Selmer Bringsjord

"This state of affairs wont change unless a number of key challenges are conquered; and conquering them will require some seminal advances in the intersection of artificial intelligence (AI) and narrative. (E.g., since interactive digital narrative will need to be crafted and massaged as the story is unfolding, computers, not slow-by-comparison humans, will need to be enlisted as at least decent dramatists -- but getting a computer to be a dramatist requires remarkable AI.) In this paper, I discuss one of these challenges for the start of the new millennium: the problem of building dramatically compelling virtual characters. Within this challenge I focus upon one property such characters presumably must have: viz., autonomy."

"Let make explicit one presupposition before we begin in earnest. I assume that AI in general, along with the AI part of the gaming industry in particular, have realized that non-logicist AI isnt magic. What do I mean? I mean that people should be smart enough now to concede that no system is going to automatically learn, through subsymbolic, numerical processing, how to, say (and this is the challenge Im going to focus on below), assemble robust, autonomous characters in a game. Automated learning through artificial neural networks and genetic algorithms are fine for certain applications, but gone, I assume, are the days of wild optimism about the ability of such learning techniques to automatically yield (to stick with this example) full-blooded NPCs.4 We really only have two overall approaches to AI: one based on logic, and one based on subsymbolic processing. This paper is written from the perspective of logic. If you think that you have a way of solving the problems Im talking about without using logic, I wish you well."

"The first reason is that lots of people are either ignorant of or tendentiously choose to ignore the underlying mathematical facts. These facts include that some problems can be solved by computing machines, and others cant, and that most cant. So whenever you ask whether a problem P can be solved by a computing machine, where P is such that it isnt known that there is an algorithm for solving it, honesty should imply wait-and-see agnosticism.3"

"We have dramatically compelling non-interactive digital (= electronic) entertainment: sit down and watch The Matrix (see Figure 1) or T2 (see Figure 2). We have compelling interactive digital entertainment; my favorites include console-based sports games.2 We have dramatically compelling interactive entertainment; for example, improvizational theatre. What we dont have is dramatically compelling interactive digital entertainment. People sometimes tell me that there is a counter-example to my negative view to be found in online multi-player games, but though such games are digital, interactive, and entertaining -- they have zero literary power (which explains why, though T2 engages young kids through at least middle-aged professors, such games are demographically one-dimensional). The same can be said, by my lights, for all other electronic genres. Can we build systems that imply and affirmative answer to the title of this paper? My answer is: ``Maybe, maybe not; but at any rate I can tell you, at least in broad strokes, what some of the hurdles are, from the standpoint of AI. And I can tell you, today, about one specific hurdle."