Identifying human-like qualities for non-player characters

Abstract

This research investigates qualities common to non-player characters in games can contribute towards an in-game agents believability. A number of NPCs with different initial setups were evaluated using a blind Turing Test style evaluation, in which players were pitted against another human and different AI controlled NPCs over six rounds. At the end of each round, players were asked to rate both opponents, unknowing which was human or NPC. The results demonstrated that none of the agents were able to reliably fool the player, however, some combinations of characteristics were successful in generating behaviour that was perceived as ‘human-like’.