A Meta-Analysis of Factors Affecting Trust in Human-Robot Interaction

Peter A. Hancock, Deborah R. Billings, Kristin E. Schaefer, Jessie Y.C. Chen, Ewart J. de Visser, Raja Parasuraman

Human Factors: The Journal of the Human Factors and Ergonomics Society, September 2011

Summary

                  Robots are commonly used in unreachable or unsafe environments for human beings, more an more, starting from the military, robots are views as a tool to be manipulated by humans to accomplish specific discrete functions. With the growing of robot capabilities, robots are starting to be implemented more and more in team working environments where there team mates are humans, cases like this already are reality in military applications (an interesting example is the Sgt. Talon robot, which even received promotions and Purple Hearts for its stellar bomb disposal performance). In team working with a robot, as for human-human interaction, necessary is that a certain level of trust is perceived, it is demonstrated (Park, Jenkins, & Jiang, 2008) that trust affects decision making, particularly in uncertainty or risky conditions.

There are cases of untrusted robots, as for SWORD (Special Weapons Observation Reconnaissance Detection), which had to be implemented in Iraq, but soldiers did not trust it in the case of dangerous operation environments due to technological malfunctions which caused errors in its movements.

An inappropriate trust level in machines and robots can cause several problems even on the organizational side, in fact it results in overreliance on and misuse of the system (for over-trusting cases) or in disuse of the system entirely (for under-trusting cases); therefore there must be a certain level of tolerance to be appropriately calibrated according to the capabilities of the robot itself.

The authors identify three possible antecedents of trust: robot-related factors, human-related factors and environmental-related factors; according on these three they enable a quantitative review of the predictive strength of these respective trust factors in human-robot teams.

·       Method

Studies regarded correlation and group design data, so that multiple meta-analytic methods have been applied (correlation and Cohen’s d). The correlation is needed to illustrate the effects of a give factors., while Cohen’s d indicates the standard difference between two means in standard deviation units. Cohen’s d (1988) have established ranges: small (d<.20;r<.10), medium (d=.50;r=25) and large (d>.80;r>.40).

Variance Estimates were also calculated: s²_g (variability on the effect sizes themselves) and s²_e (variability due to sampling error); these two allow to obtain the residual variance (s²δ), which is an indication of the hetereogenousity of the effect sizes (requiring in affermative case one more variable to moderate a particular effect). A final check is the homogeneity of variance (s²_e /s²_g), for which, according to Hunter and Schmidt (2004), a value greater than 0.75 represents homogeneity.

Key Concepts

Antecedent of trust in Human-Robot team working.

Key Results

On 10 studies, the results reported that there is an moderate global effect between trust and all factos (r=+0.26), small is the effect regarding human dimensions and environmental characteristics (r=+0.11), while robot characteristics are reported to affect moderately (r=+0.24). It appears that performance factors are more influential than robot attributes (r=+0.34 against r=+0.03). The experimental analysis, which reported group studied and were carried according to Cohen’s d, illustrated greater influence of the global effect (d=+0.67), morderate for environmental factors (d=+0.47), very small for human factors (d=-0.02), great for robot performances (d=+0.71) and moderate for robot attributes (d=+0.47).