FocusHUMAN-ROBOT INTERACTION

Measures of incentives and confidence in using a social robot

See allHide authors and affiliations

Science Robotics  22 Aug 2018:
Vol. 3, Issue 21, eaat6963
DOI: 10.1126/scirobotics.aat6963

Abstract

Measures of incentives and confidence in using a social robot were stable, predictive, and sensitive to changes in robot behaviors.

Rapid, recent advances in applications of social robots run the risk of exceeding potential users’ willingness to engage with them. This highlights a need to assess factors that influence their acceptance and use (1). Current measures of users’ perceptions typically assess a single psychological dimension (2), confound multiple constructs (3), assess global attitudes toward robots (4), and/or do not directly measure the level of social and emotional connection with social robots. To advance research in this area, an assessment measure that can quantify changes in social robot characteristics or behaviors and predict willingness to use social robots is required.

These new measures should have their basis in well-established psychological theory about the prediction of behavior. Two predictors have proved particularly powerful: self-efficacy (confidence in meeting current performance demands) and incentives (expectations about outcomes from an action) (5, 6). Currently, we know of no published measure of self-efficacy or incentives about interpersonal interactions with a social robot. We developed and tested measures to fill this gap, examining their sensitivity to contrasting robot behaviors and their prediction of intentions to use a social robot.

Initial development of the measures involved administering them to groups of 202 female high school students who had observed an interaction with a NAO robot. Exploratory factor analyses indicated that the Robot Incentives Scale (RIS) had three subscales: emotional (liking or enjoyment of social robots), utility (whether they were useful or solved problems), and social/relational (their potential for social connection). The Robot Self-Efficacy Scale (RSES) had two subscales: operation (confidence in operating a social robot) and application (confidence in completing a task or goal with use of the robot). Intentions to use a social robot formed a single scale. The internal structures of the scales were confirmed in an online adult sample of 404 participants (Table 1), although the RSES required one item to be removed and errors on two pairs of similar items to be correlated before an acceptable fit was obtained. The internal consistencies of the subscales for all three scales were very high.

Table 1 The robot incentives scale, robot self-efficacy scale, and robot usage intention items.

All items were rated on a 0 to 10 scale, with only end points labeled. RSES items are rated from “Sure I can’t” to “Sure I can,” and RIS and RUI items are rated from “Not at all” to “Definitely.”

View this table:

We tested whether the new scales were sensitive to more mechanical or humanoid behaviors from the NAO robot by using a student sample. The more humanoid robot received higher scores on all three measures but only when the students saw both types of behavior. The RIS and RSES subscales jointly predicted 78% of the variance in intentions to use the social robot in the student group. On their own, RIS subscales predicted 77% of the variance, whereas RSES subscales predicted only 40%. The adult group gave similar results, with RIS and RSES predicting 83% of the variance in intentions when used together (82 and 54% separately). In both groups, all RIS and RSES subscales except RSES operation contributed unique predictive variance.

In summary, the three scales provided a coherent and stable factor structure across the studies despite differences in the nature of the samples and observed interactions. However, the RSES required omission of one item to finalize an acceptable fit, suggesting a need for further replication. High internal consistencies of RSES and RIS subscales suggest a potential for shortening the length of assessment measures without affecting reliability. The measures were sensitive to comparisons of different social robot behaviors when individuals were able to contrast the behaviors. Because most participants had little prior exposure to robots, observations of the interactions were presumably dominated by their novelty. Future groups that have extensive experience with robotics or human-robot interactions may not need contrasting interactions to obtain differential ratings, because comparisons with previous experience would be available.

All RIS and RSES subscales made unique contributions to a concurrent prediction of intentions to use a social robot, except for RSES operation. However, almost all of the predictive power was from the RIS subscales, suggesting that assessments of incentives may be sufficient to predict intentions to use a robot for a social interaction. The limited prediction from self-efficacy was unexpected, because it is typically a stronger predictor of performance attainments than incentives (6). However, the focus of the intentions measure was on a social interaction rather than on controlling or programming the robot, making self-efficacy less relevant than incentives. If participants were required to undertake a more demanding role, then a different pattern of results may be obtained. As yet, we have not tested the ability of the measures to predict actual use of the robot. Generalization of results to other social robots, characteristics, and contexts also await determination. These studies provide strong initial support for the new measures, and the RIS may have wide potential application in assessing the acceptability of social robots.

SUPPLEMENTARY MATERIALS

robotics.sciencemag.org/cgi/content/full/3/21/eaat6963/DC1

Text S1. Study design, materials, and methods.

Text S2. Exploratory factor analyses: Further notes.

Text S3. Comparisons of mechanical and humanoid robot behaviors.

Text S4. Confirmatory factor analysis.

Text S5. Prediction of intentions from self-efficacy and incentives.

Fig. S1. Ratings of more mechanical and humanoid robot interactions.

Table S1. Results of exploratory factor analyses: Robot Incentives Scale (RIS).

Table S2. Results of exploratory factor analyses: Robot Self-Efficacy Scale (RSES).

Table S3. Results of exploratory factor analyses: Robot Usage Intention (RUI).

Table S4. Between-group differences on each subscale.

Table S5. Between-group differences on each subscale after the first robot observation.

Table S6. Time by condition results for each subscale.

Table S7. Results of confirmatory factor analyses.

Table S8. Multiple regressions predicting intentions from self-efficacy and incentives in the student sample.

Table S9. Multiple regressions predicting intentions from self-efficacy and incentives in the adult sample.

Data S1. Raw scores for scale development and experimental studies.

Reference (7)

REFERENCES AND NOTES

Acknowledgments: We thank the Australian Centre for Robotic Vision, Centre for Children’s Health Research, Institute of Health and Biomedical Innovation, and Queensland University of Technology for their support in this project. Funding: L.H. is funded by a National Health and Medical Research Council Senior Research Fellowship and by Lives Lived Well, a non-profit charity. N.L.R. was supported by the Research Training Program (RTP) scheme on behalf of the Department of Education and Training. Author contributions: N.L.R. designed and conducted the studies, data collection, analyzed the data, and completed the first draft of this paper. J.C. contributed to the design, supervision of the conduct of the studies, and the analyses and gave comments on the paper. G.M.J. and Y.K. assisted with collection of the student data and commented on the draft of the manuscript. L.H. contributed to the analyses and commented on the paper. D.J.K. was the senior researcher on this project and supervised all stages of it. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. The data for this study have been deposited on https://github.com/nrbsn/scirobo. Copies of the code and surveys for the two studies are available from the corresponding author.

Stay Connected to Science Robotics

Navigate This Article