Simulating Human Imprecision in Temporal Statements of Intelligent Virtual Agents

Abstract

Research on intelligent virtual agents (IVAs) often concerns the implementation of human-like behavior by integrating artificial intelligence algorithms. Thus far, few studies focused on mimicry of cognitive imperfections inherent to humans in IVAs. Neglecting to implement such imperfect behavior in IVAs might result in less believable or engaging human-agent interactions. In this paper, we simulate human imprecision in conversational IVAs' temporal statements. We conducted a survey to identify temporal statement patterns, transferred them to a conversational IVA, and conducted a user study evaluating the effects of time precision on perceived anthropomorphism and usefulness. Statistical analyses reveal significant interaction between time precision and agents' use of memory aids, indicating that (i) imprecise agents are perceived as more human-like than precise agents when responding immediately, and (ii) unnaturally high levels of temporal precision can be compensated for by memory aid use. Further findings underscore the value of continued inquiry into cultural variations.