Install Steam
login
|
language
简体中文 (Simplified Chinese)
繁體中文 (Traditional Chinese)
日本語 (Japanese)
한국어 (Korean)
ไทย (Thai)
Български (Bulgarian)
Čeština (Czech)
Dansk (Danish)
Deutsch (German)
Español - España (Spanish - Spain)
Español - Latinoamérica (Spanish - Latin America)
Ελληνικά (Greek)
Français (French)
Italiano (Italian)
Bahasa Indonesia (Indonesian)
Magyar (Hungarian)
Nederlands (Dutch)
Norsk (Norwegian)
Polski (Polish)
Português (Portuguese - Portugal)
Português - Brasil (Portuguese - Brazil)
Română (Romanian)
Русский (Russian)
Suomi (Finnish)
Svenska (Swedish)
Türkçe (Turkish)
Tiếng Việt (Vietnamese)
Українська (Ukrainian)
Report a translation problem
This topic has been pinned, so it's probably important 
Discussions Rules and Guidelines
19 
Report this post
Revised is the reason TripleA exists and probably Abattlemap too (which was a popular PBEM platform before TripleA), because there was just nowhere to play that game online. It also had a relatively low bid for Axis, and some fun strategic options for trying to set up a Pacific play. It still had most of the good Atlantic shuck shuck moves of Classics, but a different dynamic in many other areas of the map especially the Eastern Front. It had technology and introduced the artillery unit which was a huge game changer compared with Classic. Ultimately the Revised game still hinged on the JTDTM (Japanese tank drive to Moscow), vs the KGF press (Kill Germany First), where Allies try to push on Karelia/Baltic States and set up killing plays on Paris or Berlin while trying to stack Western Units in the USSR. Those could both feel a little gamey after a while. But it also did have an actual KJF (Kill Japan First) strategy that Allies could employ if they were slick, or a way to stall Japan with a pacific game while not totally giving up on Europe. It was pretty well balanced and had a good run. And yeah, the A&A Online color scheme does remind somewhat of Revised. Especially with the Blue-Gray of the German units, though I'm glad they didn't go quite so dark on the sea zones hehe.
Some players who collect sculpts for the boardgames still prize this Revised edition for the lime green British units, and the uniquely colored German units.
I still think that AA50 was probably Larry's finest mid scale board, because you just really got the sense that he took his time with it. The boxed materials were beautifully presented. It had stuff like unit boxes for each nation with the set up cards printed on them, cash money, a bunch of cool special rules, two start dates and the like. Most hardcore players moved on to Global, but I still really enjoy the AA50 game. Especially the 1942 start date for that map, which was kind of neglected, since most people were playing the 1941 start date when it launched.
I really like the idea of a Legacy Edition of A&A that includes all the major titles in one place. In the box and online. If I'm being honest, that is what I'm hoping Beamdog might help to make happen. Even if I wouldn't really put 1942.2 in that list. My hope was that just like 1942 second Edition is meant to be the entry level game for new players, perhaps it would could be the same thing here for the game developers. Where Beamdog learns how to program a more bare bones stripped down version of A&A at first, but then adds on to that, to create a series of games using the same platform. Given where we are right now, its probably better that they didn't take on a game that had technologies, national objectives, or china rules right away. But I'm still hoping that's where it goes eventually. Or that we at least get a toolset where enthusiastic modders can recreate those games using the basic system that Beamdog sets up.
I'd honestly pay a lot more than 20 bucks for that. The reissue of AA50 still retails for like 80 dollars, and before it was reprinted they were going on ebay for like a grand. People who like these games will pay a pretty penny for this stuff, but only if its done up right. We are still in the trial run here. I don't want to see another slow burn like GTO (game table online) which promised much but didn't really deliver. So I'm a little jaded. I put a lot of heart and a lot of time into TripleA, which I still think is the benchmark, but I know it will never achieve the playerbase that a Steam title with official support can help to create and sustain. I'm hoping they're in it for the long haul.
Plus I'm just a massive Baldur's Gate geek, so I kind of dig this intersection of two of my favorite games being revisited by the same company. I know people have gripes, but I really enjoyed Siege of Dragonspear and the EE games and the fact that they have multiplayer support again. I haven't seen the work on NWN (which was also big for me) but at least this team seems to like the kind of games I like. And I appreciate that they seem to be making a place for cool PC games to continue on in the after life. I wouldn't mind them taking a similar approach to Axis and Allies, if it means a way to play with a modern OS. I feel like there is maybe an analogy to be made between Advanced 2nd Edition D&D, and Axis and Allies on the computer. I'm pretty sure Iron Blitz and Tales of the Sword Coast dropped in the same summer. Or at least, I remember having them both in the big CD case lol. So yeah, feels kind of natural. Hoping it all gets tied up with a neat bow, so I don't have to go everywhere on the internet to do my thing. I'll just set up shop here, would be nice.
