Lt. Kije Posted October 31, 2002 Share Posted October 31, 2002 OK. Since Bernoulli's Law of Large Numbers came about because people were trying to get a better grip on how to gamble successfully, Lt. Kije is now taking bets for how the "AI cheats?" massive thread is going to turn out. (No sense in making that thread longer and harder to understand by placing this vote inside it.) Here are the first set of choices. You may suggest other choices, which I may or may not add to this pool. Outcome A --------------------------------------- The eventual Treeburst-method finding is that the AI advantage is: 1. less than or equal to one percent 2. more than one, less than or equal to three percent 3. more than three percent Outcome B --------------------------------------- Treeburst, Cameroon, Maastrician, Warren Peace differing methods: 1. Turn out to give about the same picture of AI vs. human abilities 2. Generate results that are simply confusing 3. Generate results that differ in a way that spurs theory development. (e.g. Warren Peace isolates a situational variable that changes meaningfully the magnitude of the AI advantage. Via deductive inference from this critical finding, he proposes why the AI has an advantage sometimes, not other.) Outcome C --------------------------------------- The AI cheats? thread: 1. Continues as Peng has 2. Reaches a satisfying conclusion 3. Dies from sheer fatigue, in a muddle -- Lt. Kije Scorekeeper and Historian Link to comment Share on other sites More sharing options...
CombinedArms Posted October 31, 2002 Share Posted October 31, 2002 OK, I'm ready to bet a virtual wooden nickle on the following outcomes: Outcome A 1. Less than or equal to 1%--i.e. nada Outcome B 2. Simply confusing--hard to sort out the signficance due to differing methodologies (though all tending toward nada)--however, I concede that 1 & 3 are both possible. Outcome C I think that will vary with the participant. I'm already satisfied that there's not much difference between AI and human performance if the skewing variables are isolated, but some may never be satisfied and the Peng element will creep into anything that's truly controversial. [ October 31, 2002, 04:30 PM: Message edited by: CombinedArms ] Link to comment Share on other sites More sharing options...
Lars Posted October 31, 2002 Share Posted October 31, 2002 Dealer, I put everything on 300 posts and then the lock. Link to comment Share on other sites More sharing options...
CavalryMan Posted October 31, 2002 Share Posted October 31, 2002 Lordy knows. It just goes to show what happens when you exceed the critical mass of statisticians. I am just waiting for a pure mathematician to wade in and start demanding proofs.. Coffin 'Enry where are you? Link to comment Share on other sites More sharing options...
Apache Posted October 31, 2002 Share Posted October 31, 2002 16 times out of 20 when I asked this question the thread got locked. Therefore, the AI cheats Link to comment Share on other sites More sharing options...
mike the wino2 Posted October 31, 2002 Share Posted October 31, 2002 For what it's worth, not much BTW Outcome #1 <1% diff Outcome #2 Gotta be 3. No way to duplicate with the randomness of CMBB. This is gonna be long and ugly, with the occasional witty comment...where is Seanachi when we need him most? Outcome #3 It will BECOME the Peng thread for statiscians (sic)...they really need their own forum as evidenced by this thread Link to comment Share on other sites More sharing options...
Zarquon Posted October 31, 2002 Share Posted October 31, 2002 My nickel on C3. It has reached the point of statistics for statistic's sake already. I just imagine Steve and Charles reading this thread, shaking their heads a little, then closing their browsers and going back to work again... Link to comment Share on other sites More sharing options...
Sgt Joch Posted November 1, 2002 Share Posted November 1, 2002 my money is on C3, as Benjamin Disraeli said:¨"There are three kinds of lies: lies, damned lies, and statistics." Link to comment Share on other sites More sharing options...
Guest Mike Posted November 1, 2002 Share Posted November 1, 2002 My guess is that it will become a self-replicating thread limited to 300 posts perincarnatino, with much competition for inventing the title ofteh next one, next-to-no actual useful data, and a great place to go if you really need to know the significance of your 300th rejection for a date this week. Oh and there'll be people in there who are never seen anywhere else, or who type in funny accents or quote silly statements by otehrs in their sigs. But that's just my best guess!! Link to comment Share on other sites More sharing options...
Coffin 'Enry Posted November 1, 2002 Share Posted November 1, 2002 Originally posted by CavalryMan: Lordy knows. It just goes to show what happens when you exceed the critical mass of statisticians. I am just waiting for a pure mathematician to wade in and start demanding proofs.. Coffin 'Enry where are you?Right here, and I am staying well out of it. People in that thread are having fun, nothing that I can say will increase their enjoyment Link to comment Share on other sites More sharing options...
gunnersman Posted November 1, 2002 Share Posted November 1, 2002 Bell curve, what about the bell curve? Has anyone mentioned anything about a bell curve? I think it will end up in a bell curve. That is a handy statician tool isnt it. Yea! Thats it! Bell curve. The other Matt Link to comment Share on other sites More sharing options...
Eden Smallwood Posted November 1, 2002 Share Posted November 1, 2002 If BTS says it doesn't cheat, then it doesn't, AFAIC. Above and beyond given the befefit of the doubt to someone I don't know, say "Charles", the issue of whether the program cheats is in a very different context here- this is CMBB. This is not Civilization III, where I have no doubt that the program cheats- there is not even a "no bias" setting there. The desire of all is to have this game be completely historically accurate, as far as it can be done. So if the AI did cheat, it would no longer be an accurate representation of what that battle should have been like. It just doesn't "fit" the whole idea in this game. The AI could certainly be better, but to have it resort to cheating is clearly going down a road which leads to historically inaccurate behaviour. Quod Erat Demonstrandum, As Far As I'm Concernedum. Eden Link to comment Share on other sites More sharing options...
