AI Interesting Moves and Points of Improvement on Revised


  • Admin

    Played against AI on revised - @redrum it was pretty fun and some challenge : )

    There were a few interesting moves and some adjustments that could have made the AI stronger. Save game attached and below called out some of the interesting moves made.

    ai_could_do_better.tsvg

    Round 1: Germany uses a tank to capture archangel. There is only a single 50/50 change that it'll be an even trade assuming the tank can kill an infantry while on the defensive. Meanwhile AI then loses the tactical advantage of having a tank near the front line. Attacking the lone tank is a good trade for russia provided the air power and can do so with better than a 50/50 chance of winning.

    Round 2: Japan lost Manchuria with a factory in round 1 against improbable odds. Instead of retaking in round 2, it attacks Buryatia

    Round 3: Germany does SBR in Karelia instead of attacking BeloRussia with one infantry and bomber.

    Round 5: Japan buys an AA gun even though there are no bombers in range


  • Admin

    @LaFayette Thanks for the feedback and showing no mercy to the AI with the full map paint! Here are some responses:

    Round 1: Yeah, this is a tough one. I actually usually make that move with the tank myself and see a decent number of good players do it. While you probably lose a little bit of TUV initially you force Russia to use units/planes in another place. Another way to think about this as well is if you make Archangel worth 3 or 4 instead of 2 then it would probably be a no brainer to capture with a tank which I think shows how close this type of move is. I prefer the AI be a bit over aggressive rather than not take enough territories and this is difficult to perfectly balance across maps given different territory and unit values.

    Round 2: Yeah, that's really bad and geez you had quite some luck round 1/2. I think most Axis players would rage quit if they had the openings that germany and japan did 🙂 This has to do with the AI trying to attack too many territories and ends up aborting 1 of the attacks and in this case it was Manchuria (not a good choice). I'll take a closer look at this.

    Round 3: Agree. This one has to do with the AI wanting to attack Western Europe and Algeria so it deprioritizes other attacks. Then it decides it can't get quite enough units to attack Algeria and aborts that attack. Some of this behavior is caused by dice rather than LL which causes more randomness in the AI. But this is a good turn to train the AI on as there are lots of places it can attack.

    Round 5: The AA gun logic is pretty poor. It tends to be overly conservative and in this case sees the UK fighter in Caucasus as a potential threat where it could lose say Sinkiang or another territory within landing range of bombing Kwangtung. This could be improved a bit but not as high priority IMO compared to your examples in Round 2/3.


  • Admin

    @redrum another test game attached: vs_ai.tsvg

    AI had some pretty good play. Eventually Japan blundered in round 5 effectively handing the game to allies. AI could have made some other good moves, notes below:

    Round 1: Germany allows 2 AA to be in Germany, it may as well move the AA to eastern europe or to Balkans

    Round 2: single tank attack at likely negative net PU. Could conquer West Russia with tank and then retreat

    Round 4: germany used 2 tanks to attack Karelia, bad trade-off as UK can take the units. Losing 10 PU for max likely trade of 8 PU (Germany kills one infantry (+3), territory capture (+2), and likely kills one infantry on defense (+3))

    Round 4: Germany non-combat, moves big tank stack without infantry cover

    Round 4: Japan leaves undefended AA gun in India

    Round 5: 2 tanks and bomber attack archangel, should probably be one tank

    Round 5: BLUNDER: Japan divides fleet and loses most of it to a 99% combat with +60 TUV expected for Americans.

    Round 8: 2 tanks taking siankang, excessive.

    Round 11: Germany retake of Western europe with 1 arty and 2 figs. Should use 2 infantry with figs.



  • I'm not sure if there's interest in additional feedback about the AI - or whether the issues I'm reporting have already been fixed, but here's a saved game from Global 40 2nd e where there were a number of bad moves by the AI. This is still from 1.9, so if there are AI improvements in 2.0, this was not benefiting from them.

    For context, in this game all countries were Hard AI except for Germany and Japan which were being played by human players.

    Round 10:
    Japan completely eliminates a US fleet in Sea Zone 11 that the AI placed in a vulnerable location (US AI placed them there on turn 9).
    A similar thing happens on Round 16 in Sea Zone 14.

    Round 14:
    UK bombs Western Germany with 1 Bomber, which is intercepted by 4 fighters and dies. Seems like a very wasteful move.

    Round 16:
    US has 28 transports next to Gibraltar and 74 units in Gibraltar (of which 56 can be transported).

    It can land them in Normandy or Southern France uncontested. Possibly also in Northern Italy and Southern Italy with a small combat (assuming the British naval base can boost their movement).

