|
In A Force More Powerful, the behavior of the adversary, the Regime, is controlled by the "artificial intelligence" (AI) built into the software. Creating AI for A Force More Powerful posed a unique challenge.
In most strategy games the goal for the AI is that it should play as well as possible within the bounds of a difficulty level set by the user. At the most difficult levels the AI should play optimally, never wasting a day or a dollar and always taking the choices that offer it the best chances of success, mathematically compared to every other option. But real-life governments rarely do this, and dysfunctional or corrupt governments are least likely to do so. Thus, we needed to create an AI that didn't act optimally, yet clearly reflected the internal dynamics within the adversary. One of the most important points AFMP tries to make with its models is that human institutions aren't monolithic (even in an authoritarian system); instead, they're made up of groups and personalities that compete with one another as much as they cooperate, and promote their own agendas just as much as they promote the whole coalition's agenda.
Hence the AI needed to dramatically change its behavior based on the Regime leader's individual quirks as well as its overall character as determined by a weighted average of all the different attributes of all the people and groups in the Regime. This variation in behavior extended to what Tactics the Regime prefers, to how far it needs to be pushed before it will engage in repression or start granting concessions to the movement, to how efficient it is at managing its resources.
Another nuance in the way AI in AFMP works is how it reacts to political pressure. Some Regimes react brutally to political pressure - this is determined not just by the Regime's characteristics but also by the Status Quo, since the Status Quo determines what measures the Regime is allowed to take. (Of course, the Regime also has the power to change the Status Quo to suit itself!) Others will be more likely to react to pressure by giving the movement some of what it wants.
This highlights another important difference between the AI in this game and in other strategy games: in the latter, "winning" for the AI is the same thing as "losing" for the player. In empire building or conquest games, the AI's gain is always the player's loss. In AFMP, though, the adversary's goal is to stay in power or maintain certain elements of the status quo - goals which are not necessarily 100% at odds with the player's goals. It was not the goal of Dr. King to overthrow the U.S. government, for example. Therefore, we had to take care to make sure the AI in AFMP stayed focused on what its real-world goal would be, and didn't just go after the player for the sake of making the player lose. Hence the AI in AFMP will often, when pushed to the breaking point, do just as real governments do: try to give concessions so that the pressure will be relieved.
Moreover, we also had to make sure that the AI didn't go after the player unrealistically. In other games, the AI will take advantage of a player starting in a weak position and pounce immediately. Real-world adversaries don't do that - if the opposition is under the radar, they'll tend to ignore it until it makes itself visible. We accomplished this by creating a concept called "Focus on the Movement", which measures the adversary's nervousness about the movement. As Focus rises, more Tactics become available for the adversary, but concessions are also more likely. The Difficulty Level selected at the beginning of each scenario has an effect on how fast Focus can rise and fall, as does the Regime leader's personality type (a Bureaucrat or Kleptocrat will be slower to react to a threat, but also slower to re-adjust once the threat has been dealt with; an Ideologue or Pragmatist will be faster in both cases).
Finally, we had to balance the AI over months of testing. This was separate from the task of balancing the Tactics - we also had to make sure that the AI knew when to use each Tactic and didn't overuse any in particular. We think we did a good job at designing in enough variation for the AI's behavior while keeping it realistic and challenging.
The AI was one of the most important parts of the design and the game, and we put a high priority on getting it right. We hope it is up to the challenge of cracking down on thousands of virtual movements on people's computers around the world!
|
