ResearchPod Summary
Researchers conducted four studies in India and the Netherlands to examine whether individuals apply the same prosocial economic preferences used in human-human interactions to robot recipients. Using the dictator game paradigm—a standard economic tool where one person (the dictator) decides how to split a stake with another—the study explored how people allocate money to a diverse set of 18 different commercial robots.
Participants interacted with 18 distinct robots, ranging from consumer appliances to humanoid social robots. The researchers used a data-driven approach to categorize participant perceptions of these robots, rather than relying on predefined scales. By analyzing responses to twelve items concerning robot characteristics, they identified three key predictive factors: Likeability, Anthropomorphism (human-likeness), and Social Functionality. The studies systematically tested whether these factors, along with variations in the game structure (hypothetical versus real money, one-step versus two-step decisions), influenced how much money participants were willing to offer the robots.
Participants consistently allocated money to robots, demonstrating that economic prosociality extends to artificial agents. This behavior mirrored well-documented trends in human-human dictator games, such as reduced generosity when stakes are real rather than hypothetical, and further reductions when the task is structured as a two-step decision process. Among the identified factors, likeability emerged as the strongest predictor of monetary offers. Notably, these results were consistent across participants from India and the Netherlands, suggesting that while robot perceptions may vary, the underlying economic decision-making process toward robots is stable across these populations.
As robots become more integrated into daily life, understanding the social and economic norms people apply to these machines is crucial. This research provides evidence that people treat robots as social actors and that economic preferences toward them are not random. These insights can inform the design of future socially interactive robots, highlighting that physical appearance (anthropomorphism) is not the sole driver of human generosity; rather, perceived likeability and functionality play significant, independent roles in how humans value and interact with technology.
Sam: Students gave money to robots, even when keeping it would leave them better off. But their offers shrank when the money became real, and again when refusing was made explicit.
Alex: That sounds like a result about both robots and experimental design. What did the researchers actually establish?
Sam: Across four experiments involving 451 students, offers tracked how participants perceived the robots, as well as how the choice was presented. The work is by Avantika Dev, Roy de Kleijn, and Sumitava Mukherjee, published in the International Journal of Social Robotics.
Alex: So researchers designing robot interactions need to ask more than whether people will share. They need to know what that sharing measures.
Sam: That’s the central issue. This study suggests human likeness is only part of the story; likeability and perceived usefulness matter too. We’ll unpack the money-sharing task, the evidence behind those associations, and why giving does not necessarily mean caring.
Alex: Start with the task itself. How do you make an economic decision involving a robot comparable to one involving another person?
Sam: The authors use the dictator game: one person receives money and decides how much to give a recipient. The recipient must accept whatever is offered. There’s no bargaining or retaliation built into the game, so keeping everything carries no strategic penalty.
Alex: Which makes giving potentially informative, but not automatically altruistic. Someone might share because that feels like the expected thing to do.
Sam: The authors make that distinction throughout. Offers can be a limited proxy for prosocial preferences, meaning willingness to benefit others. But they may also reflect social norms or the expectations created by an experiment.
Alex: Researchers had already used this task with robots. What does this study add beyond showing that people sometimes give machines money?
Sam: Earlier studies covered relatively few robots, often examining human likeness or particular robot models. This project used 18 commercially available robots with very different appearances and functions. The set ranged from humanoids and companion-like machines to industrial and medical robots.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.
Alex: That broader range lets them ask whether a pattern belongs to one appealing robot or extends across different designs. How did participants encounter them?
Sam: Through static images, not live encounters. Each participant evaluated every robot, with presentation order randomized. They rated perceived characteristics and made an allocation decision for each recipient.
Alex: If human likeness was already a leading explanation, did the authors simply measure that more broadly?
Sam: They deliberately cast a wider net. Their questions covered friendliness, creepiness, wanting to touch or own the robot, and perceived abilities to think or feel. Other questions asked about allowing robots to perform tasks such as cooking, vacuuming, or caring for family.
