Friar Bellows Posted August 21, 2002 Share Posted August 21, 2002 Maybe some of these things have been mentioned before: 1. The AI has trouble with transporting reinforcements from the USA to Britain. As the Axis player, put a couple of strength-15 subs, for example, in the region between Ireland and Britain and the US transports blindly suprise-encounter them and are wiped out immediately. That this happens once in the same turn is OK, but that it happens 3 times is not good. It's like the AI has a bad memory and forgets what happens during the course of a single turn. Maybe have the transports vary their course if the previous one caused pain? 2. When I have my Italian navy try to break out from the Mediterranean to the Atlantic, the Allied navy in the Mediterranean (I think they're near Egypt) totally ignores them, yet they should know about them, I think, because there's an Allied unit in Gibraltar. 3. The AI of land units needs to be improved. I think Hubert should concentrate most of his energies on this. I am well aware of the difficulty in designing an AI which can attack with reasonable skill, so that's why I think Hubert should concentrate on improving the AI defense. After all, I think most of us prefer to attack against the AI (e.g. We play the Axis in 1939/40 campaigns and the Allies in the 1943/44 campaigns). And with a static map, maybe Hubert can "babysit" the AI by creating defense "scripts" for it follow. I dunno, I'm just thinking out loud. Link to comment Share on other sites More sharing options...
Rediroc Posted August 21, 2002 Share Posted August 21, 2002 All good, but AI will never be as challenging as a human opponent. Link to comment Share on other sites More sharing options...
Bruce70 Posted August 21, 2002 Share Posted August 21, 2002 Originally posted by Rediroc: All good, but AI will never be as challenging as a human opponent.I made a very long post(s) on the CM forum about a comment like this so on this forum I'll just pose a rhetorical question "Why not?" Link to comment Share on other sites More sharing options...
Gunslingr3 Posted August 21, 2002 Share Posted August 21, 2002 I made a very long post(s) on the CM forum about a comment like this so on this forum I'll just pose a rhetorical question "Why not?" Lack of intuition. You can look at the map and imagine how it will be (or how you hope it will be) several turns hence. With a limited game, chess for example, there are only 64 locations, and a max of 6 unit types and 32 units at start. In a situation like this the AI can attempt to resolve the future via brute force - literally plotting out millions of combinations of future moves - and make it's own moves accordingly. Imagine how much more complex a relatively simple wargame like SC is. How many moves are possible? How does the computer know what moves are useless and should not even be considered? Hubert can answer definitively, but I'd be suprised if the AI in SC 'sees' past the possibilities of a single turn (it does seem hard headedly resolute in continuing amphibious invasions, but this might be due to the fact it already has the men on boats and each turn it 'sees' sending them in as the best option). AI will continue to be improved, but I think it's application is to roles of limited scale. AI is fantastic for quickly prioritizing limited options, but I've yet to see any demonstration of how AI could generate and implement strategy. I think it just works, turn by turn, for a more favorable position. I haven't seen it do anything indirectly clever. Advance two turns in one direction simply to create a larger pocket before swinging the gate shut, etc. In the end AI (thus far) simply represents the situations considered by the programmer, and responses he has prioritized. Gunslinger Link to comment Share on other sites More sharing options...
danscan Posted August 21, 2002 Share Posted August 21, 2002 I have been working on pattern recognition AI that would help in the AI world. I was able to create an AI that learned to play blackjack by both remebering what it did last time the situation came up (if the exact situation has come up before). It will also look for similiar situations and then use the same strategy it used then in the new situation and then store that situation in memory for next time. I think a more challenging AI could be made I just wish I could build an AI for a war game and have someone else handle all the graphic crap that I am just no good at handling. I have lots of ideas and frameworks built but I need to apply that two a game that has an open AI. Oh well. If you want the BlackJack game tell me and I will send you the code if you are so interested. Link to comment Share on other sites More sharing options...
husky65 Posted August 21, 2002 Share Posted August 21, 2002 Originally posted by Gunslingr3: [QB] So why don't you think a programmer can arrange for an AI to set short term goals in support of a long term strategy? AI will remain bad as long as people expect it to be bad and accept it being bad as a matter of course. (this is a general comment, not specifically directed at SC). Link to comment Share on other sites More sharing options...
