PaperlayerView exact revision on arXiv

Source-linked AI summary

Trust as indicator of robot functional and social acceptance. An experimental study on user conformation to the iCub's answers

Ilaria Gaudiello, Elisabetta Zibetti, Sebastien Lefort, Mohamed Chetouani, Serena Ivaldi

arXiv:1510.03678v1cs.ROcs.CYcs.HC

TL;DR

The study examines acceptance of humanoid robots as potential partners, including whether users conform to a robot’s answers in functional and social tasks. Its findings indicate that iCub was more readily accepted in functional than social tasks, while the study identifies additional acceptance factors for future investigation.

  • Problem

    Acceptance of robots as potential partners remains a research challenge, particularly for human–humanoid interaction.

  • Method

    The study measured acceptance through participants’ conformity to iCub’s answers in decision-making tasks involving perceptual and socio-cognitive uncertainty.

  • Results

    Participants were more readily accepted iCub’s answers in functional than social tasks, with conformity generally uncommon and concentrated in the functional task.

  • Takeaways & Limitations

    The findings place emphasis on the nature of robot acceptance and indicate that iCub was not yet a reliable partner in social tasks.

  • Takeaways & Limitations

    The study mainly focuses on robot functional and social acceptance, leaving additional aspects to be investigated.

Abstract

from arXiv · show

To investigate the functional and social acceptance of a humanoid robot, we carried out an experimental study with 56 adult participants and the iCub robot. Trust in the robot has been considered as a main indicator of acceptance in decision-making tasks characterized by perceptual uncertainty (e.g., evaluating the weight of two objects) and socio-cognitive uncertainty (e.g., evaluating which is the most suitable item in a specific context), and measured by the participants' conformation to the iCub's answers to specific questions. In particular, we were interested in understanding whether specific (i) user-related features (i.e. desire for control), (ii) robot-related features (i.e., attitude towards social influence of robots), and (iii) context-related features (i.e., collaborative vs. competitive scenario), may influence their trust towards the iCub robot. We found that participants conformed more to the iCub's answers when their decisions were about functional issues than when they were about social issues. Moreover, the few participants conforming to the iCub's answers for social issues also conformed less for functional issues. Trust in the robot's functional savvy does not thus seem to be a pre-requisite for trust in its social savvy. Finally, desire for control, attitude towards social influence of robots and type of interaction scenario did not influence the trust in iCub. Results are discussed with relation to methodology of HRI research.

Introduction

Robot acceptance includes functional and social dimensions, but existing approaches often treat robots mainly as technologies rather than social agents. This study examines trust through users’ behavioral conformation to iCub’s decisions across functional and social tasks.

  • Functional acceptance concerns perceived usefulness, ease of use, accuracy, and innovativeness, whereas social acceptance concerns presence, sociability, and social intelligence.
  • Industrial and research trends increasingly target robots that combine functional and social abilities.
  • Trust is treated as a shared ingredient of functional and social acceptance and as a potentially useful indicator of sustained human–robot interaction.
  • Functional acceptance is defined as trust in the robot’s functional savvy, while social acceptance is defined as trust in its social savvy.
  • The authors adapt the Media Equation paradigm by using users’ conformation to robot decisions as an objective trust measure across functional and social tasks.
  • The study asks whether users trust robots’ functional and social savvy, whether functional trust precedes social trust, and which individual or contextual factors influence trust.

1. Trust as a fundamental indicator of acceptance

Trust is presented as a behavioral and psychological indicator of human–robot acceptance, especially when users rely on robots under uncertainty. The study motivates conformation as an objective measure that complements explicit trust reports.

  • Trust involves confidence that a system will perform appropriate actions and is relevant when people rely on robots for decisions under uncertainty.
  • The literature distinguishes subjective trust measures, such as questionnaires, from objective measures based on behavioral data.
  • Questionnaires can alter response spontaneity and may not reveal effective trust, while objective and explicit measures often correlate weakly.
  • Prior Media Equation research found greater computer trust and conformity when task success depended on collaboration with the computer.
  • Conformation means withdrawing one’s decision to comply with the machine’s decision, thereby indicating whether users trust the agent’s savvy over their own.
  • Conformation is especially relevant for collaborative or delegated HRI tasks that require sharing duties, responsibilities, and decision authority.

