Making a Better Artificial Ignorance

Making a Better Artificial Ignorance (Full Version)

All Forums >> [New Releases from Matrix Games] >> WarPlan

Message

Omnius -> Making a Better Artificial Ignorance (11/30/2020 3:35:42 PM)
Just for grins I played the Axis Artificial Ignorance on the highest settings and found it to be seriously pathetic. There are a few things that could be done to improve AI play. The first major problem was what was done on turn 1 with the German subs. Putting them out near Scapa Flow in Fleet mode is the stupidest move of all! It took me two turns to wipe out both German subs with my two British carriers. The Axis AI should keep subs in Raider mode and go for attacking MS routes solely. I remember someone at Talonsoft who kept saying that a busy AI is a good AI. I found through many decades of experience that a busy AI is a pathetic AI if the game measures fatigue or effectiveness. All being busy does is exhaust AI units so they're always too pooped to pop effectively. I watched the pathetic AI make multiple attacks at bad odds losing large numbers of units because it just keeps attacking until it exhausts operations points. I rarely make an attack below 4-1, and am careful to limit bad odds attacks. In the USSR the German AI stupidly tried to send the panzers in between Moscow and Leningrad, out into the woods where there are no good targets to capture. Oh it managed to squeeze two panzer corps past my lines but I quickly surrounded and killed each of them, a total waste! Instead of trying to sneak through cracks in lines in the middle of nowhere the AI should have focused the panzers at Moscow or other cities of economic value. The Axis AI did manage to capture France from me at the very end of 1940, but by then it had no merchant shipping because I used the France '39 exploit of getting Italy into the war early so I could use my French fleet to trash Axis MS. I had the British capture Tripoli about that time. In 1941 while Germany got ready to invade the USSR I was invading Italy and was happy to see that Taranto is the magic key to getting Italy to surrender early. I managed to hold the AI to around the historical line in the USSR in 1941. In 1942 I easily withstood an German attack and then went on the offensive in the summer of 1942. I had the British coming up out of Italy into central Germany. I had the Americans and Canadians coming from France to meet up with the British. By September of 1942 I had the Soviets pushing the Germans out of the USSR and had managed a paradrop on an unguarded Ploesti on the last clear turn. By then it was really game over and decided I'd had enough of whipping a pathetic Artificial Ignorance.

AlvaroSousa -> RE: Making a Better Artificial Ignorance (11/30/2020 4:37:57 PM)
A.I. takes time to develop at a decent level for any game of this scale. There is a reason why players think the HoI A.I. is horrible or the Civilization A.I. is horrible. It is too difficult to create one for a game of this scope and size with complex rules. Even World in Flames has no A.I. because it is simply impossible to do for that game. Also consider the player. If you are an excellent WarPlan player who beats every human. What chance does an A.I. of an already complex game have vs you? There is literally no challenge. Of all the A.I. I have seen from these kinds of grand strategy games I'd say SC3 has the best one. But it is due to over a decade of modifying and improving. Even with SC's A.I. and with me being just an above average wargamer I managed to hold off the Axis in France till 1941 and kill a lot of their subs at the standard level difficulty level. WarPlan A.I. slowly improves as I get constructive feedback from players.

Omnius -> RE: Making a Better Artificial Ignorance (11/30/2020 4:43:45 PM)
I sure hope I've given you some ideas on how to improve the WarPlan AI. Cutting down on stupid attacks that lose strength for the AI without causing any or enough would be one way, make the AI more willing to reserve operations points to preserve that all important Effectiveness. Also put all sub groups into Raider mode from the start of a scenario. I agree that getting an AI to credibly play complex games is a tough nut to crack, I'm glad the programmer for WiF didn't waste time on an AI when there's so much else for him to program for optional rules and units. One thing I always found useful in designing scenario was a game that allowed for AI vs AI play. Put in that feature so you can better judge the AI play for each side. Plus it gives us a tool for testing out our designs if we do them.

