AI improovements (Full Version)

All Forums >> [Current Games From Matrix.] >> [World War II] >> Steel Panthers World At War & Mega Campaigns



Message

nimu -> AI improovements (4/24/2001 5:36:00 PM)

Will Ver 5.0 improove also the AI?I think It should a bit.I'm not saying the Ai way of fighting is poor or stupid,but several times the computer acts in a too foreboding way.After one year af playing you always know where it's going to mass the troops.It also never uses special forces abilities,never paratroops launches or anithing else to rock yor defensive preparation.I'm not an expert,so i do not know if it's difficult or not to improove it a bit.You're doing a wonderful job guys.i do not deny that.But i think people that use to play in the single player mode will go crazy with some improovements with the AI.That's all(I swar yhis is not a complain!!!!). Thanks Best regards :D :D :D ;)



Brutto-Bob -> (4/24/2001 6:10:00 PM)

I think that AI is the worse element of SPWAW game. It's very predictable: CPU opponent always spread forces to attack the whole front line using a bulk of armors in the centre, never mount a strong and powerful attack on one limited zone to create a hole in the defensive line from where surronding the enemy. All units go straight to objectives without searching cover or height advantages. Never adopt a flanking tactic. All units start advance at the same time so engineers and recon infantry are overunned by armors that then blow on undescovered mines. AI do not use gliders or paratroops. To be onest, AI is the main problem of all strategical games but this is the very improvement area of SPWAW not OOB modifications or icons design.



Charles2222 -> (4/24/2001 8:05:00 PM)

Brutto-Bob: Sorry Bob, you are quite wrong on where the AI attacks, from what I've seen. I don't know why the complaint you make keeps coming up from various people and I've just never seen that going on. I typically play with three frontal objective areas (maybe if we all played this way your complaint would vanish) made randomly in campaigns. The AI "almost" never attacks all three of them at the same time, and does indeed attack any combination of the three on any given game. If you are playing a predesigned game, or are playing only 40 hexes wide, with only two frontal objective clusters, that may be why the AI does what you claim (BTW, I don't experience tanks bulked in the center either - only on occassion, just like their advances). I must say it would be fascinating to see an occassional unexpected airdrop from the AI, but I suspect the human player would mop it up so easily that complaints against the AI would mount still higher. [ April 24, 2001: Message edited by: Charles_22 ]



Brutto-Bob -> (4/24/2001 9:08:00 PM)

Uhmm... What I've writed is referred to defend scenarious in WWII CPU generated long campaigns. That scenary type is too easy to win with few losses. In these map the AI is very important because is called to make tactic choices while in others engagements simply react to human actions. And always, in these scenarious, there are a front line with two objective bunches nearly on the same vertical row. So AI build up an assault on both, as you have confirmed, so it's very easy to stop it with a mine field line and a statical defence. And, due to the vicinity of the opposite starting lines, border CPU troops before reaching objective are engaged by the defensive line causing the spread of the attack on the whole front line. Always human border units take out he thin line of foes and start to surround the centre ones, weakened by fight, to an easy victory. In Human assault vs CPU defence the way is similar. CPU build up a strong centre with pillbox, at-guns and armors. And protect the front with one thin line of mines. Once removed, is always simple to break through the map borders with fast units and encircle the defences until elimination (CPU never perform tactical retreats).



Paul Vebber -> (4/24/2001 9:13:00 PM)

IF we could write an AI that fought "intellignetly" we not be making games, but making millions form the DoD... Creating a remotely "intelligent" computer opponent is something that would take years of work, and is way beyond the scope of effort that can be given away. We have improved the AI in a number of ways - completely overhauled its buying habits, made it able to at least have a snowballs chance where bridges are involved and and made it a bit less predictable in its use of reserves. But if you want an intelligent opponant, there is no substitute fo a real human!



WW2'er -> (4/24/2001 9:55:00 PM)

quote:

Originally posted by Paul Vebber: IF we could write an AI that fought "intellignetly"
Ok, this is NOTa slam, but I just couldn't help but find humor in the statement above and it's spelling. Oh, the irony! :D Love ya Paul!...Really! :p :D WW2'er



nimu -> (4/24/2001 10:22:00 PM)