2. Factors influencing robot trust

The paper considers human-, robot-, and environment-related influences on trust, focusing on desire for control, attitudes toward robot social influence, and interaction scenario. These factors are motivated by prior findings about autonomy, attitudes, and cooperation.

  • Following a three-factor model, the study examines human-related, robot-related, and environment-related influences on trust.
  • 2.1 Personality traits: desire for control: Desire for control is an intrinsic tendency to control life events and may affect willingness to let robots take charge of tasks.
  • 2.1 Personality traits: desire for control: The study tests whether desire for control negatively correlates with conformity, defined as giving the robot the last word in decision-making.
  • 2.2 Attitudes towards robots: social influence: The NARS assesses negative attitudes toward robot interaction, social influence, and emotions through Likert-rated subscales.
  • 2.2 Attitudes towards robots: social influence: Negative attitudes toward robot social influence are expected to reduce trust in robots’ decisions about social issues and reduce conformity.
  • 2.3 Interaction scenario: Prior scenario studies reported mixed effects, with collaboration sometimes raising trust and imagined-contact effects differing across cooperative, competitive, and neutral conditions.

3. Our study

The study examined trust in iCub’s functional and social savvy, testing task type, participant characteristics, and interaction scenario as potential influences on robot acceptance.

  • Aims and trust measure: The study investigated whether participants conformed to iCub’s answers differently for functional and social decision-making under uncertainty.Conformation was defined as modifying an initial decision to comply with the robot’s answer and was used as a trust measure.
  • Individual and contextual factors: The study tested whether desire for control, negative attitudes toward robot social influence, and collaborative versus competitive scenarios affected trust.These factors were treated as possible influences on functional and social conformation.
  • Decision-making tasks: Participants completed functional judgments about perceptual characteristics and social judgments about item suitability in specific contexts.Functional judgments concerned weight, pitch, and dominant color; social judgments concerned paired items and social contexts.
  • Decision-making tasks: The robot answered the same questions after participants made initial decisions, after which participants could retain or change their answers.This procedure operationalized trust through agreement or disagreement with iCub’s responses.
  • Main hypotheses: Participants conformed more to the robot’s functional than social answers, confirming the first hypothesis.The study therefore found stronger trust in iCub’s functional savvy than in its social savvy.

H3 Participants who imagine a collaborative HRI scenario tend to trust more the robot

The study hypothesized that imagining a collaborative interaction would increase participants’ trust in iCub, while negative attitudes toward robot social influence would reduce social conformation.

  • Interaction scenario: The collaborative-scenario hypothesis predicted greater conformity to iCub’s answers in both functional and social tasks.This prediction was based on evidence that imagined collaborative contact can increase confidence toward a real robot.
  • Attitudes toward social influence: The study also predicted that stronger negative attitudes toward robots’ social influence would reduce conformity in social tasks.The relevant NARS subscale addressed concerns that robots might influence, dominate, or create dependence.

H5 The more participants show a strong desire for control as a personality trait, the less

The study treated desire for control as a possible individual difference affecting trust, while documenting the participant sample and controlled iCub setup used for the experiment.

  • Desire for control: The desire-for-control hypothesis predicted that higher DFC scores would be associated with less conformity to iCub in functional and social tasks.The prediction specified negative correlations between DFC scores and both conformation scores.
  • Participants: Fifty-six healthy French-speaking adults participated, including 37 women and 19 men aged 19 to 65.Participants were recruited from two universities and an online cognitive-sciences network.
  • Experimental setup: The experiment used a fixed iCub setup with standardized seating, a display screen, cameras, and a hidden operator controlling robot behavior.The Wizard of Oz paradigm made participants believe they were interacting with an autonomous system while an operator remotely controlled the robot.
  • Robot behavior: The robot used neutral or positive expressions and preprogrammed verbal answers during the tasks.The operator could also enter new sentences during unforeseen questions.
  • Measures: Participants completed NARS and DFC questionnaires to measure attitudes toward robot social influence and desire for control.Both questionnaires used Likert-type response scales; NARS’s second subscale measured attitudes toward social influence.

