A List of Some Challenges

Very well. So, why is it that building dramatically compelling interactive digital entertainment (in the form, e.g., of games) is so difficult? There are many reasons, among which fall the following.

C1: Formalizing Literary Themes.

If Dave Ferrucci and I are right, plotlines and so-called 3-dimensional characters aren't enough for a computer to generate first-rate narrative: you also need to instantiate immemorial themes -- betrayal (the one we focus on in (Bringsjord & Ferrucci, 2000)), self-deception (a theme we employ in BRUTUS), unrequited love, revenge, and so on. If such themes are to be used by story-managing machines, they must be represented; if they are to be represented and exploited, they need to be rigorously represented and reasoned over. Such rigorous representation and reasoning is very hard to come by.

C2: Story Mastery.

After interactive drama begins, things can devolve toward the uninteresting. If ``hack-and-slash" is all that is sought from an interactive game, then such devolution may be acceptable. But if genuine drama is desired, then something or someone must ensure that what happens is dramatically interesting. One possibility (with respect to a multi-player online game) is to have a human or humans oversee the action as it unfolds, and make changes that keep things ``on track." For obvious reasons, in a rapidly progressing game with thousands of human players, this is impracticable; the possibility of human oversight is purely conceptual. So we must turn to computers to automate the process. But how? How is the automation to work? I've assumed that if a program could be built that writes compelling fiction, we might thereby have taken significant steps toward a program that can serve as a story master.

C3: Building Robust, Autonomous Characters.

A sine qua non for compelling literature and drama is the presence of robust, autonomous, and doxastically sophisticated⁵ characters. In short, such literature and drama exploits the central properties of being a person. (In many cases, great stories come to be remembered in terms of great characters.) This presents a problem for interactive electronic entertainment: how do we build an electronic character that has those attributes that are central to personhood, and whose interaction with those humans who enter the virtual worlds is thereby compelling?

C4: Personalization.

If virtual characters are going to react intelligently to you as user or gamer, they must, in some sense, understand you. This problem is directly related to C3, because the characters must have sophisticated beliefs about you and your beliefs, etc.

In the remainder of this paper, I focus on C3, and moreover I focus within this challenge on the specific problem of building autonomous characters. The plan is as follows. I begin by reviewing the concept of an intelligent agent in AI. I then explain the clash between this limited concept and the kind of properties that are distinctive of personhood; one of these properties is autonomy, or ``free will." In order to highlight the problem of imparting autonomy to a virtual character, I turn to what I have dubbed ``The Lovelace Test." I conclude with a disturbing argument that seems to show that virtual characters, as intelligent agents, can't be autonomous, because they would inevitably fail this test. I do intimate my own reaction to this argument.