I agree with what Bob said.I never saw a totally unpredictable attack by the AI and never been struggled by a real "circle"attack.In difending i think AI acts better but it's clearly easier to do that.But i also understand that Paul is right when he said that cpu couldn't be as smart and intelligent like a human opponent.I just guess if it could be possible to make the cpu use better the "waypoints" screen when creating a battlefield and really use all the units in the right way:so,ingenieers in front of the formations to detect minefields,poratroops,special forces,sappers to blow up the bridges when defending.... Just a suggest.I love this game and i'd like to see it growin' up the more and more.Keep doin' like this.I thank u all Matrix team guys :D ;) :D ;)



Paul Vebber -> (4/24/2001 10:40:00 PM)

Yes..WHERE IS THAT SPELL CHECK :) Those of us with "typlexsia" and proofreading time could use one ;) Take the case of when do you blow up a bridge - think about all the ambiguous cases where a human may or may not try to blow a bridge and how to program that into a "ruleset" that the Computer can blindly follow! IT will be some time before we get beyond 'rule-based' AI and until we do, a human can figure out the 'rules' and use that to his advantage! We have greatly improved the way waypoints work - this will be improved in ver 5! [ April 24, 2001: Message edited by: Paul Vebber ]



nimu -> (4/24/2001 11:03:00 PM)

Ok,thank u Paul,i got it.



timc -> (4/24/2001 11:16:00 PM)

While I agree that the AI is too predictable, I also understand the limitations of programming the computer opponent. It is what it is. A word, though, on the subject of infiltration, gliders and paratroopers: All generated battles against the AI take place along the front lines (FEBA, Forward Edge of the Battle Area for the oldtimers. The map only spans a couple of thousand meters at the most in each direction. An airdrop that close to the front would be tactically preposterous and would be less historically accurate than anything the AI already does. Airborne operations should be left to scenarios in which a human operator can set reasonable objectives.



Charles2222 -> (4/25/2001 12:02:00 AM)

Brutto-Bob: Let's see if I understand you correctly. Are you saying that when you play campaigns, that it's always two objective clusters at the front? If so, there is an option to make that three clusters, with two in the back. I can't tell you what the option is that makes this possible, but it surely makes for more unpredictable advances from the AI (though always en masse).



murx -> (4/25/2001 12:17:00 AM)

I don't know in which way the AI is controlled - but if it's a kind of script maybe making the source available (the script not how to execute the actions) would help. So players can program own scipts that has a broader knowlege of actions. I never seen the AI pull back to lure me in, or try to time flanking maneuver more efficiently - leave alone in setup check for logical areas of LOS... the AI seems to love to put AA scattered in the backarea in the midth of forrests neglecting to put them at least at hilltops or the like. If You at Matrix Game are allowed to publish the script it might really help to improve it (at least it doesn't put naval units on land... :) ) murx



A_B -> (4/25/2001 12:19:00 AM)

I'm in agreement that the AI is pretty weak in the long campaign. I've put up several post on this in the recent past. A solution brought up to me is to have a 'partner' create a randam map scenario, choose and quickly deploy the opposing forces, and emial it to you. You can then plug it in to a long campaign, as the 'next' battle. Technically, it works no problem, if the OOB's are the same. This setup will also be much better than an ai could ever do, with objectives in reasonable places, such a the far side of a village, or the hills overlooking the crossroads. It would be some work, but probably the only way to keep a long campaign interesting. I've found a guy to partner with, once 5 comes out, and i'd suggest you do the same. Plus, you can pull some good pre-made maps to use as well :)



panda124c -> (4/25/2001 2:10:00 AM)

First let me make it clear that I feel it is easier to rewrite the entire game than to rewrite just the AI. However there are some 'tricks' that I have not seen in game that would make them more interesting. For example if someone were to program a game to remember how the human deployed his forces in, say last three games, and apply this as the basis for the AI's descisions on where to attack and what forces to use. This would force the human player to keep changing his tactics to confuse the AI. And just image if the last three games were played by someone else. :D These machines are too stupid to be of any practical use. :D



john g -> (4/25/2001 4:48:00 AM)

quote:

