I refer to this kind of gameplay as “Design by Telepathy.” The designer creates an obscure “puzzle” with exactly one solution. Inevitably, the level designer was smoking massive quantities of crack when he decided on that particular solution and declared it to be “intuitive.” You must discover by means of telepathy (or, if you can’t afford the phone call to Miss Cleo, trial-and-error) that the right way to solve the Korothian Cow Ninja puzzle is to tie the leprechaun to the tin-foil cowskull and microwave it in the Plasmotic Defibrillator to open the pod bay doors and thereby sneak past the Angry Styrofoam Space Dragon.
Our studio prides itself on gameplay based on open-ended simulation. We don’t want to dictate exactly how you play the game. We present the player with a set of obstacles to overcome and give you a whole bunch of different tools that you can use to get past those obstacles. Some players will do it in one of the standard ways we anticipated, but sometimes they’ll come up with some clever new solution we never could have dreamed up ourselves.
This kind of simulation-based gameplay requires the AI to be a lot more flexible. For the most part, we don’t know exactly how the player is going to proceed through the level or the specific order in which he or she will overcome the obstacles we put in the game. So you need to develop an AI which has some degree of “generic intelligence.” The AI needs to understand enough about the game world it inhabits that you don’t need to use a lot of scripting to make it look intelligent.
A level designer should be able to just drop an AI into a level somewhere and that AI should automatically behave in a reasonably intelligent fashion with no additional scripting required. If it’s a guard, it should act like a guard. If it’s a burrick, it should know what a burrick does and act like a burrick.
Thanks,