    If 56 units are moved, the counter strike from Germany is only 44% with a TUV swing of -125 (i.e. greatly favouring US). Also, UK gets a turn before Germany's next turn and can reinforce with 2 planes and 4 infantry.

    Also, if a Southern Italy landing is possible, then there is essentially no counter-attack available to the Axis.

    I believe this was a must-do for the US player, but the AI did not take the opportunity. Instead, it moved its 28 transports away and left its 70 units stranded in Gibraltar for the rest of the game, with the Axis ultimately winning.

    Other things:

    • ANZAC was messing around all over the map, like sending troops to Africa and Russia, instead of doing something that would be more useful to it locally.
    • US send some troops to Australia at some point, for no reason that I could see.

    As I said, I'm sure there's lots of other blunders, but those were some of the biggest ones I noticed.

    Global_Oct14.tsvg


  • Admin

    Interesting, I've almost noticed that the AI "does not go for the win". It lacks a drive for a death-blow.

    For example, a recent game with allies on revised, AI Germany was doing really well after some bad Russian dice rolls and was pushing to Moscow. It had enough tanks where Russia could have been dead in 2 or 3 rounds. Instead of pushing tanks next to Moscow, building for a round and killing Russia, the AI started to divert towards the middle east for more PUs.

    It makes me think there should be more weight for advancing towards a capital and even more weight for capturing a capital. The subtraction of enemy PUs by not collecting income is very valuable, far outweighing say a +10 or +20 from capturing more territory.


  • Admin

    @Alexei-Svitkine AI testing is always helpful but doing it with the pre-release is definitely more useful as there are a decent number of AI updates there and stable release save games aren't compatible with the pre-release so hard to then retest. I'm primarily waiting til we release 2.0 until doing more AI changes.

    @LaFayette The AI doesn't really plan more than a turn at a time and prefers to attack anything that is beneficial immediately rather than building up for a larger attack. Advancing towards capitals and factories is already pretty high but that won't stop the AI from attacking things it can win immediately.



  • @redrum said in AI Interesting Moves and Points of Improvement on Revised:

    @Alexei-Svitkine AI testing is always helpful but doing it with the pre-release is definitely more useful as there are a decent number of AI updates there and stable release save games aren't compatible with the pre-release so hard to then retest. I'm primarily waiting til we release 2.0 until doing more AI changes.

    The purpose of the game wasn't to do testing, but to play a game. 🙂 I just shared it, in case it was helpful.

    We played on 1.9 because that's the last stable release. (For example, until a couple of days ago, 2.0 had a bug where the game could not advance past the Repair phase - which I just fixed recently in https://github.com/triplea-game/triplea/commit/fcf04c96796bac7bda8b586518b1087ef83009fc.)

    It's unfortunate that 2.0 save game incompatibility is blocking AI improvements.

    Have we considered having 2.0 be able to load 1.9 saves? For example, via a helper .jar built from the 1.9 code that can export to 2.0 format?



  • Okay, I have a save game from Great War map on 2.0 (so save game compatible), where AI played badly.
    For context, in this game, Ottomans and Austria is played by humans and all other countries played by AI.

    Here are some particular dumb things:

    • Early on - e.g. Round 8 - Germany could attack US transports with some subs and sink troops. Yes, transports defend on a 1 on this map, but still, its worth it to sink troops. Also, US should be defending its navy better.
      Screen Shot 2019-11-02 at 7.48.28 PM.png

    • In Round 8, US builds a plane in US. There are no ACs on this map and that plane literally cannot make it out of there. Later in the game, it builds another.

    • In Round 10, Russia built some boats. Why?

    • In Round 11, Germany attacks with a bad force. It has lots of Artillery and infantry, but chooses to attack mostly with infantry. In general, it seems AI was not pairing up Artillery and Infantry in this game - and just left lots of Artillery behind - and moved them into the same territory after in non-combat. Why? (20 Artillery left behind:)
      Screen Shot 2019-11-02 at 7.39.57 PM.png

    • Round 13: Germany does a dumb attack on Crimea when Russia has a strong counter attack
      Screen Shot 2019-11-02 at 7.45.46 PM.png

    • Round 18: Russia is fighting for its life. It decides to buy Cavalry instead of defensive units.

    • In Round 20, lots of Artillery left behind. Attack with 22 infantry and 2 out of 31 artillery. WTF? (The red army is Ottoman who is on the same side as Germany, so no counter attack there.)
      Screen Shot 2019-10-31 at 8.38.53 PM.png

    There's a bunch of other silly stuff in there, but those are some of the most dubious.

    Some other things:

    • AI was basically never buying planes in Europe.
    • Western front battles were very wonky (between France AI and Germany AI). I think neither side played well.
    • Germany spent lots of resources building fleets in the Mediterranean where it was not useful most of the game.

Log in to reply