Originally posted by Brutto-Bob: Uhmm... What I've writed is referred to defend scenarious in WWII CPU generated long campaigns. That scenary type is too easy to win with few losses. In these map the AI is very important because is called to make tactic choices while in others engagements simply react to human actions. And always, in these scenarious, there are a front line with two objective bunches nearly on the same vertical row. So AI build up an assault on both, as you have confirmed, so it's very easy to stop it with a mine field line and a statical defence. And, due to the vicinity of the opposite starting lines, border CPU troops before reaching objective are engaged by the defensive line causing the spread of the attack on the whole front line. Always human border units take out he thin line of foes and start to surround the centre ones, weakened by fight, to an easy victory. In Human assault vs CPU defence the way is similar. CPU build up a strong centre with pillbox, at-guns and armors. And protect the front with one thin line of mines. Once removed, is always simple to break through the map borders with fast units and encircle the defences until elimination (CPU never perform tactical retreats).
I guess you just haven't played enough to see what the ai is capable of. I always play with 5 victory clusters and have seen the ai pull back from their front line to defend their victory hexes. Once as a German assault vs US I was able to infiltrate armor back to the rear 2 victory clusters taking out the sp artillery at the same time. The rest of the battle was a matter of defending those hexes against the ai as they attempted to retake them. In another British assault vs Italian ai, I had my core infantry decimated by the Italian 305mm naval guns at which time the ai moved all its infantry back to to the victory clusters where they piled up making my remaining armor useless against the threat of dozens of close assaults for each hex moved forward. As for minefields, in assault battles I have seen the ai put in as many as 3 rows of mined hexes with random mined hexes on roads behind that as well. Though the edge hexes always seem to be unmined. Of course if the random nature of the ai in placing units offends you, there are always the scenarios. Try Chiangs Nightmare with all the realism settings including C&C on as the Japanese. Let me know if you can take and hold all the victory hexes by the end of the game, I got all but one and I consider myself as a pretty good player. The ai doesn't play that scenario as well as a human gamer would, but probobly does as well as a Chinese general would have at the time. thanks, John.



Greg McCarty -> (4/25/2001 7:38:00 AM)

[QUOTE]Originally posted by Paul Vebber: [QB]IF we could write an AI that fought "intellignetly" we not be making games, but making millions form the DoD... /QUOTE] This is certainly the case. I have had some experience in progamming, and even brilliant coders will tell you that true AI; the kind we'd like to have, is still a long way off. The best use we can make of the AI as it exists now (and it aint bad by the way) is to incorporate it into human designed scenarios or campaigns which are designed with the AI strengths and limitations in mind. In most cases, this approach will work out quite well, and should be challeging enough for anyone.



murx -> (4/25/2001 8:11:00 AM)

The perfect AI ... Have a neuronal net (or simulate it, like in the games Creatures or the fresh Black & White). Only problem is Creatures had like 6000 distinct neurons (1000 per creatur) with up to 20 connections per neuron ... Now estimate the number of neurons for a strategical or tactical game, estimate the number of connections needed per neuron and realize that you need a large cluster to have the cpu-power to run the game. To have a comparison, the average human has about 100 billion neurons (10 to the power 11),a given neuron is connected to about ten thousand other neurons. (source http://www.phy.syr.edu/courses/modules/MM/Biology/biology.html ) So the way to have a 'better' AI is scripting - like in Chess games where it just draws of a LARGE number of successful move combinations (the source are chess games played by champions) and correct evaluation of the best move based on prior moves and the database. But then chess 'only' uses 32 'units' and the AI has only 16 units to choose. Now try this for 200+ units, terrain, smoke, firing arcs and angles, morale, hit probabilities and the chance of unseen units (and all this has to be estimated for the enemy units too to have the 'right' decisions). After all I still would like to have the scipting because a greater number of 'scripter' can produce much more scipts that apply different tactical situations leaving the main problem to have a script evaluating which is the 'best' script for the actual situation and when to change the used script to a better suited one. murx [ April 24, 2001: Message edited by: murx ]



Charles2222 -> (4/25/2001 8:50:00 AM)

john g: Well thank you, thank you very much. It seems you're the first one I've seen that would agree that the AI isn't doing the same blasted thing every time out. It seems a lot of people haven't played with the five objective clusters. Sadly, they don't know what they're missing. I never would've suspected that it would make a difference for the AI, but I knew it would likely make it tougher for me, so I took it.



rfox -> (4/26/2001 2:57:00 AM)

The AI does, in a sense, seem to create random tactical situations, Charles_22. The problem that I've seen, and this mainly applies to AI assaults, is that its randomness is its weakness. It has no 'purpose' when it is attacking, unless that 'purpose' is scripted in a carefully designed scenario. I don't want to get into mind/body relationships, philophical dualism vs. monism, the origin of human intelligence, or any of that stuff. However, a computer, as long as it is running a program, will never ever be able to make independent judgements even if you are using a 'neural net'. That's why scripted scenarios will always be the biggest challenge and the most successfull aspect of SPWAW. The scenarios involve a human mind telling a machine what to do under certain circumstances and in a controlled, pre-planned setting. I'm not really a programmer, but I wonder what it would involve to create a set of 'scripts' for the computer to use dynamically as chess programs do. The number of variables involved in SPWAW is just immense, though, as opposed to a chess board with relatively few. The thing that allowed Big Blue to defeat Kasparov wasn't its abillity to truely 'think', but its ability to orgranize and draw upon, and analyze patterns very, very quickly. They are patterns, none the less. John g makes a good point. Enough of my rantings...for now.



