Introduction
In social interactions, humans naturally attribute "mind" to others, inferring their beliefs, desires, and intentions. This attribution is a cornerstone of cooperation. However, a "mind gap" often exists between humans and technology; people tend to attribute less mind to computers than to humans, leading to lower levels of trust and cooperation. While AI research has traditionally focused on Agency (rational, goal-oriented behavior), it has largely neglected Experience (the affective side of intelligence). This study investigates whether manipulating a user's perception of an agent's cognitive and emotional abilities directly changes how that user treats the agent in a financial decision-making task.
Methods
The researchers conducted a controlled experiment with 187 participants using a 2 × 2 between-participants factorial design, crossing Intentionality (Agency) (Random vs. Intentional) and Emotions (Experience) (Non-emotional vs. Emotional).
Participants played the Ultimatum Game as the "sender." They were given 20 tickets and decided how many to offer an AI "receiver." If the AI accepted, the split occurred; if rejected, both received zero.
- Agency Manipulation: "Intentional" agents were described as making decisions based on intent, while "Random" agents were described as using a die roll to decide. Visual cues reinforced these descriptions.
- Experience Manipulation: "Emotional" agents used virtual faces to express joy, sadness, or anger based on the fairness of the offer. "Non-emotional" agents remained neutral.
Results
The study found significant main effects for both factors. Participants offered significantly more to intentional agents ($M = 39.36\%$) than to random agents ($M = 29.36\%$). Similarly, emotional agents received higher offers ($M = 36.75\%$) than non-emotional ones ($M = 31.86\%$).
Subjectively, intentional agents were perceived as more fair and trustworthy. Emotional agents were rated as more fair, trustworthy, likable, and cooperative. Interestingly, only 18.6% of participants attempted the "rational" game-theoretic move of offering zero to a random agent, a figure that decreased further when agents displayed intentionality or emotion.
Discussion and Key Insights
The results confirm that Agency and Experience are distinct, additive dimensions of mind perception. There was no interaction effect, suggesting that emotional expression increases cooperation regardless of whether the agent is perceived as rational or random. This implies that designers have independent levers to build trust.
Furthermore, the study highlights that perception is reality in human-agent interaction. The agents did not need to be truly sentient; simple cues like an hourglass (for intentionality) or a sad face (for experience) were sufficient to shift human behavior toward higher cooperation, moving closer to the "fair split" (50%) typically seen in human-human interactions.
Limitations and Future Work
While promising, the authors note that these effects may vary depending on the social context (e.g., the Prisoner’s Dilemma). Future research should explore the "depth vs. breadth" tradeoff—determining if more sophisticated emotional models yield even higher cooperation—and the impact of physical presence in social robotics.
Summary updated