Everyone takes losses in nearly identical fashion. It is extremely rare to take losses in any other way. This is not a failure or detriment
I dont want to play multiplayer, just vs the PC.
Axis and Allies, or Axis and Allies Iron Blitz (the old Hasbro CD and its expansion 1998-2004): This computer game featured the Classic version of the A&A boardgame released in 1984, and its various editions.
Axis and Allies GTO (Game Table Online 2004-2014): This computer game featured a couple versions of the A&A boardgame including Revised released in 2004, and Spring 1942 released in 2009.
TripleA (an unofficial open source game maintained by enthusiasts, which includes Mods of A&A maps 2004-Current): This computer game has pretty much all versions of A&A with the exception of 1914 and Zombies listed under the World War II heading.
In addition to these there was also ABattleMap (2004-2015), which was more of a digital mapping or board tracker utility, but which could be used with dice rollers for PBEM (play by email.)
Each of these digital versions of A&A has some unique features and things to recommend them, as well as a few drawbacks. I'm curious which if any of these platforms the Devs have looked into? I had assumed that someone on the development team would be tasked with reverse engineering some of these earlier games to see which features have been popular among digital A&A players over the years, or using them as like roadmaps for the kind of core functionality we're after. Of the games listed above, TripleA is the only one that still has an active user base, primarily using the PBF (play by forums) method, though the platform does support PBEM and Live Play via a lobby as well. It has the most competent AI of any Axis and Allies computer game that I know of courtesy of Redrum's HardAI. I would suggest that the devs look into it for ideas regarding GUI and in-game Utilities, as well as the AI stuff. I don't think we'll have Deep Blue for A&A for a while, because the base game is so damn complicated haha, but it made a pretty good start. For general gameplay principles at the tactical level it's aces, but at the broader strategic level there are still things that the HardAI in tripleA doesn't take into account, the main one being Victory Cities. Anyhow I have a lot of experience playtesting that one, and some ideas on how to take it to the next level, if A&A Online's computer AI ever gets to that point.
"Complicated" is just an aggregate of individually simple things with a prioritization list.
As to machine learning AI, you take a big block of data, feed it in, then look for correlations between data and wins. There's a few refinements in there but it's a different approach that doesn't need nearly as much analytical power built into the AI. (edit - the AI ends up being a list of if-then instructions created from the program that crunches the data. The data crunching program needs plenty of power; the end-product AI less so.)
Say Russia's first turn. Look at all possible moves and all enemy countermoves, calculate estimated utility of various attacks and withdrawal contingencies, estimate optimal forces for optimal gain, account for opponent counters including "can opener" sequential attacks and their utilities, reserve units as required to minimize impact of enemy attacks, distribute attack forces optimally, carry out attacks, attack or retreat as appropriate over the course of an attack depending on dice results, at noncombat calculate odds on opponent attacks at the new board state and move units appropriately to minimize impact of enemy attacks including to-be-mobilized units in account, mobilize units, end. I mean yeah bunch of other things in there but basically.
OR computer reads Russia players that buy 8 infantry aggregate have a 54% win rate, 4 infantry 3 artillery a 48% win rate, 3 infantry 3 tanks a 49% win rate, battleship and artillery a 0% win rate blah blah you get the picture. Then computer looks at what combat moves players with highest win rates performed . . . it's just monkey see monkey do, attack, retreat, whatever, all based on recorded game metrics, player win rates, and so on.
I mean yeah, bunch of other things in there but . . . basically.
==
My thought is what's holding AI back isn't "complexity" in the sense that it's often used as in "there's too many things to consider to make sense of it", but in the sense that programmers want to cut computation speeds. But I figure maybe instead of brute force binomial distributions performed each time, an AI could use multivariate calculus for optimization and look up values to save time.
Though for machine learning the AI really comes down to being a glorified reference table so I don't know that there's that much computation time required in terms of instructing the computer to take action based on game metrics - there's plenty of computation time required for the data but they're two different things.