The functional task

The functional task measured whether participants changed judgments about perceptual properties after hearing iCub’s answers, using ambiguous and unambiguous stimuli.

  • Task structure: Participants compared physical stimuli to judge relative weight, pitch, or dominant image color.The task included image, sound, and weight subtasks, each requiring a choice between paired alternatives.
  • Task procedure: After each participant answered, the experimenter asked iCub the same question and invited the participant to confirm or change the initial decision.Participants and the robot could therefore agree or disagree on each evaluation.
  • Stimuli: Each subtask contained four paired evaluations, including three ambiguous and one unambiguous stimulus pair.Ambiguous pairs were used to assess behavior under strong uncertainty.
  • Robot responses: For ambiguous items, the robot was programmed to contradict the participant, while it agreed on completely unambiguous items.This design reduced the risk that participants would interpret the robot as always contradicting humans or as faulty.
  • Materials: The image materials comprised four evaluations of dominant color, while the weight materials comprised bottle pairs differing in similarity and weight.The figures document the images and bottle stimuli used in the functional task.

The social task

The social task assessed trust in the iCub’s social savvy through context-based judgments about which of two objects was more appropriate. Participants answered first, heard the robot’s contradictory response, and could confirm or disagree.

  • Task design: The social task was designed to assess trust in the robot’s social savvy through subjective, context-dependent judgments.This contrasted with the functional task, which evaluated perceptual characteristics using objective measures.
  • Task design: Participants selected the more appropriate of two objects across three everyday contexts, where there was no right answer.Contexts included school, a swimming pool, and a rainy day.
  • Task design: The experimenter asked the question to the participant, then the robot, and finally asked the participant to confirm or reject the robot’s answer.The robot’s answer was chosen to contradict the participant’s initial response.
  • Experimental design: All 56 participants completed both the functional and social tasks, presented in random order.Participants were assigned to collaborative, competitive, or neutral imagined scenarios.
  • Measures: Trust was quantified as a conformation score: changes matching the robot divided by instances where the robot disagreed with the initial answer.The score ranged from 0, disagreement, to 1, conformation.

4. Results

Participants conformed more to iCub’s functional than social answers, but overall conformation remained below the study’s trust threshold. Imagined interaction scenario also showed no reliable effect on either task, although the scenario manipulation was not fully remembered as intended.

  • Functional versus social trust: 0.315 (σ=0.201) was the average functional conformation score, versus 0.199 (σ=0.221) for the social task.Both means were below the 0.5 threshold used to indicate trust.
  • Functional versus social trust: V=981; p<.001 indicated a significant difference, with greater conformity to functional than social answers.The result confirmed the hypothesis that participants trusted the iCub’s functional savvy more than its social savvy.
  • Social-task distribution: 92.8% of participants had a low social-task conformation score, with 48.2% always disagreeing and 44.6% conforming on one of three disagreements.Only 1.8% always conformed, while 1.8% scored 0.5 and 3.6% scored 0.67.
  • Social-task distribution: The few participants with high social-task conformation generally showed lower functional-task scores, so the second hypothesis was not confirmed.Two of the three participants with high social conformation had lower functional scores, while one scored 0.5.
  • Imagined scenario: χ2=1.69; p=N.S. for the functional task and χ2=1.63; p=N.S. for the social task showed no effect of imagined scenario.Collaborative, competitive, and neutral conditions therefore did not produce different conformation scores.
  • Imagined scenario: Mcompet=3.57, Mneutral=4.76, and Mcollabe=5.21; χ2=13.83; p<.001 showed that participants remembered the imagined scenarios differently.The competitive group nevertheless described its scenario as less competitive than instructed, motivating caution in interpreting the null trust effects.

trust in the robot social savvy? (H4)

The study tested whether negative attitudes toward robot social influence or desire for control predicted trust in iCub’s functional or social savvy. Neither individual difference measure significantly correlated with conformation in either task.

