Because nobody really knows what to do about it.
It is very common to hear or read complaints about the AI in 4X gaming communities. It is, in fact, almost ubiquitous; go and take a look on almost any 4X gaming message board and you will find plenty of discussion and complaints about the inability of the AI to put up any sort of real fight. I will wait…
Of course if you are reading this article you are probably already a fan of the 4X genre and a pretty experienced player and AI issues are nothing new to you. But what is interesting is that if you play a lot of 4X games, or hang out on a lot of 4X message boards, the same complaints turn up over and over again:
1) The AI is a cheater
2) The AI is too passive/does nothing
3) The AI builds dumb things in its cities
4) The AI builds cities in odd places
5) The AI is terrible at tactical combatAnd so on and so forth.Why are these complaints so common? Why do they seem to cut across the genre, rarely leaving a game unscathed?
Many of these AI weaknesses are due, in large part, to the fact that 4X games are pretty complex. Even relatively simple 4X games will have growth, production, and research mechanics that operate independently and interact with one another at the same time. Terrain and other environmental factors, often randomly generated, will also have an impact on how those mechanics play out in a given gaming session. Some sort of unrest/happiness mechanic is common. Many games also have a trade route mechanic. Sometimes, there is a command point limit for units to be accounted for as well.
Diplomacy is another complex arena for 4X games. Players expect the AI to behave in a relatively intelligent manner and to not, for example, give away all of its strategic resources for a song and a pretty-please-with-sugar-on-top from the player. Players also expect that the various AI players in a game will act and react in a somewhat predictable and realistic manner when it comes to diplomacy. No one likes an AI that proposes an alliance on turn 15 and then declares war on turn 17 for no apparent reason.
In tactical combat, the AI faces a similar set of obstacles. The interplay of unit movement and attack rules, special abilities, terrain modifiers, and global effects (i.e. spells) often proves too much for most AIs to play intelligently from a human perspective.
All of the complexity that attracts us 4X lovers to the genre is the same complexity that causes the AI such grave difficulty.
Meanwhile, it is important for 4X players to remember that computer AIs have struggled with games with far simpler rule sets, like chess. In 1997, when the IBM computer Deep Blue defeated chess champion and Grandmaster Garry Kasparov, the victory was less than decisive; the AI won 2 games, Kasparov won 1 game, and there were 3 games played to a stalemate in the 6 game match. Not only that, Deep Blue received programming updates between games from IBM personnel. Later, in 2003, Kasparov played Deep Junior, the software descendant of the hardware-based Deep Blue, and fought the computer to a 3-3 game tie. (source). Humanity’s record in chess versus the machines since then has been mixed. (source). Point being, if computers struggle with a relatively simple game like chess, how can we humans reasonably expect them to do any better in more complex 4X games?
In my opinion, the answer is simply that we should not expect near-human level AI in 4X games and, further, that the AI in a 4X game doesn’t have to be anywhere near that competent in order to be fun and challenging to the player.
Perhaps when we ask, “Why is the AI in game X or game Y so stupid?” we are asking the wrong question. The venerable philosopher Obi-Wan Kenobi once said that “…many of the truths we cling to depend greatly on our own point of view.” We should shift our point of view and stop judging the AI by human player standards. Instead, the truth that we should recognize is that a 4X game is a strategy game for the player, not the AI.
So what is it that a 4X AI needs to be able to do? Just how good does it have to be in order to be sufficient? What allowances should game developers make for the AI’s obvious weaknesses against human opponents?
The primary thing that any good 4X AI should be able to do is to be an obstacle for the player. If the player can march through the entire game while completely ignoring the AI, then what is the point of having AI players at all? What we should be looking for is an AI competent enough to get in the player’s way. In other words, the AI should pose a significant obstacle to the player’s strategies and designs for the game. To be satisfying, all the AI must do is to force the player to strategize around its presence and actions. If the AI can pose a significant strategic obstacle for most players to overcome (at most difficulty levels), then near-Kasparov levels of intelligence are unnecessary.
Meanwhile, the AI has to present a decent military threat to the player as well. Tactical combat systems in 4X games are often complex and units have a wide variety of strengths, weaknesses, and special abilities for the AI to keep track of and use. But remember, the 4X game is a strategy game for the human player, not the AI; tactical genius is not required for a satisfying experience. What a good AI should do is pose a decent military obstacle for the player to strategize his or her way around. Looked at in this way, sheer numbers of military units can be enough to pose a significant strategic challenge for the player. An AI that is programmed to make semi-effective use of units’ special abilities can also force the human player to change tactics in combat, build up unit resistances, or build units that would otherwise remain unused.
Of course more could be said about other common 4X game mechanics with regard to the AI. The point, however, is that if we continue to look for 4X games that exhibit a level of intelligence and play similar to that of a human player we are doomed to disappointment for the foreseeable future. Instead, it is important to remember that a 4X game is meant to be a strategy game for its human players and not for the AI. The question we should be asking is not whether the AI makes decisions the way that we human players would make those decisions, but whether or not the AI is sufficiently intelligent to pose a meaningful strategic obstacle for the human player to overcome. What we should be looking for is an AI that brings such a challenge to the strategic table as well as some color and life to the world in which the game takes place.
Maybe if we look at AI in that light then that pesky metaphorical elephant won’t take up so much of the room. But we might still need a new coffee table.