Bruce70 Posted August 21, 2002 Share Posted August 21, 2002 Lack of intuition. You can look at the map and imagine how it will be (or how you hope it will be) several turns hence.Did you read the posts I made on the CM forum? That is exactly what reinforcement learning does. In fact there is no need to trust me, try reading this book by Sutton and in particular this chapter about planning and learning. This book was written a long time ago (in computing terms). Remember that the original statement involved the phrase "AI will never...", never is a very long time. Link to comment Share on other sites More sharing options...
Hubert Cater Posted August 21, 2002 Share Posted August 21, 2002 1. The AI has trouble with transporting reinforcements from the USA to Britain. As the Axis player, put a couple of strength-15 subs, for example, in the region between Ireland and Britain and the US transports blindly suprise-encounter them and are wiped out immediately. That this happens once in the same turn is OK, but that it happens 3 times is not good. It's like the AI has a bad memory and forgets what happens during the course of a single turn. Maybe have the transports vary their course if the previous one caused pain?What version are you playing? There was a small bug up until v1.03 that should have fixed this, if not let me know and I'll take another look. Hubert Link to comment Share on other sites More sharing options...
Gunslingr3 Posted August 21, 2002 Share Posted August 21, 2002 Remember that the original statement involved the phrase "AI will never...", never is a very long time. I'm talking about what is, not what we will create. I believe the AI in Strategic Command seeks opportunities referencing a list of prioritized goals (seizing a city, reducing a resouce value, attacking a unit) based on a combination of hard coded responses and cost benefit analysis. A competent human being differs by seeking not simply to seize the next city or kill the next unit within sight, but by envisioning a strategic outcome and then deducing how to reach it from the opportunities available. Those opportunities change from turn to turn, but the strategic aim tends to shift much less. Some opportunities present themselves and outweigh the existing strategy. Response still has to be prioritized. I'll give you an example. My first hotseat game was '39 as the Axis. My overall plan was to hit France, Russia, then England & America. I rolled Poland, Denmark and Norway hastily, seeking the plunder for financing a strong attack on France. The British responded to the invasion of Norway, but my aircraft stationed in Denmark were able to work in concert with the Kriegsmarine and inflict considerable losses on the Royal Navy. Knowing that the Royal Navy was reduced, and knowing that some British infantry were going to be lost in Norway I began to consider adjusting the order of my overall strategy, and fortunately hadn't spent from my plunder. Instead of beefing up the corps and tank armies for an early invasion of Russia I was going to maintain my core forces strength and concentrate on airpower to take advantage of Britain's early setbacks. To make a long story short, the plan worked. I was able to drop highly experienced troops with heavy air support into Britain and secure the island before America or Russia could effectively respond. I don't think the AI in Strategic Command operates in this broad of a scope. Perhaps I'm wrong, I would love to hear from those who know. In the meantime I'll make the suggestions I think can be implemented in Strategic Command on the Strategic Command board (like concentrating fire to kill units, not strategic bombing in the face of better targets, etc.) and leave the AI philosophy to the phd's. Gunslinger Link to comment Share on other sites More sharing options...
Friar Bellows Posted August 21, 2002 Author Share Posted August 21, 2002 Originally posted by Hubert Cater: What version are you playing? There was a small bug up until v1.03 that should have fixed this, if not let me know and I'll take another look. HubertAh great, I'm still playing with v1.02. Thanks for fixing it. I figured that it was the sort of thing that was fixable. Link to comment Share on other sites More sharing options...
