AI and non-human agents in virtual worlds
Many non-human agents – such as chatbots, service robots, and AI-based car driving assistance systems – are designed for use in real life. Non-human agents, however, exist in various virtual worlds, too – as in computer games, in which they are typically called non-player characters (NPCs). Such agents also exist in other virtual worlds created for entertainment, for example, in virtual clubbing and virtual porn. And they appear in worlds created for education and training purposes. One specific application is the virtual city of Uruk 3000 B.C., in which students of history can be immersed as avatars and thereby hopefully become more engaged compared to when they read about Uruk in a textbook. And one way to accomplish this is to populate the virtual city with interactive non-human agents, such as Uruk fishermen and pot makers, who can take on the role of tutors. In another application, a virtual audience was created to allow users to practice presentation skills; the members of the audience could be manipulated to look bored, yawn, and ask questions.
Yet, to date, many of the non-human agents in virtual worlds have not been particularly realistic. And they have been less interactive than the non-human agents that users meet IRL. In computer games, in which most NPCs are based on handcrafted behavioral scripts, one reason is that AI-powered autonomous virtual agents may engage in non-predictable behavior, which can attenuate the gaming experience. Truly intelligent NPCs, fully capable of learning, are of course also less optimal from a cost effectiveness point of view.
This, however, is likely to change: many observers argue that it is time to better capitalize on AI when it comes to its possibilities to improve the behavior of non-human agents in virtual worlds. It would reduce the gap in relation to the increasingly realistic physical parts of the environment in many virtual worlds (such as cars, airplanes, weapons, and buildings) and the gap in relation to avatars, which represent and are controlled by real humans, so it has the potential to offer a more congruent user experience. It would indeed allow for more sophisticated interactions between the user and virtual agents. In a virtual clubbing setting, for example, the user’s experience is likely to be boosted (or at least likely to be different) when a virtual dance partner behaves like a human on the dancefloor and can engage in humanlike conversation. This, then, would blur the line between social encounters IRL and in virtual worlds.
AI-powered virtual agents in virtual worlds would also blur the line in development processes for agents to be used in virtual worlds and IRL – and blur the line between AI and virtual world technologies such as Virtual Reality. And more realistic virtual agents in virtual worlds open the door for researchers who want to study humans’ social behavior in simulated (but realistic) settings rather than IRL. Maybe it is also in a virtual world that we should train our AI-based agents – such as service robots helping us in our homes – before they are let loose for IRL usage.