Do people understand irony from computers?

In this paper, we empirically investigate whether people understand irony from computers in order to test the recent argument for an egocentric tendency in irony comprehension. In the experiment, participants took a timed math test comprising 10 questions of 3-digit by 2-digit multiplication. After that, they received a feedback comment on their performance (including potentially ironic sentences) from either an intelligent evaluation system with an AI engine (AI condition), a nonintelligent automatic evaluation system (Auto condition), or a human judge connected via the network (Human condition). The result was that the participants in the AI and Auto conditions understood the comment as ironic as those in the Human condition, and the participants in the AI condition perceived more sarcasm than other participants. Because people know that computers cannot think just as humans do, these results can be regarded as evidence for the egocentric tendency in irony comprehension, indicating that participants understood irony egocentrically from their own perspective without taking into account the mental state of the ironic speaker. These findings are also consistent with the “media equation” theory, from which we can suggest implications for the media equation, anthropomorphism, and computer-mediated communication of irony.