Steve weighs in deeply here. I suppose he knows much about chess. I suppose I would wager a nice sum he could not beat me. But he is full of opinions so enjoy.
English version
1/ For you, what’s really an AI in a wargame. Several scripts and algorythms, of course, but anything more ?
This is going to sound odd, but there is something else… « artistic creativity ». If you boil things down they usually can be categorized as scripts or algorithms, however the devil is always in the details. Especially with AI. To get complex behavior the AI not only has to have a very strong set of scripted behaviors to select from, but it also has to understand when to use and not use them. It’s the last part that’s so difficult to do in the real world. Limited time budgets, competing resource needs within the game, and the enormity of the task (i.e. duplicating Human intelligence) require a clever approach with a very skilled programmer. This is why academic AI programmers rarely venture into commercial gaming.
Cheating is a tool that some games use to give the player more of a challenge without having to do lots of complex AI programming. However, players quickly see through this and find it angering. Gamers have very low patience for cheaters. Which is why so few wargames use much, if any, cheating.
2/ For a PC wargame, what are the differents AI models ? And the differents software tools for developers ?
AIs generally fall into two basic types; on-the-fly and scripted. The first method means the program conducts it’s actions without any direct help from a Human. Most wargames favor this approach. The scripted concept means the AI takes cues from Humans who developed the battles being played out. This requires some degree of end user tools and the skills to use them well. Most non-wargames favor this approach. Then there are the hybrids which lean one way or the other, but combine elements of both.
The primary advantage of an on-the-fly system is more flexible and requires no end user effort. The disadvantage is the programmed AI is now solely responsible for figuring out how to approach a specific battle and how to deal with any number of different opponent actions. Scripted, on the other hand, has the advantage of a Human instructing the AI how to approach a specific battle based on his understanding of how that battle is likely to play out. The disadvantage is that battles take more time to create and (usually) AI scripting isn’t a skill most end users have. Which means reducing the numbers of potential battle makers simply because many don’t have the patience/skill to create AI scripts.
3/ What is possible with the computing power of a modern PC (more or less 2 years old) of good quality ? 3D asides, what other technicals constraints is there ?
There’s a very old axiom in computer technology… every new computing advance that comes out there’s already a game in the marketplace that can overload it. This means all new technologies allow games to advance in quality, but not as much as either game developers or game makers want. I’d say right now the two most important advances in the past few years are cheap RAM and higher core clock speeds. Video card technology hasn’t improved as much as it has in the past.
4/ Are some kind of wargames, so of historical conflicts, or scales of games, more appropriate for (current) AI ? If so, which one ?
Higher level games are easier to program AI for because the scale is easier to « fudge ». Tactical games are the hardest because the number of elements the AI has to successfully handle is much larger. The more « melee » a game is the easier because (again) the range of elements is far fewer than others. In fact, I was thinking about this last night as I played an RTS game that has only 4 types of units (Pikemen, Swordsmen, Archers, and Cavalry). One form of ranged fire, everything else is brute force hand to hand. Compare this to tactical modern warfare with it’s range of weaponry and unit specializations. Not even in the same ballpark from an AI standpoint.
5/ From your experience in game design, what technicals challenges have you meet ? Which one did you overcome ? Which one are still a barrier ?
Player expectations are always higher than any game can deliver, so that’s a barrier we still have. If we overcome it we’d probably be the first ever, which is why we don’t judge our games’ success based on such an unobtainable standard. What we have succeeded in doing is providing players with a very realistic game environment that produces reasonably realistic behaviors according to reasonable expectations. Where we need to do more work is giving scenario designers the ability to have their AI scripts be more adaptable based on their designs and unexpected situations.
6/ Is an AI allied with the player (for a game including alliances and several factions) more simple or more complicated to build than a unique AI enemy of the player (for a game where there’s mainly one adversary) ?
Whether a Human is directing AI or the AI is acting on it’s own, they must work from the same basic AI programming. Which means, basically, the two are identical in terms of difficulty. Remember, an allied AI that doesn’t behavior well without Human interaction is almost useless. The player might just as well be commanding all those units himself! And there is where the big difficulty comes about. In order for a Human to have a competent allied AI there needs to be a very strong capability of both Human and AI to interact with each other. What’s the use of an allied AI player if it’s getting beat up and doesn’t ask for your help? Or if it sends resources to help you and then doesn’t have enough to do anything for itself, which then requires you to send stuff back to it? How likely is the game to succeed if you can’t coordinate his actions with those of his ally’s? It requires an entirely unique type of UI and AI for this to work.
7/ In strategy games and wargames AI are very often better in attack than in defense, why ?
Defense, always. It’s the same in real life as well. The attacker is the one that has all the pressure on it to come up with something brilliant. The defender simply reacts to it. Much easier to react than to act.
8/ To what extent the perception of the AI from the player has to be taken into account ? Can an AI in a game just be less good strategically (than a human adversary), as long as it offers a good gameplay with its actions ? And so at the end, for a game, offers an entertaining result, thus something satisfying ?
AI must be at least entertaining, or it’s a waste of time. Anything beyond that is a bonus.
9/ For wargames, what development can we expect for AI in a relatively near future ? A crazy idea probably, if that was possible technically, could a wargame having part of its AI in the future (and rather hypothetical) « cloud » gain an advantage (ex : computing power, comparing moves in same other games) ?
If a wargame company could double it’s sales by making better AI then you’d see a number of really great innovations. Especially what is loosely called « learning » behaviors. They offer the best possibilities for major advancement, but they also require massive programming skills and time plus vast amounts of computer resources to make them work effectively. Since there is no financial incentive to improve AI these things will not happen. Or happen any time soon, at least.
10/ A chess game has a board with 64 squares and 32 pieces. So a base less important than a good wargame, and far much less than some « monster wargames ».
With 256 processors in parallel (allowing 200 millions move / sec.), in 1996-97 Deep Blue has finally beaten the world chess champion Gary Kasparov. In theory, well programmed, a computer do not make mistakes. But even with a super-computer, the machine still lacks a strategical vision, a global view ? If we increase the size of the « board game », for a wargaming simulation is it just a question of computing power ?
Computing power is, certainly, a major limiting factor. But the major factor is the the lack of what is called « outcome based AI ». In chess the elements have very specific capabilities and the variables boil down to which piece is moved and which direction and how many spaces does it move. There are only so many combinations of these things that can possibly happen given a specific circumstance. Deep Blue was programmed to search through all the patterns, identify potential moves, then play out those moves long term to predict where that one action might lead to many turns later. It’s only possible in Chess because the elements are so limited and predictable. Any significant simulation of warfare is vastly more complex.
The only real solution is to program a computer to think like a Human. Billions of Dollars could be given to the most advanced AI research centers in the world and still come up short of expectations.
11/ Tiebreaker question : If in a wargame an AI could beat certainly an experienced wargamer, whatever the simulated theatre of operations, to be a good game does it have to limit the AI to necessarily let the player win ?
Players don’t like to lose. They say they want really good AIs, but if they are given one they would quickly lose interest. So yes, when it comes down to it players don’t want great AI. They want AI that’s just good enough to give them a realistic challenge and then lose. At least most of the time.
Our thanks to Battlefront for taking the time to answer these questions.