作者：Rian Quenlin

Mazda MX-5是否刚赶超你的神速Polyphony F1赛车？在你获得特斯拉线圈前敌人是否就已配备军队？AI是否透过墙面直接看到你？有的象棋游戏是否允许系统进行违规的移动操作？那么你面对的是个无疑是个作弊AI。

AI的操作内容存在限制，例如需要适应随机环境，或动态环境，或是陷入无处可逃的回路中。有些游戏的AI非常杰出，例如《最终幻想战略版》，但除了独特类型的敌人，它们几乎遵循相同的规则。

screenshot from igxpro.net

但我觉得一个作弊AI就是个糟糕的游戏设计。有些琐碎的地方我可以理解，例如让他们看穿战争迷雾，因为AI会烦于追逐和预测你的操作，但有些内容就太具欺骗性了。

作弊AI的典型例子显而易见。最典型的表现在策略游戏中，其AI完全忽略资源限制（游戏邦注：这在《Red Alert: Retaliation》中表现尤其明显），能够指挥整个军队，仿佛是一大群玩家在操纵同一个阵营，完全无视创建规则（《泰伯利亚之日》就是如此）。但玩家作为人类也存在建墙或亚对策等诸多有利条件。

这是个糟糕的游戏设计，但极端的欺骗行为，即不按理出牌的AI才是邪恶的违规游戏设计。这种AI就像是橡皮筋，你越是拉伸，它们形成的抵触就越大。快速根据你的技术水平而变化是它们在理想世界中的运作情况，但有时它们的表现有些近乎疯狂。

这在赛车游戏中表现得最为明显，有些赛车以其车型完全无法达到的速度从你身旁呼啸而过。你对AI稍不留神，它们就已剥夺你在技能游戏中所具备的有利条件。

有些游戏则更为慷慨，会通过动态难度水平设置游戏进程，这样你就始终面临挑战，但不会处在完全不公的状态中。但这是否属于优秀游戏设计仍有待商榷。

这之所以被视作糟糕的游戏设计是因为在你所体验的游戏中，你所面临的规则和对方相同，而优秀的游戏通常会让你和对方处在势均力敌的地位，或是略微不利的处境。

而在有些游戏中，打破规则的作弊AI具有可行性。例如，《最终幻想战略版》就包含若干不受规则约束的玩家，但只要你的团体得到Cid，你就能够扭转时局，让自己免受Ivalice规则的限制。

作弊AI也能够获得合法地位，因为即便玩家具备非常模糊的亚策略知识，他们也能够轻松主导游戏。例如，《最终幻想X》虽然是款简单的游戏，但其包含众多只能通过魔法进行攻击的怪兽（游戏邦注：例如Flan）。若Flan过于强大，玩家无法将其置于死地，你会怎么做？将其MP降成0。遗憾的是，由于《最终幻想X》的AI是个作弊角色，玩家可从中抽取的MP有限，但他们其实并不需要使用MP施放魔法。庆幸的是，这一情况在《最终幻想X-2》中得到改善，但效果并不显著。（本文为游戏邦/gamerboom.com编译，拒绝任何不保留版权的转载，如需转载请联系：游戏邦）

Cheating AI is Bad Game Design

By Rian Quenlin

Did a Mazda MX-5 just overtake your lightning-fast Polyphony F1 car? Does the enemy have a literal army before you even get a Tesla coil up? Is the AI looking through walls at you? Is a chess game making OUTRIGHT ILLEGAL MOVES? Congratulations you are up against a cheating AI.

The trope name for this is The Computer Is A Cheating Bastard, and the length of the page on TVTropes is fairly long. There are limits to what an AI can do, such as having to adapt to a random environment, or a dynamic environment, or getting caught up in a loop where they have no means to escape. There are games with excellent AI such as Final Fantasy Tactics, but other than unique classes on the side of the enemy, they play (mostly) by the same rules as you.

I feel however that a cheating AI is simply poor game design. I can understand some minor things like having them see through fog of war since it would be a pain in the backside for the AI to have to track and predict your moves, but some things however are outright cheating.

Notable examples of a cheating AI are quite obvious. The most obvious ones can be seen in strategy games, where the AI completely disregards resource limits (extremely obvious in Red Alert: Retaliation), can command an entire army as though an entire team of players were handling one side, and laugh at build rules (Tiberian Sun does this). The player does have many advantages as a human (such as building walls and metagame knowledge) however.

That is poor game design, but foul game design is the ultimate cheatery, a rubberband AI. Named for rubberbands that resist harder the more you pull them. Adapting quickly to your skill level is how it would work in an ideal world (I believe Lego Star Wars does this), but sometimes they just pull craziness out of their backsides.

It’s most obvious in racing games when a car speeds past you when such a speed is mechanically impossible for that kind of car. Take your eyes off the AI for one second they’re right along your bumper, robbing you of any advantage you have in a skill-based game.

Some games are a bit more generous and will ease up with a dynamic difficulty that fluctuates so that you’re always challenged, but not to where it’s downright unfair. Whether this is good game design or hand-holding however is debatable.

The reason this is bad game design is because if you’re playing a game where you’re up against the same rules as the other side, then a good game would have you be evenly matched or at a slight disadvantage.

There have been some games where the cheating AI breaking the rules makes sense in-universe. Final Fantasy Tactics Advance for example has certain people be exempt from the law, but once you get Cid in your party, you can turn the tables and give yourself a degree of exemption to the laws of Ivalice.

A cheating AI can also be justified because someone with even a inkling of metagame knowledge would be able to dominate the game effortlessly. An example, Final Fantasy X, though an easy game, has a handful of monsters who can only attack via magic, like Flans. If a Flan was too powerful to kill, what would you do? Osmose his MP down to zero. Unfortunately, because the AI in FFX is a cheating bastard, there is a limit to how much MP you can suck out of them but they actually don’t need MP to cast. This was thankfully corrected in FFX-2 but honestly not effective.

Do not get me started on Final Fantasy 8′s card game when the hands are concealed. I’m sure that’s being a royal cheater as well with pulling out the perfect card like they were in an episode of Yu-Gi-Oh.

It is possible to develop a good AI, but to make a cheating one? You’re just a hack.（Source：igxpro）