Alex: Those mix appearance, emotional response, and usefulness. How did they turn that collection into something interpretable?
Sam: They used factor analyses, which look for shared dimensions in patterns of responses. They interpreted the resulting groups as likeability, anthropomorphism, and social functionality. Anthropomorphism means attributing human-like qualities; here, that included perceived thinking and feeling, not just physical resemblance.
Alex: And social functionality means perceived suitability for helping people, rather than demonstrated technical performance?
Sam: Yes—the index captured participants’ willingness to allow those tasks. It was not a performance test. The questionnaire was exploratory, and the authors explicitly say they were not developing a fully validated measurement scale.
Alex: Let’s follow the money. How generous were participants when their decisions had no financial consequences?
Sam: In the initial hypothetical study, they offered about 48 percent of the stake. A subsequent study involving students in India and the Netherlands produced a similar overall average. Asking for robot ratings before rather than after the allocation did not significantly change offers in the initial study.
Alex: Hypothetical generosity can be cheap. What changed when participants actually stood to lose money?
Sam: In the real-money study, one allocation was randomly selected for payment. Participants received what they had chosen to keep on that decision. Average offers were about 35 percent of the stake, lower than in the hypothetical version.
Alex: Was that the same group changing its mind once payment became real? That would affect how directly we interpret the comparison.
Sam: No. These were separate studies with different participant samples. The authors found a statistically significant difference between the hypothetical and real-money samples, but this wasn’t a within-person switch.
Alex: Then they changed the choice itself. What did making refusal explicit involve?
Sam: Instead of immediately requesting an amount, they first asked whether participants wanted to give money to that robot. Only those answering yes proceeded to enter an allocation. With real money still at stake, average offers across decisions fell to about 18 percent.
Alex: So an explicit invitation not to give was associated with much less giving. That makes the task’s implied expectation hard to ignore.
Sam: The authors interpret it that way too. They discuss demand characteristics: cues about how participants think an experiment expects them to behave. The contrast suggests that some giving may come from the setup, rather than a stable preference to benefit robots.
Alex: If the task changes offers so much, do the robot characteristics still tell us anything consistent?
Sam: They do. Across the study variants, higher likeability, human likeness, and social functionality ratings each predicted higher offers in the authors’ models. Likeability showed the strongest association in the later studies’ reported correlations, and in the combined analysis of offer amounts.
Alex: Could a designer read that as evidence that making a robot friendlier will cause people to give it more?
Sam: Not from this design. Participants’ perceptions and offers were measured together; the researchers did not independently manipulate likeability. The evidence supports an association, not a causal recipe for increasing generosity.
Alex: Did the findings travel across countries, or were they mainly driven by one national sample?
Sam: The broad patterns were similar in India and the Netherlands. Country was not a significant predictor of the offered proportion in the later studies’ models, which included robot perceptions. That supports consistency across these samples, not a claim that culture never matters.
Alex: And when the authors compare this with giving to humans, is there a human-recipient condition inside these experiments?
Sam: There isn’t. All recipients here were robots. The similarities to human giving come from comparisons with published dictator-game research, not a direct test against human recipients under the same conditions.
Alex: That leaves a useful but bounded conclusion: predictable allocations to pictured robots, without proof that people treat robots and humans alike.
Sam: And without proof that the allocations express concern for a robot’s welfare. The authors note that robots lack human material needs. Student-only samples, static images, and rating questions that might cue particular judgments further limit generalization to everyday interaction.
Alex: Who should read the full document, and where should they begin?
Sam: Researchers studying robot perception, social robot design, or economic-game methods should read it. Start with the real-money study and the comparison between the one-step and two-step tasks. Then read the aggregate analysis for the perception associations, followed by the discussion’s cautions about experimental expectations and the exploratory questionnaire.
Alex: And for someone taking just one conclusion back to their colleagues?
Sam: Money offered to a robot tells us about both the robot people perceive and the choice we give them.
Alex: Keep both in view when you read the result.