I must say, I have no idea right now why Solo players vs the AI are forced to play only to standard victory. To me probably the whole point of playing a computer that doesn't quit would be to see all the weird endgame type scenarios that you don't typically get to experience in PvP play. Like what happens after the capitals fall or start being traded, how liberation works after the capitals start changing hands, or just how the production fronts re-align after the center collapses. I think if anything we should be encouraging players to play Total Victory vs the AI rather than Standard Victory, because it will probably end up being more rewarding for them as an play experience, and at least provide something different than they are likely to get vs another human. That'd be my angle anyway. Because you could set up AI challenges that really getting people enthusiastic about playing vs the AI (even a middling AI) as a way to help improving it over time, if they get to do badass steamroll type world conquest in the process.
I think it needs to be passing fair at launch to attract that kind of attention initially, so you could get it to a point where the data its collecting is actually useful for improving the game vs experienced players. So that's the only reason I mentioned. HardAI in tripleA came a long way over the past few years, going from pretty terrible to passing fair. I'd say right now Axis and Allies Online is still in the pretty terrible phase. I'm mopping the floor with it currently. In tripleA it takes about a 10-20% income bonus per round for the computer to overcome its deficit. Axis and Allies online feels like you'd need much more, maybe even like 50% to make it competitive right now, but of course those sort of income modifiers aren't possible here. So it's just a blowout.
When beating up on the computer that hard, its like you might as well be trying do crazy stuff like Invasion USA, or conquering Japan with Russia or wild endgame stuff like that, under total victory conditions. At least until it gets to the point where the Standard VC game has us breaking a sweat hehe.
I actually enjoy playing against the AI and picking it apart. I'd like to help the devs see places where the machine is falling short and work to improve it just by playing against it a ton and pointing out where it makes mistakes. Gloating over the machine is easy, cause it doesn't take things personally. But to do that you need to able to provide the developers with a game save. So they can see the state of the game when things went awry, what the machine failed to prioritize or prioritized in error, and what it should have done differently to have a better result. But right now I wouldn't know how to do that, other than abstract anecdotes or screenshots or something, which isn't really going to do the trick. The first step before anything really, and for general debugging of the game too, is just to give players a way to export saves, so we can see/show/share what's happening with each other and with the devs. To me honestly, that's the whole point and the whole promise of having a digital version of this game to begin with, beyond just a faster way to set the board, it's so that the moves can recorded and reviewed and learned from. That has been an important part of each digital iteration and it's part of what makes playing A&A on the computer cool. Because you don't have to go by anecdotal testimony, you can say 'just show me the gamesaves' and prove it.
A ground floor of general principles is not needed. The game rules do not need to be programmed in. The computer AI will not break down because of dice roll deviations and player purchases and moves over the course of 10 or even 50 or any number of rounds.
Those are not issues because machine learning AIs do not ask "why?".
A traditional AI looks at the board state, ideally computes every combination of possibilities of every unit moving to every different territory, the consequences of those moves, the consequences of combats, probabilities, and so forth. When a decision is made, it comes down to something like "If Berlin is lost, 45 IPC go to US, Germany loses 45 IPC, Germany can't produce units on its turn, compute expected gain versus possible loss and likelihood of successful Allied multi-attack on Berlin. However if the Axis can secure 9 victory cities uncontested by the end of the turn, override IPC considerations." There is always a "why" with traditional AIs. It may not be the right "why", or arriving at "why" may take too long, but there is a "why".
A machine learning AI looks at the board state, looks at what the highest rated players did in those board states, and takes whatever actions, including retreats, those players did. It doesn't think about the complexities, it only looks for highest win percentage.
The machine learning AI doesn't even require exact board positions to repeat. It can be programmed to understand analogous positions. Which is not the easiest thing to do, but hey.
The question comes up, what happens when you have situations that are totally without precedent? Well that's why a large bank of game data is needed. It will come up sooner or later.
Another question might come up, what happens if you have a completely different map? What if new units are introduced? Technology? National advantages? Well fundamental differences in gameplay results in fundamentally different generated data so the optimization algorithm needs to be run again to generate the machine learning based AI. But that's as it goes.
I'd like to get my hands on the algorithms and computation run times of different assessment methods. And a pizza.
Well I can get a pizza at least.
What I'm really interested in knowing is whether the ban on using allied transports and carriers (a huge game change), defensive hit allocation and defensive profiles, and floating fighters after defending carriers destroyed, is related to the AI. Because I would just as soon not have an AI, and have a game that implements the actual tabletop rules.
I'd be totally fine with it if using the AI required adding back the whole allied transports / carriers, defensive hit allocation, and floating fighters thing. But at least allow human players to play by the actual rules.