Academic Background

In human social interactions, gaze is one of the most important ways to convey information. Research has shown that human gaze can influence others’ attention, cognition, and even preferences. For example, when a person looks at an object, the observer tends to believe that the object is attractive to the gazer, which in turn influences the observer’s own preference formation. However, with the rapid development of robotics, robots are gradually acquiring gaze behaviors similar to humans. So, can a robot’s gaze influence others’ preferences like a human’s gaze? This question not only involves how humans perceive robotic behavior but also relates to the design and optimization of future human-robot interaction (HRI).

Moreover, Theory of Mind (ToM), the core ability to understand others’ mental states, including inferring their intentions, beliefs, and emotions, plays a crucial role in human social interactions. However, its role in the effects of robotic gaze has not been fully studied. Therefore, this study aims to explore the differences between human and robot gaze in preference formation and analyze the role of ToM abilities in this process.

Source of the Paper

This research was jointly conducted by scholars Federico Manzi, Mitsuhiko Ishikawa, Cinzia Di Dio, et al., from institutions such as Università Cattolica del Sacro Cuore in Italy and Hitotsubashi University in Japan. The paper was published in IEEE Transactions on Affective Computing, an important journal in the field of affective computing that focuses on interdisciplinary research between emotion and computational technology. The publication date of the paper is 2025, with the DOI: 10.1109/TAFFC.2025.3531945.

Research Process

1. Participant Recruitment and Experimental Design

The study recruited 79 Italian adult participants, with a gender ratio of 40:39, and an average age of 23.83 years. All participants were native Italian speakers and signed informed consent forms. The experiment was conducted online via the Qualtrics platform, with participants randomly assigned to different tasks.

2. Theory of Mind Ability Tests

The study employed two ToM ability tests: - Reading the Mind in the Eyes (RME): Participants were required to infer the emotional or mental state of characters based on 36 eye-region images. Each image had four options, and participants selected the most appropriate one. Scores ranged from 0 to 36, with higher scores indicating stronger ability to interpret others’ emotional states. - Perspective Taking (PT): Participants needed to infer the size or shape of objects from another person’s perspective. The task included five trials, each showing a bookshelf, and participants had to choose the target object from the perspective of a person behind the bookshelf. Scores ranged from 0 to 5, with higher scores indicating stronger perspective-taking ability.

3. Gaze Preference Task

The experiment used a 2×2 repeated-measures design, with two factors: “Agent” (human vs. robot) and “Question Type” (gazer preference vs. participant preference). Participants watched 10 videos, each lasting 10 seconds, where either a human or a robot (Robovie) looked at one of two objects. After each video, participants were randomly asked: “Which object does the robot/girl like?” (gazer preference) or “Which object do you like?” (participant preference). Scores for each question type ranged from 0 to 5, with higher scores indicating a greater influence of gaze on preferences.

4. Data Analysis

The study used Generalized Linear Mixed Models (GLMM) to analyze data from the gaze preference task, with agent, question type, and gender as fixed effects and participant ID as a random effect. Additionally, Generalized Linear Models (GLM) were used to analyze the impact of ToM abilities on gaze effects.

Main Results

1. Gaze Preference Task

The results showed that, whether from a human or a robot, gaze had a significantly greater influence on gazer preference than on participant preference. This indicates that gaze serves as a strong social signal in conveying the gazer’s intent, but its influence on individual preferences is weaker.

2. Influence of ToM Abilities on Gaze Effects

In the robot condition, PT ability significantly positively predicted gazer preference scores (p = 0.05), while RME ability significantly negatively predicted participant preference scores (p = 0.032). This means that individuals with stronger perspective-taking abilities were better at inferring the robot’s preferences, while those with stronger emotion-reading abilities were less influenced by the robot’s gaze. In the human condition, ToM abilities had no significant effect on gaze effects.

Conclusions and Implications

1. Main Conclusions

• Adults process robot gaze similarly to human gaze, interpreting it as a signal of psychological states (e.g., preference for an object).
• Gaze alone is insufficient to significantly influence individual preferences, regardless of whether it comes from a human or a robot.
• In the robot condition, different dimensions of ToM abilities (perspective-taking and emotion reading) predict gazer preference and participant preference, respectively.

2. Scientific Value

This study is the first to reveal the role of ToM abilities in robot gaze effects, providing new perspectives for HRI research. It shows that humans can interpret a robot’s gaze as a signal of psychological states, but its influence on individual preferences is weak, possibly due to the cognitive distance associated with robots as non-human entities.

3. Practical Applications

The findings have important implications for robot design. For instance, in sensitive scenarios such as healthcare or education, robots could more effectively convey intentions by combining gaze with emotional signals (e.g., smiling or frowning), thereby influencing human behavior and preferences.

Research Highlights

1. Novel Perspective: This is the first study to explore the role of ToM abilities in robot gaze effects, filling a gap in HRI research.
2. Interdisciplinary Approach: By combining psychology and robotics, the study reveals the psychological mechanisms humans use to process robot gaze.
3. Practical Application Potential: The findings provide specific recommendations for robot design, helping improve robots’ performance in social contexts.

Other Valuable Information

The study also notes that cultural factors may influence how humans interpret robot gaze. For example, East Asian cultures, with higher acceptance of robotic technology, may be more likely to interpret robot gaze as a signal of psychological states. Future research could further explore the impact of cultural differences on robot gaze effects.