Introduction

With the increasing commotion about the importance of Artificial Intelligence in games, there is little doubt that it will be a key component of best-sellers in the near future. However, the manner in which this innovative Artificial Intelligence will be incorporated into current game frameworks is not so clear. Indeed, as you may expect, the wide variety of AI solutions available implies many possible approaches, ranging from game logic that learns to intelligent environments, including autonomous intelligent agents... All these techniques are justifiable, and in fact they often accompany each other. This essay looks into one of them: intelligent agents. We focus on essential characters in a storyline, and to a lesser extent background NPC.

Image 1: A creature in Black & White. State of the art intelligent agents?

First, we'll start by looking into current in-game agents and show how they can, to some extent, be considered intelligent. However, since they are extremely inflexible and tedious to develop, we'll show what is to be gained by expanding the capabilities of these characters to human-level intelligence. Learning will be emphasized as one of the crucial components. We'll then briefly look into AI techniques that allow such complex behaviours. Finally, from a game developer point of view, we'll look into the benefits and problems with all this new technology.

Intelligence and Black Boxes

In the games industry, programmers pride themselves to be very practical people. And indeed, when you've got strict requirements and rigid deadlines, that's an advantage. From that point of view, it's often not important how things work under the hood, as long as the outcome is acceptable. Intelligence can be defined in this very same way...

As a matter of fact, this approach to Artificial Intelligence is nothing new. One of the great minds of the last century, Alan Turing, devised a test for intelligence based on speech. The idea is to place the candidate behind a curtain, and ask it questions. If it can convince the other person that it is human, then it is considered intelligent. In practice, the evaluation of the candidate is sometimes made less subjective by introducing a set of impartial judges. Additionally, this test has been expanded upon for more than just speech... (chess and aerobics spring to mind :)

Now seems a good time to mention the implications of this policy. For AI developers, this means that hard-coded scripts are fine, if they meet your requirements. There's no reason to implement a state of the art algorithm to perform a task that would take 1 line of specific ANSI C code. The same applies to academia, though it has the reputation of being snobbish with regards to the design of solutions. This is not intrinsically the case, but the use of elaborate algorithms is usually a necessary consequence of tougher specifications and requirements. The games industry can accommodate these, since the deadline is often more important, as opposed to academia which prefers getting part of the way to a well identified goal rather than changing it...

As mentioned in the next section, the more realistic our agents become, the more they'll need to rely on academic approaches

Remember you can visit the Message Store to discuss this essay. Comments are always welcome!. There are already replies in the thread, why not join in?