Desert Dave Posted August 21, 2002 Share Posted August 21, 2002 A couple of comments on the AI: 1) I think that this was mentioned before, but one AI tactic that I would like to see changed -- when Germany has Russia on the ropes, they always seem to end up with too many Air Fleets and HQs, and not enough infantry. When you finally reach the Urals or Caucasus area, they are left defending vital areas with only the air and HQs. Perhaps there could be some kind of a ratio coded in, so that the AI wouldn't buy more Air Fleets without first checking to insure that they have enough Armies? (... this also seems to be true of USA and Britain -- an awful lot of Air, and not enough tanks or expedition armies) Or, place more corps in that area and just leave them so that their entrenchment is such that they could offer stiffer resistance. Those mountains around the Urals could be a truly tough area to conquer. :eek: 2) Also, could there be some kind of choice made so that once in a while the Allies will save up resources and invade North Afrika or even Italy, instead of always making a '42 invasion of France? Seems as though this ultimately fails every time -- it doesn't seem to require enough of a German defense so that the effort in Russian is hindered much, even when the Siberian reinforcement arrives. Link to comment Share on other sites More sharing options...
Friar Bellows Posted August 22, 2002 Author Share Posted August 22, 2002 Originally posted by Immer Etwas: When you finally reach the Urals or Caucasus area, they are left defending vital areas with only the air and HQs.On each of the two occasions that I reached the Urals, there were enemy infantry units there that I had to overcome. The first time, it took quite a while for me to force the Soviet government to move to Stalingrad (which I had ignored on my drive to Moscow). Link to comment Share on other sites More sharing options...
Guest Mike Posted August 22, 2002 Share Posted August 22, 2002 Umm....well one of the reasons why there are often so many HQ's left "at the death" for any nation is that you've bloody well killed everything else!!! Sheesh........ [ August 21, 2002, 09:58 PM: Message edited by: Mike ] Link to comment Share on other sites More sharing options...
John DiFool Posted August 22, 2002 Share Posted August 22, 2002 Originally posted by Mike: Umm....well one of the reasons why there are often so many HQ's left "at the death" for any nation is that you've bloody well killed everything else!!! Sheesh........True. I was thinking today about exactly what kind of Body Count (yes going ahead in time to Vietnam) the Germans need to inflict on the Russkies in order to prevail. The Russians typically get 450 MPPs or so per turn at the start, so you need to kill the equivalent of 4 1/2 corps (cheaper for Russia due to their Industrial Tech) per turn, which pretty much prevents them from assembling a reserve and/or building higher quality units than wimpy little corps. The key for the German then is taking (or taking and forcing the scorching of) those 3 mines which are close to the front. That drops Russia down to 360 MPPs or so a turn, and after that the attrition will weigh more and more heavily on them until their front cracks for good. In a human vs. human game defending these may be more vital than defending cities (since they are x2 their value in MPPs). JD Link to comment Share on other sites More sharing options...
Bruce70 Posted August 22, 2002 Share Posted August 22, 2002 quote: -------------------------------------------------------------------------------- Remember that the original statement involved the phrase "AI will never...", never is a very long time. -------------------------------------------------------------------------------- I'm talking about what is, not what we will create. I believe the AI in Strategic Command... Fair enough and I agree. It is likely that the AI in SC works "pretty much" as you suggested and this is not likely to perform as well as a human opponent no matter how much it is tweaked. It is sweeping statements about AI (not made by you) that I have a problem with. As Husky said "AI will remain bad as long as people expect it to be bad". The AI in SC is pretty good but we don't want Hubert to become complacent for SC2 now do we? Link to comment Share on other sites More sharing options...
Desert Dave Posted August 22, 2002 Share Posted August 22, 2002 As originally posted by Mike: Umm....well one of the reasons why there are often so many HQ's left "at the death" for any nation is that you've bloody well killed everything else!!! Sheesh........ True that you would normally find HQs and Air Fleets preserved, so that they would be the last targets, but I was thinking that the Urals and Caucasus tend to be poorly defended by infantry. Friar Bellows reports that he has seen infantry placed there, but in all the games I have played I have only seen ONE infantry corps guarding each area as the game winds down. What I was suggesting was for the AI to plan ahead and say something like this -- "hmmm, I am in danger of losing Moscow, and they are even coming 'round the bend toward my precious oil reserves, so I will forego some of the other reinforcements and begin building perimeter defenses for these two EXTREMELY vital areas." So that you would end up with the HQs and Air Fleets protected within a ring of entrenched corps -- see what I mean? Link to comment Share on other sites More sharing options...
Recommended Posts