Automation 1
- 1. Automation: not all trust is created equal Vicki Chiang 10/1/13
- 2. Not All Trust Is Created Equal: Dispositional and History-Based Trust in Human-Automation Interactions Stephanie M. Merritt and Daniel R. Ilgen Human Factors: The Journal of the Human Factors and Ergonomics Society 2008 Central goal of the paper is to show empirically the connections between individual user perceptions and trust in automation interactions • User personality traits such as extraversion mediates trust • It is understood that there is a correlation between trust in machines and trust in other people, but individual traits are not parsed out (Parasuraman and Miller, 2004 and Parasuraman and Riley, 1997) • Trust is dynamic and can be divided into dispositional/initial trust (interaction naïve) and post-task trust (after interaction)
- 3. 13 Hypotheses ! 1: Initial trust is positively related to automation use on a subsequent task. 2: Propensity to trust machines is more strongly related to initial trust than to post-task trust. 3: Initial trust mediates the relationship between propensity to trust machines and automation use. 4: Extraversion is positively related to propensity to trust machines. 5: Initial trust in an automated system is positively associated with post-task trust. 6: Initial trust and post-task trust show evidence of discriminant validity. 7: Machine characteristics are more strongly related to post-task trust than to initial trust. 8: Automation use moderates the relationship of machine characteristics and post-task trust, such that machine characteristics have a greater effect on post-task trust when use is high. 9: Significant variance in perceptions of machine characteristics exists even when actual machine characteristics are constant. 10: Machine characteristics moderate the relationship of propensity to trust machines and post-task trust. 11: Machine characteristics will moderate the relationship of propensity to trust machines and perceptions of machine characteristics, such that more extreme levels of post-task trust will result when both variables are either low or high. 12: Perceptions of machine characteristics account for additional variance in post-task trust beyond the effects of the actual machine characteristics. 13: Perceptions of machine characteristics mediate the relationship between the interaction of propensity to trust with actual machine characteristics and post-task trust.
- 4. Experimental design • X-Ray screening task o Computer based simulation of x-ray baggage screening o X-ray images of suitcases with and without weapons (limited to guns and knives) are shown on a monitor o Study participants must quickly choose to “clear” or “search” the baggage, based on whether or not they believe that weapons are present. o Participants are told that there is an automated machine available to assist them (Automatic Weapons Detector or AWD)
- 5. Experimental Design Continued • 255 undergraduate student participants o Study participants are separated into two groups: high machine function and low machine function o Study participants also answer a questionnaire to determine their “propensity to trust machines”, extraversion as well as control variables o Participants receive AWD instruction and 1 minute demo that included an error or breakdown in the AWD, after which they rated their levels of trust. o 20 minutes to screen as many bags as they could o Lots of stats and correlations collected on participants
- 6. X-Ray Screening Task Figure 1. X-Ray Screening Task screenshot. Downloaded from hfs.sagepub.com by HFES General on June 15, 2012 To simulate real world situations the operator must be quick as well as accurate so a points system is assigned. Top scorer wins $50.
- 7. Perceptions on machine characteristics for high and low AWD function groups are statistically significant AWD Machine Characteristics High vs Low Function
- 8. Factors and Correlations Difference between initial trust and post-task trust is significant Propensity to trust machines strongly correlates with initial trust, but does not correlate with post-task trust. Extraversion positively correlates to propensity to trust machines.
- 9. Automation use, machine characteristics and trust • Automation use moderates the relationship of machine characteristics and post-task trust, such that machine characteristics have a greater effect on post-task trust when use is high. Figure 2. Interaction between machine characteristics and automation use predicting post-task trust. Figure 3. Interaction between propensity to trust machines and machine characteristics predicting post-task trust. • Machine characteristics moderate the relationship of propensity to trust machines and post-task trust.
- 10. User Perceptions and Post-task Trust • Hypothesis 12: Perceptions of machine characteristics account for additional variance in post-task trust beyond the effects of the actual machine characteristics • Hierarchical linear regression model was used and “the effects of the actual machine characteristics were significant (β = .51, p < .01)”. • When perceptions were accounted for and added to the model, it resulted in a large increase in variance (∆R2 = .52, p < .01, 52%). Overall, the model accounted for 78% of the variance in post- task trust.
- 11. Conclusions • The study empirically supports differences in trust in human-automation interaction o Trust is dynamic and affected by different factors • User propensity to trust automation and machine characteristics affect trust and vice versa o Post-task trust is negatively affected when users with higher initial trust are paired with a less reliable machine. o Post-task trust is less effected by user propensity to trust machines with more machine use • User perceptions of automation are important and measurable o The same machine will be perceived differently by individual users o Individual perceptions affect variance in trust by 52%
- 12. Future Research • Research should take user’s individual perceptions into account when studying trust • Trust should be measured at different points in time for greater accuracy • Training on machines should be more individualized to take user expectations and perceptions into account o Is this practical to implement?
- 13. Thank You
Editor's Notes
- #2: hello
- #3: Parasuraman and Miller showed that etiquette in machine communication affected trust in automationParasuraman and Riley reviewed several papers that showed a connection between trust and automation reliance in different situations, but not through the lens of individual differences.
- #4: All hypothesis point to individual perceptions of automation and the relationship between those perceptions and the different stages of trust.
- #6: After 20 minutes questioned post-task
- #7: Here is a picture of what the X-ray baggage simulation would look likeAnd the points system to encourage accuracy and speed, just like in real life airport baggage screening. The top score won a prize.
- #8: The participants were split into two groups, high machine function and low machine function, where the machine characteristics were different.They checked to make sure that in each characteristic there was a statistically significant difference between high and low groups.
- #9: Table shows all the factors measured for the participants and the correlation between several factors.Difference between initial trust and post-task trust is significant-assessments of trust taken at different times seem to represent two different qualitative forms of trustTraits such as extraversion positively correlates to propensity to trust machines, which correlates with initial trust, but does not correlate with post-task trust. An example of how individual user differences affect trust. These means were used to calculate several models to prove or disprove hypothesis.
- #10: On the flip side,Automation use and machine characteristics also influence trust. Low automation use, post-task trust is essentially the same, but when use is high the difference in trust is significantWhen the propensity to trust machines is high, trust is greater in a high function machine, but drops much lower in the low function machine setting. The authors believe this has to do with the expectation of machine function that does not match the outcome, so the user loses trust in the machine.
- #11: However, that although the actual machine characteristics were fixed and the same (within each high/low function condition), the perception of the machine characteristics varied in post-task trust.