitters or dont attack because he wont be using them anyway.
    It is like playing chess with random factors blended in your AI will have weak spots unfortunately and people will abuse them.

  • 2021 2020 '19 '18 '17 '16 '15

    Ok. Kind of besides the topic……  The hard AI is
    1. waaaaay better than medium AI
    2. Builds too much navy as germany
    3. Does not build Chinese units, just collects money
    4. Leaves transports and bombers unprotected
    5. Often no war between soviet and Germany, leaving east germany / west soviet unprotected
    6. Often no war between uk and Japan/ Japanese does not take DEI
    7. AI attacks true neutrals waaaaay too often.

    Good things. Does alot of unpredictable stuff. Not all of it is bad. Makes you reconsider the established truths


  • @cyanight:

    @ItIsILeClerc:

    Abuse of Victory Conditions. Knowing when a disadvantagous attack on a VC still means winning the game. Disadvantagous as in: battle itself can be won but will be crippling in the long run. Still, very high chance of winning the game by occupying the VC without the other side being able to do something about it the next turn. Closely related to can opening and blocking concepts.

    Example:
    USA went KJF, so Europe is a German dreamscenario. 7 VCs are already under Axis control, including Moscow. The IJN, IJAF and IJA (in ~10 transports) are on the run for the USN and fled into the ME area. Now they kamikaze-attack Cairo where all invading Japanese units are destroyed. So very disadvantagous for Japan, but if Germany (or even Italy after a similar German kamikaze) can grab Cairo and hold it for just 1 turn, they win the game.

    I did that exact maneuver in a league game and won. I called it the “throw Japan under the bus” maneuver. I even took planes as casualties to maximize my occupying force of land units.

    :-).

    I did a ‘reverse maneuver’ once, during a KGF. Took the entire German army from Bryansk and marched onto an already isolated India to take Calcutta. That allowed Japan to focus more on the pacific itself and take Hawaii very early and build up pressure on Sydney, SF/Mexico/Panama and Alaska. Convoyed ANZAC into a non-existent income of 3IPCs/turn anyway.

    So I kinda threw Germany under the bus FTW ;-).

  • '19 '15 '14

    Great stuff! That last concept of understanding the Victory conditions, and knowing when to throw everything at it is especially important for global. Though technically it should be the same for all games since Revised, even if a lot of people ignore VC wins and go for unconditional surrender on the basic boards. I think respecting the VC win will be important for the AI games. Basically if the HardAI sneaks in an “Honorable Victory” on you, then you should concede to the machine, respectfully, without flipping over the table 😉 Hopefully HardAI can be made to gun a bit harder in such situations, and know when to throw everything at the win. Whether its by VC techinical wins, or just the normal capital capture exchange, something that a human will take a high risk on would be nice to emulate.

    I guess also for G40, understanding things like the DoW is pretty critical too.

    One other thing, more generally for massing or magnifying unit abilities by grouping them in large numbers. In other threads I’ve called these…

    Magnified Builds or grouping units of the same type together: Basically the difference between having a couple bombers scattered here and there, and a dozen bombers all in one spot all focused on one target. It works for any unit type, whether fighters or tanks or artillery or infantry, or at sea with Carriers and the like. The AI actually seems to highlight this already, as they will often stack like that, with their Air, at least on attack.

    I suppose the alternative would be appreciating something like…

    Mixed Builds or grouping units of a different type together: knowing when its better to have mixed forces, and exploit a different kind of magnifier, when you have less total units in the expeditionary force. Like for example the mech + tank + planes combo, or dd + cv + transport combo and things of that sort. Building basically at different scales… where, if its a smaller force, then the mixed type works better, but as the force gets larger, then the power of having one unit magnified can be more potent. Not sure if I’m phrasing that correctly, but something about it seems somehow on target. Basically the Minimum mix necessary for a unit group to be effective on its own, vs how to take that mininum mix and then Max it out, by adding a stack of a particular unit to it.

    I think what both of those ideas come down to is know when you can split forces up, and when its better to keep them together, and how to support either approach through building strategy. Either piecemeal or all at once. Along those lines, another concept related to production might be…

    Saving IPCs for the heavy drop Knowing when _no_t to spend every last dime right now, but instead save up for a multi-round purchase. This has been a popular strategy for humans, especially with Nations like UK, or sometimes Germany (sea lion focus, or late game entry into the Naval arena). Usually this is when buying the same units little by little would put your newly purchased force at risk of destruction, so instead you save all your cash for a round, and then “drop them” all at once. Thinking here especially of UK Carriers with US fighters, or German transports etc. Another component of this type of thinking, would be “not showing your hand, until you’re ready.” USA often does this, if they want to keep the opponent guessing about which theater they will go after.

    As Shadowhawk notes, its pretty hard to get the AI to appreciate all this stuff, but I still think its helpful to think about and to try and tease out general concepts, from the more map specific strategic stuff. And who knows, maybe someday the AI will be schooling all of us. I mean hell, just look what Google DeepMind is doing with those old Atari games! There are probably “Pong Tunnels”, or Centipede style “Mushroom Blobs” for A&A too, and we just haven’t gone through enough iterations to see them all yet hehe. Even now, looking at the way HardAI transports ground and pushes their stacks can be fascinating to observe.

  • '17 '16 '15

    I think you’re phrasing it just fine.
    Unit concentration could be another way of wording it. Based on the old adage “concentration of force” . 🙂

  • 2020 '19 '18 '17 '16

    In both 1940 and 1942 Globals I’ve found Hard Japan sells out on the Pacific to go all in against China and Russia with mainland IC builds and huge concentrations of air units. In 42 it’s especially tough since Germany will go all in as well against Russia and Japan starts out with a huge income edge that must be addressed immediately by the US.

    In 40 it is interesting to see the Hard German AI track the UK player’s builds to determine whether to make massive investments into Sea Lion. It’s not particularly creative in building a fleet, it just believes in battleship escorts for transports. What is odd is sometimes it will attack London or Scotland but if their defenses are relatively strong it will just send the fleet into the North Atlantic with fully loaded transports begging to be attacked.

  • '19 '15 '14

    Might interest some people here that the latest pre-release Jar for hardAI includes scrambling logic. See the AI dev thread if you want to download it and give it a go. I believe Redrum is working on fighter landing and transports next. Have fun all

Suggested Topics

  • 1
  • 1
  • 10
  • 15
  • 1
  • 40
  • 1
  • 3
I Will Never Grow Up Games
Axis & Allies Boardgaming Custom Painted Miniatures
Dean's Army Guys

92
Online

15.1k
Users

35.9k
Topics

1.5m
Posts