Omnius -> RE: Making a Better Artificial Ignorance (11/30/2020 5:42:21 PM)
It would be best for the AI to always choose the Elite Specialty upgrade, it's my only choice for spending Specialty Points. It always gives a combat bonus whether on offense or defense so gives the most bang for the buck. The other choices tend to be dependent on the situation as to whether they provide a benefit or not. I noticed that the Axis AI doesn't build any merchant shipping. Once Germany loses all of it's MS then it loses that 25 iron ore resource input which definitely hurts the German AI. I'd think that building some MS would be beneficial to the Axis AI, I bet the Allied AI builds MS.

ncc1701e -> RE: Making a Better Artificial Ignorance (11/30/2020 7:32:08 PM)
I agree that the AI moving all the time is causing a fatal drop in Effectiveness. Sometimes, it is better to decide to do nothing...

Omnius -> RE: Making a Better Artificial Ignorance (12/2/2020 2:53:53 PM)
I would also slow down the AI during bad weather turns. While there is some effectiveness recovery during rain and heavy rain it's reduced if I'm not mistaken. During snow and heavy snow effectiveness is reduced, making winter offensives very costly on effectiveness as there is no recovery unless one wastes tons of trucks. I basically slow down during the fall to start effectiveness recovery and pass on attacking in winter so I recover more effectiveness in the spring without wasting a ton of supply trucks. Then I go full bore during the summer. Just isn't worth it to attack during the bad weather.

Protonic2020 -> RE: Making a Better Artificial Ignorance (12/4/2020 9:46:42 AM)
quote: ORIGINAL: AlvaroSousa A.I. takes time to develop at a decent level for any game of this scale. There is a reason why players think the HoI A.I. is horrible or the Civilization A.I. is horrible. It is too difficult to create one for a game of this scope and size with complex rules. Even World in Flames has no A.I. because it is simply impossible to do for that game. Also consider the player. If you are an excellent WarPlan player who beats every human. What chance does an A.I. of an already complex game have vs you? There is literally no challenge. A reason for why, if anything, i would want to get down deep and dirty with programming games these days as a backseat observer only of all games so called AI, not typing so called AI programming now since that would just sound like brute criticism of all who do it as a profession. But AI by definition should at least until it´s capable of playing a game and making smart decisions and learn to improve it´s game play NOT be called AI since intelligence by definition is to be able to evaluate it´s performance and be able to adjust to the opponent it´s engaging. To adapt and have a goal set for it to reach meanwhile having the tools to know it´s environment and be able to use and navigate within the environment as well as it possibly can. But i´m totally illiterate personally when it comes to game programming so excuse me? (Thanks) [sm=00000947.gif] quote: ORIGINAL: AlvaroSousa Of all the A.I. I have seen from these kinds of grand strategy games I'd say SC3 has the best one. But it is due to over a decade of modifying and improving. What game is "SC3" short for? I certainly will check it out if it rhymes with my preferences overall. I have to ponder on it, can´t be too difficult to figure out on my own now. (Yes it was, guessing it. I give up). [&:] Strategic Command 3! Perhaps, i´m on it....

AlvaroSousa -> RE: Making a Better Artificial Ignorance (12/4/2020 1:38:39 PM)
I use A.I. because it is a simple term understood by everyone what it means and doesn't necessarily encompass machine learning. Or I could say a long winded statement each time like patterned logic tree algorithm mapping and coding, or PLTMAC, and everyone would go huh? So I choose A.I. to sum that up. [;)] A.I. doesn't only mean machine learning. Big Blue was PLTMAC and not machine learning. Alpha Go is machine learning. And are you kidding me? I would LOOOOOVE to have machine learning for WarPlan. But that just isn't in the cards unless everyone had a Google Deep mind computer system. A.I. turns would take a week on a normal PC. SC3 is Strategic Command 3. Comparatively if I put my PLTMAC vs its PLTMAC we would be about even. Mine has better defensive controls through order maps His has better fuzzy logic controls and 2-3 tricks up his sleeve. SC3 is an excellent game with event systems and I would say better graphics. It has 3d models. But it is less complex than WarPlan. WarPlan offers more strategic and operational decision planning. Depends on your liking. I play SC3 on the side myself.

