Computer-generated characters are becoming increasingly lifellike. Yet even highly human-like agents differ from humans in several significant ways: First, ontologically (being human vs. being computer-generated); second, looking human (looking exactly like a person vs. looking very much - but not exactly - like a person); and third, displaying appropriate human-like behaviors (behaving socially human vs. acting like a computer).

One open question is how one should combine human-like and computer-like attributes when designing lifelike agents, given the fact that it is currently impossible to create characters that are completely human-like on every dimension.

Two competing theories predict the effects of combining human and non-human qualities in a single interface agent: Consistency theory states that the consistency of perceived traits leads to higher levels of positive affect. Thus, consistency theory would predict that "purely" human-like or computer-like agents would elicit greater positive affect than lifelike agents displaying mixed traits.

Alternatively, the CASA (computers are social actors) paradigm argues that people perceive mediated cues the same way they perceive cues from real life. This paradigm would predict a simple preference for higher quality agents, regardless of whether they are perceived to be human or computer-generated.

This experiment series examined the effects of different combinations of human-like and computer-like traits in embodied agents on users' perceptions of the agent, memory of content presented by the agent, and buying behavior in an e-commerce context.