View source for Abubshait, A. & Wiese, E. (2017). You Look Human, But Act Like a Machine: Agent Appearance and Behavior Modulate Different Aspects of Human–Robot Interaction.

In their paper, A. Abubshait & E. Wiese investigate what is necessary for humans to perceive a “mind” in an artificial agent, and how an agent can be developed to be as socially capable as possible. One of the most important and frequent parts of human communication is eye movement/gazing. The gazing behavior of an individual is incredibly important in social settings, since it can signal different things to other humans. Because of the sheer importance of gazing, it will be used as a way to investigate the interaction between agents and humans.


An experiment was devised that aimed to pin down how agent behavior and appearance trigger mind perception, and how it influenced social-cognitive performance like gaze following. Participants were shown a face on a screen, either very human like, or very robot like. This face would look left or right, and then a letter would appear either left or right. How often the face looks in the same direction as where the letter appears is called reliability. An agent that looks in the same direction as where the letter is to appear often, is deemed reliable. To test for reliability, two different groups were created: one with 80% reliability, and one with 50% reliability. 


Two main things were set to be decided from this experiment: how does appearance (human-like vs. robot-like) influence mind perception (1), and how does reliability influence mind perception. After having performed the experiment, the researchers found that reliability, but not appearance, influenced performance. This means that gaze following was better when the agent was more reliable. On the other hand, appearance had a significant effect on mind rating, while reliability had only a small effect.