kennonlightfoot -> RE: Making a Better Artificial Ignorance (12/4/2020 2:23:10 PM)
Here is an idea I have been bouncing around between my ears when I was trying to come up with some way to put decent AI into a game I have been developing off and on. Implement the AI Player as a PBEM opponent. Have Matrix run a server with enough power to run a high end AI system and have it handle the AI Player end of a PBEM. It could be shared across all the games sold by Matrix to minimize the cost. It would allow the developer to uses a really powerful AI opponent in their games. And, it could have a dedicated programmer that really knew how to make the AI work. Or even partner with Google (Google AI) and have them provide the "AI PBEM Player".

Omnius -> RE: Making a Better Artificial Ignorance (12/4/2020 2:49:58 PM)
After experiencing the pitiful first perversion of Strategic Command I would never waste time or money on the following ones. The original versions of SC had one of the most pathetic AI's ever programmed and I doubt that SC3 was much better. SC series was a child's game, something paradox would puke out. While the AI in WarPlan is better than in SC it's still rather weak at providing a challenge without revving it up on steroids. Even then it's not much of a challenge long term.

Protonic2020 -> RE: Making a Better Artificial Ignorance (12/4/2020 3:21:07 PM)
quote: ORIGINAL: AlvaroSousa I use A.I. because it is a simple term understood by everyone what it means and doesn't necessarily encompass machine learning. Or I could say a long winded statement each time like patterned logic tree algorithm mapping and coding, or PLTMAC, and everyone would go huh? So I choose A.I. to sum that up. [;)] A.I. doesn't only mean machine learning. Big Blue was PLTMAC and not machine learning. Alpha Go is machine learning. And are you kidding me? I would LOOOOOVE to have machine learning for WarPlan. But that just isn't in the cards unless everyone had a Google Deep mind computer system. A.I. turns would take a week on a normal PC. SC3 is Strategic Command 3. Comparatively if I put my PLTMAC vs its PLTMAC we would be about even. Mine has better defensive controls through order maps His has better fuzzy logic controls and 2-3 tricks up his sleeve. SC3 is an excellent game with event systems and I would say better graphics. It has 3d models. But it is less complex than WarPlan. WarPlan offers more strategic and operational decision planning. Depends on your liking. I play SC3 on the side myself. Thank you for that explanation now sir, really. AI then, still. I tried to play smart with the best and it´s like i take on challenges IRL. Geronimo! Friskt vågat hälften vunnet, it means "he who dares wins half the victory", something like that. Cheers

AlvaroSousa -> RE: Making a Better Artificial Ignorance (12/4/2020 4:07:28 PM)
Quite welcome.

ncc1701e -> RE: Making a Better Artificial Ignorance (12/4/2020 8:40:38 PM)
quote: ORIGINAL: AlvaroSousa And are you kidding me? I would LOOOOOVE to have machine learning for WarPlan. But that just isn't in the cards unless everyone had a Google Deep mind computer system. A.I. turns would take a week on a normal PC. Well, I am not so sure. Did you try TensorFlow with Python? https://www.tensorflow.org/?hl=en System requirement is not so huge.

Protonic2020 -> RE: Making a Better Artificial Ignorance (12/4/2020 11:16:50 PM)
quote: ORIGINAL: Omnius By then it was really game over and decided I'd had enough of whipping a pathetic Artificial Ignorance. Yes, and no offense to any AI programmers. But system events and scripted stuff is where it is at today still obviously. I better wait for games that can actually do "machine learning". PBEM, play by e-mail right? Now how long for those games to progress plus i don´t like to work around others schedules. By the looks of it and considering i have got several holiday games to deal with already and learn this one might not hold up for decent single player and what i´m looking for.

Page: [1]

Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI

0.7011719