Diceman Posted November 1, 2002 Share Posted November 1, 2002 Unmentioned possible outcome: the thread only serves to demonstrate that to hit calculations are complex enough, with enough variables to require a huge sample to get the margin of error down to less than 1% Just my 2cents. Link to comment Share on other sites More sharing options...
Andrew H. Posted November 1, 2002 Share Posted November 1, 2002 I think that the patch will come out and the stats thread will fall into obscurity. Then 3 months later, someone will *bump* it. Link to comment Share on other sites More sharing options...
Bruno Weiss Posted November 1, 2002 Share Posted November 1, 2002 Ah excuse me, is this the bus to Hoboken... Link to comment Share on other sites More sharing options...
Michael Emrys Posted November 1, 2002 Share Posted November 1, 2002 Impossible to predict. The question has gone non-linear, bifurcated, and gotten completely chaotic. It was always sensitive to initial conditions and these cannot be determined since Warren won't tell us what he had for breakfast last week. Michael [ October 31, 2002, 10:29 PM: Message edited by: Michael emrys ] Link to comment Share on other sites More sharing options...
Warren Peace Posted November 1, 2002 Share Posted November 1, 2002 I had frosted mini-wheats! Link to comment Share on other sites More sharing options...
Terence Posted November 1, 2002 Share Posted November 1, 2002 The thread is garbage, and only I know why. Its because nobody is factoring in relativistic acceleration. Thats why. Link to comment Share on other sites More sharing options...
Michael Emrys Posted November 1, 2002 Share Posted November 1, 2002 Originally posted by Warren Peace: I had frosted mini-wheats!Well no wonder! If you'd only told us that at the beginning, there'd have been no need for all this pother. Another thought: You neglected to mention whether you had them with cream, milk, or dry. Also, did you have juice (designate kind), coffee, tea, or some other beverage (designate kind)? Fruit? Toast? [German professor accent]How can we reach reliable conclusions with incomplete data sets?[/German professor accent] Michael [ November 01, 2002, 12:25 AM: Message edited by: Michael emrys ] Link to comment Share on other sites More sharing options...
chrisl Posted November 1, 2002 Share Posted November 1, 2002 AI doesn't cheat. Why: Because the way it's most likely programmed is that the Strat AI sits in for the other player, and passes to the Turn Resolver/TacAI a set of orders just like a player would. Once the orders go to the Resolver/TacAI, the turn resolution code can't tell whether the orders were entered by Hiram, the StratAI, Fionn, or Charles (playing as the StratAI via hidden network code). It wouldn't make much sense for Charles to have programmed it any other way. Data suggest that AI probably doesn't cheat, but there are a few variables that never get isolated out that wash things out. (The Strat AI is slightly better than you think). Why:There's more going on in the game engine than people realize, and controlling out all the factors will be hard. People keep posting about the expected outcome, but we haven't seen a distribution of outcomes. If it were sharply peaked at the expected outcome we wouldn't be having any of this discussion, because the game would be nearly deterministic. It is most likely quite broad and will be take a while for the data to converge. That's why the game is fun and has a lot of replayability. It appears that Treeburst is starting to see the two numbers approach each other. The catch is that he shouldn't stop once they appear to converge, but continue taking data to see that they don't inexplicably diverge again. There is only one Peng thread, and don't start thinking you're going to go that way until you get to at least 10^4 posts. (as for the Global Morale issue-- you probably need about 10 times as many points in isolated buffer units as you have in tanks facing off in order to keep the impact of GM low. Otherwise you can still see some pretty large changes in GM ) Link to comment Share on other sites More sharing options...
WineCape Posted November 1, 2002 Share Posted November 1, 2002 THIS IS ALL BRUSSEL SPROUTS!! Charl Theron Link to comment Share on other sites More sharing options...
Michael Emrys Posted November 1, 2002 Share Posted November 1, 2002 Originally posted by WineCape: THIS IS ALL BRUSSEL SPROUTS!!Really? I was thinking it was more like kale or possibly spinach. But more like kale, I think. Michael Link to comment Share on other sites More sharing options...
rune Posted November 1, 2002 Share Posted November 1, 2002 *******The Place, a well guarded hidden complex deep in the woods of Maine. In one of the underground bunkers, the gang at BFC get together******* Charles: The AI is too weak, i must come up with something to balance out the play against gamey human players. Steve: We can give the AI secret power-up locations... Dan interrupts: Not with the power-ups AGAIN!!! Madmatt: [Cracking a Whip] Where is Rune at? Rune: [Hiding behind Moon] He went that way [Points to the left] Why is he here anyway? Moon: [Looks about innocently---well, as innocently as any CEO at a major corporation] Hurry this up, i have to take my plane in for its 3000 mile checkup...in Aruba. Charles: There must be something I can do to really upset the uber-grogs out there. Steve: I have it! Let the AI hit first exactly 1.83% more in the first hit! It will drive them absolutely nuts! ALL: [Evil laughing] Rune: [still hiding from Matt] Jefe' thought of this one, you can increase the penetration factor by Pi divided by the square root of the offset angle of armor multiplied by the thickness of armor, modified by the armor quality, and affected only slightly by the air density and mean operating temperature of the firing gun barrel. I don't know what it means, but that will confuse them all. ALL: Great idea! Aren't you supposed to be working on the scenarios? Madmatt: [Rushes in] There you are Rune! [Cracks the whip] Get back to scenario making. After all, you beta testers aren't testing, just playing... Rune: [slinks away-Whimpering] Yes Master... That is how it went...now you know... Rune Link to comment Share on other sites More sharing options...
Recommended Posts