Charles2222 -> (4/26/2001 3:50:00 AM)

Rob Fox: There are more than a few problems I have with designed scenarios as compared to campaigns. First of all scenario design is often about making it so difficult that it's challenging. In other words, it's not so much what you do the first time playing it, but what you do upon playing that same scenario over and over. To me, nothing could be more boring. I'm stuck in a situation where my forces get blitzed with unbelievable amounts of air opposition (even in severe cold scenarios) added to mines all over the place, with the added hardship of having to win it in 20 turns. I hate that the only way I can win it, isn't so much based on skill, but on memorizing, like so many Super Mario games. If I can't do it in the first attempt I at least want to wait to try again until I've forgotten completely how everything worked. That's what I enjoy about campaining, no idea what will come at you, where it'll attack (unless you're playing 4 objectives, perhaps), who the enemy will be (in the case of multiple opponents on the front), and what the map will be like. My victories are usually plentiful, but if I tried playing a minor nation it wouldn't so much be the case. Try playing a minor nation while campaigning and you will know what challenge is, each and very time, with bunches of surprises along the way. Frankly, I don't think the common en masse attack by the AI is all that bad a strategy, but so many of us are into playing nations which have such great arms that the attack has not all that good of a chance. You can evidence this by how many wilt under the French S-35 attacks. It's challenging but not even close to impossible to win it, particularly since they give up so easily. Try playing Hungary against the Soviets, and I think the challenge would be even steeper. Maybe someday I'll try that, but my tactics need a little more honing first (don't dare try it until you have the new versions rules down pat).



rfox -> (4/26/2001 5:14:00 AM)

Charles_22: I agree with you for the most part. My only point was that I'm not sure it's possible to "program" an "AI" so that it behaves like a human. I would choose the long WW2 campaign every day of the week and twice on Sunday if it wasn't so, well, predictable (in a certain sense). Also, I enjoy playing qua the Germans so a German long campaign is easy at first against the Polish, etc. In fact, I DO almost always play the long campaign (I have one going right now) as opposed to scenarios over and over again. It would be fantastic if SPWAW had the cartographic skill of the National Geographic staff and the AI of Big Blue, but it is what it is (which isn't all that bad, actually). The great thing about the long campaigns is that they are random, and the tragedy for the AI is that they are random. One improvement that I was lobbying for and that will be included in v.5.0 is the force pool choices for the computer. Hopefully now, I won't see the computer getting 75% of the entire '41 Polish armor reserves for one relatively small battle. It's interesting when that happens, but completely non-historical. For me, that takes away a lot of the fun. One battle I played seemed more like a turkey shooting excercize than a challenge. As a side note, I do enjoy a well designed scenario. I view it more as an adventure than boredom. I agree, though, that the replayability of a conquered scenario is limited. Again, my only point there is that you actually have a human mind directing events (I'm not speaking from experience since I've never designed a scenario myself) as opposed to the "blind chance" of the AI in a long campaign. Perhaps one of the designers or programmers could weigh in here and let us know what sort of algorithm the program uses to choose its objectives in a typical long campaign scenario.



nimu -> (4/26/2001 5:15:00 AM)

Well guys,i read your last replies and i feel that all of us is right in a certain sense about complaining or not the AI.I think that if we only wait some days for the delivering of Mega campaign all our doubts about long campaigns will be shot down.If i well understood in M-C you'll have so much more briefing and high-level tactic decisions that we're gonna miss the boring old AI.M-C will be as challenging as a scenario but with the "pathos" and the atmospheres of a long campaign.And if it won't be enough......we'd better change game and try to call up Gasparov for a chess match!!!??? ;) :D



murx -> (4/26/2001 7:54:00 AM)

Rob Fox, However, a computer, as long as it is running a program, will never ever be able to make independent judgements even if you are using a 'neural net'. You are so totally wrong ... neural net is YOUR brain, it enables You to judge independent - to try learned things under completely new circumstances - to learn at all, to even try complete new actions. To get a grasp on how 'independant' - 'innovative' (using completely new or concerning AI neural net - not programmed 'as is' actions) try playing Creatures or B&W. Creatures was programmed with direct support from the Oxford Institute on Brain Research... so it's not just some fancy fuzzy logic programming but 'the real thing' - like a brain. I agree that human technology is not 'short' near producing real AI - but in less then 100 years I think we will actually face feeling and thinking AIs !! How we act then will be difficult... murx