  • Negative attitudes toward robot influence: These results did not confirm that negative attitudes toward robots’ social influence predict trust in either functional or social savvy.The analysis used NARSS2, the relevant subscale of the Negative Attitude towards Robots Scale.
  • Negative attitudes toward robot influence: ρ=.127; p=N.S. for functional conformation and ρ=.127; p=N.S. for social conformation showed no significant association with NARSS2 attitudes.No difference in these correlations appeared across the three imagined scenarios.
  • Desire for control: The results did not confirm that greater desire for control negatively correlates with trust in the robot’s functional or social savvy.Desire for control was measured with the DFC scale.
  • Desire for control: ρ=.086 for functional conformation and ρ=.137 for social conformation showed no significant correlation with desire for control.The pattern did not differ across the three imagined scenarios.

5. Discussion

Participants generally trusted iCub more for functional than social decisions, while trust in functional savvy was not a prerequisite for trust in social savvy. The discussion links this pattern to task nature and to the robot’s ambiguous status as both social and functional.

  • Functional and social trust: Participants generally did not conform easily to iCub, but those who conformed did so mostly in the functional task.Participants sometimes relied on iCub’s perception rather than their own, whereas others cited task-specific expertise for rejecting its answer.
  • Functional and social trust: Functional decisions elicited more reliance on iCub because participants treated its perceptual abilities as useful for discriminating precise stimulus characteristics.Several participants preferred iCub’s perception to their own, while expertise could lead participants to reject the robot’s answer.
  • Functional and social trust: Participants rarely accepted iCub as a reliable partner for social tasks, partly because they questioned its experience and ability to apply context-specific norms.Comments referenced the robot’s lack of experience with swimming pools or rain and its inability to know relevant social practices.
  • Functional and social trust: The few participants who conformed in the social task conformed less in the functional task, contradicting the expectation that functional trust underpins social trust.This minority treated social savvy as potentially reliable without making the same assumption about technical or objective knowledge.
  • Interpretation and implications: The pattern suggests that participants may distinguish the robot as either a functional assistant or a social companion rather than integrating both roles.The authors connect this possible dichotomy to the robot’s anthropomorphic appearance and its stronger acceptance as a functional tool.
  • Individual and contextual factors: Desire for control, attitude toward robot social influence, and interaction scenario did not explain participants’ conformation to iCub.Participants with both high and low desire for control mainly tended not to conform, and the imagined scenario showed no effect on functional or social results.

6. Conclusions

The study presents robot acceptance as complex, with prevailing distrust and task-dependent trust behaviors. Participants trusted the robot more on functional than social decisions, while the examined HRI factors did not influence conformation.

  • Robot acceptance is described as a complex dynamic characterized by prevailing distrust and distinct trust behaviors across tasks.
  • Participants trusted robots more on functional decisions than on social decisions.
  • The minority trusting robots on social issues showed significant distrust on functional issues.
  • The findings suggest that trust in functional and social savvy are mutually exclusive because they involve different kinds of knowledge and skills.
  • The study argues that robot trust cannot be assimilated to mechanisms typical of trust in computers or humans.
  • Desire for control, negative attitudes toward robot social influence, and imagined interaction scenario did not influence conformation to robot decisions.

Annex 1: The WoZ GUI

The Wizard-of-Oz interface organized robot control into task-specific tabs for movement, speech, and facial expression. Operators could use predefined commands and sentences, while also entering unexpected speech responses on the fly.

  • The WoZ GUI used separate tabs to control robot movements, gaze, hands, posture, speech, and facial expressions.
  • Movement controls sent predefined commands to the actionServer and included preprogrammed whole-body postures.
  • The speech tab offered a list of predefined sentences and allowed operators to type responses to unexpected participant requests.
  • The facial-expression tab loaded LED configurations for specific expressions from a configuration file.

Annex 2 The NARS test and DFC test with their French translation

The annex describes translated NARS and Desire For Control questionnaires and three functional-trust subtasks using sounds, images, and bottle weights. These tasks varied stimulus ambiguity and required participants to compare the robot's answer with their own judgment.

  • The NARS and Desire For Control questionnaires were presented in English and double-translated French versions.
  • The functional-trust battery included sound, image, and weight evaluations.
  • Sound trials compared pairs of pure tones lasting 1.5 seconds across equal, different, and ambiguous frequency conditions.
  • Image trials asked participants to identify a dominant color from three options, with robot answers sometimes matching the participant's answer.
  • Weight trials used four bottle pairs varying in appearance and weight, after which participants reconsidered their decision following the robot's answer.
Loading 1510.03678v1…