Meta Cognition on the Internet: Expected Meta-Accuracy for Human and AI Virtual Assistants' First Impressions about Us(1,2)

On the Internet, we encounter different types of interaction partners some human, some AI. How we believe they perceive us upon our very first encounter sets the ground and may determine the outcome of the subsequently unfolding interaction. In a between-subject online experimental design, we studied the expected accuracy with which imagined human and AI Internet virtual assistants form first impressions. We measured expected meta-accuracy as the absolute difference between participants' self- and meta- (perceived interaction partner's viewpoint) perception along the two central dimensions of social cognition (warmth and competence). We anticipated that 1) expected meta-accuracy would be higher for human than for AI virtual assistants and 2) expected meta-accuracy would increase with increasing perceived human-mind-like abilities of the virtual assistant. We found proof for the second line of predictions and uncovered evidence for primacy of the warmth dimension in Internet meta cognition. Our results are in line with and extend previous knowledge about mind perception to the field of Internet meta cognition. Our work is pioneering in the study of meta cognition of first impressions on the Internet and promises an interesting route for uncovering previously unexplored aspects of online communication. With the ultimate goal of applying theoretical knowledge to the practical facilitation of Internet communication, we present ideas for further exploration of the area of online meta cognition. Our suggestions include different operationalizations of the primary phenomenon of interest, varying embodiment levels, and strengthening experimental manipulation to probe for the Uncanny Valley phenomenon in Internet meta cognition.