Warrior -> (4/26/2001 8:27:00 AM)

quote:

Originally posted by Charles_22: If so, there is an option to make that three clusters, with two in the back. I can't tell you what the option is that makes this possible, but it surely makes for more unpredictable advances from the AI (though always en masse).
"Victory Frontage" button "On." You get 5 objective flags instead of 3. [ April 25, 2001: Message edited by: REMF ]



Charles2222 -> (4/26/2001 8:50:00 AM)

I must admit that the new version's rarity factor, when implemented, could make a profound difference in the conduct not only of the campaigner, but also of the forces the AI picks. Won't it just be so amusing to pick a platoon of say PZIVCs in core, and then find that another set may not be available? With deleted formations affecting the rarity pool, you better make sure that what you are deleting is something that's dime-a-dozen. I'm wondering if we'll be made privy to what the rarity factors are for each unit beforehand?



rfox -> (4/26/2001 10:21:00 AM)

quote:

Originally posted by murx: You are so totally wrong ... neural net is YOUR brain, it enables You to judge independent - to try learned things under completely new circumstances - to learn at all, to even try complete new actions. To get a grasp on how 'independant' - 'innovative' (using completely new or concerning AI neural net - not programmed 'as is' actions) try playing Creatures or B&W. murx
Murx- I could agree with you if I could agree to reduce human thought down to simply being physical processes like neurons firing. I realize the human brain is composed of a 'neural net'. My point was that even if you could manufacture something as complex as a neural net for a computer, it would still have to 'think' according to its program. As far as I know, no one is hypothesizing that computers will some day be able to program themselves. That would be somewhat of a catch 22, chicken and the egg, etc. Both B&W (which I've played) and Creatures can reduce the 'thought' and 'learning' taking place in the game down to programmed code. Deep down, they are still using 'if-then' processing. The thinking there is not 'ex nihilo' thinking. In the end, the silicon based monkey in B&W is doing what the programmers told it to do using a set of algorithms and following paths based on user input. It does a great job of masking an actual mind, but it isn't a 'ghost in the machine'. To get even remotely back to subject: in the end, I think that the best way to improve the AI in a program such as SPWAW would be to create a database of potential 'decisions' that the AI player could make which would mask true intelligence similar to certain chess programs. I'll stop now before I wax completely philosophical.



murx -> (4/27/2001 8:13:00 AM)

OK, to make it short - the mere physical function of a simulated neural net is more or less the same way a biological neural net work - with more or less simplification. So deep down the simulated neuranal net is bits and bytes - but deep down the human brain is just burning carbonhydrate and sending proteins and other mere chemical processess. Yet one neuron be it programmed or be it biological has abilities much different to the base it is made of be it bits and bytes or chemical messengers and electrical signals. So in effect the neuron of a human brain and a properly programmed one are the same (even though programmed neurons still suffer because they have only a finit value wheras the counterpart has more 'shades of grey' but is far from the simple ones and zeros - but that depends on the available memory and processing power). So in mere functionality a large enough neural net can achieve the same 'power' of a brain. But there are still some unresolved question like how 'intuition' is achieved - is it just a not conscious stream of thoughts (very fast and using random factors) that only makes the result conscious or is it something more then mere 'physical' existenz and a 'soul' exists. murx [ April 26, 2001: Message edited by: murx ]



rfox -> (4/27/2001 9:55:00 PM)

Murx: I'm hearing you and understanding you. I don't believe that human intelligence can be boiled down to physical neuronal activity, though. To put it bluntly, I think that there is a non-physical component to human thought and human decision making. Therefore, even if you could create a simulated neural net, it wouldn't function like a human mind since the human mind is not dependent on the brain. I hope that makes sense. Verstehen? Again, I would agree with you if I believed that human thought could be reduced to human brain functionality. Now we're venturing into religious and philosophical realms that I'm sure don't directly apply to SPWAW. If you would like to continue the discussion in another context, I would be happy to. Bottom line: I don't believe that a computer/programmed AI will ever be completely unpredictable or function like a human opponent. Yes, I do believe that an AI could be programmed to be challenging, even in SPWAW with its plethora of variables. Perhaps some day we will see improvement in that arena.



Page: [1]

Valid CSS!




Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI
0.75