Carl Westin

66 Source bias effects engaged in anticipatory behavior and considered or crosschecked raw data when handling alerts. 4-2-2 Source bias controversies A recent meta-analysis of trust concluded that people in general trust themselves or other human sources more than they trust automated sources. 46 The meta-analysis was limited by the sparse amount of research available, with only nine statistics from three studies identified: Madhavan and Wiegmann; 167 de Vries et al. , 178 and Ma and Kaber. 166 When considering trust before interaction with the automated source, however, the analysis of these three studies show that the automated sources were preferred over the human sources. Note that the study by Madhavan and Wiegmann was discussed in the previous section. De Vries et al. investigated differences between student’s self-confidence in manual control and their trust in an automated navigation aid in a city-driving plan- ning task. As such, their study did not compare trust in an automated source relative to a human source. Results were consistent with previous research, that manual er- rors affected participants’ self-confidence less than automation errors affected their trust in automation. Overall, participants were considered to have higher trust in themselves (preference for manual control) than in the automated aid. 178 Ma and Kaber investigated students’ trust in a human source and automated source providing navigation support in a suburban driving scenario. Before and dur- ing initial interaction with the decision aids (before errors were made), participants expected less error and had higher trust in the automated source. After interaction and the onset of errors, trust declined for both sources. However, the difference did not vary significantly between sources, although the human source was slightly preferred. 166 A possible limitation of the study is the use of different modalities: the human source provided support over the phone, while the automated source provided support through a text-based interface. Taken together, their findings sup- port the perfect automation schema apparent before interaction, with a subsequent decrease in trust occurring when the advisers err during interaction. 4-2-3 Anthropomorphism and strategic conformance In relation to source biases, research on anthropomorphic automation (having hu- man characteristics) is interesting as it aspires to merge the boundaries that sep- arate machine from human. It is reasonable to believe that the more human-like an automated system is (e,g, by appearance, behavior, or reasoning), the more it is perceived and treated as a human. Researchers believe that anthropomorphism can benefit human-automation teamwork including trust and acceptance, a reduction of