Trust in Automation Experiments

The past decade has seen incredible strides made in the ability for computers and algorithms to forecast social events. One of the most difficult questions in the use of these algorithms, however, is the degree to which humans trust them to assist in decision- making. If the algorithm does perform better, we want a human to use the algorithm’s forecasts. We do not, however, want humans to be so reliant on the algorithm that they fail to note when the algorithm is leading them astray. These two alternatives have been labeled “algorithms aversion” and “automation bias”, respectively. While the literature on these twin dangers is already large in aerospace and is growing in the medicine and computer science, little has been done to evaluate these issues with regards to political decision-making. Moreover, there are clear blind-spots in the literature, such as the difficulty in understanding the heterogeneity of effects based on when information is presented and the area of algorithm appropriateness. Building on pilot studies we have conducted, we seek to field more experiments testing the degrees to which humans trust automation and algorithms in decision making.