Domain 1 Experiments
Experiment 1
Condition 1 (AI recommendation)
Introduction - Condition 1
Training - Condition 1
Main Task - Condition 1
Condition 2 (AI-explanation-only)
Introduction - Condition 2
Training - Condition 2
Main Task - Condition 2
Condition 3 (Hypothesis-driven)
Introduction - Condition 3
Training - Condition 3
Main Task - Condition 3
Subjective Questions: Twelve subjective questions are as follows. We evaluate Q1-4 separately to measure 4 measures (In control, Preference, Mental demand and System complexity). We aggregate Q5-12 to measure Trust. The questions for trust are based on (Hoffman et al. (2018). Metrics for explainable AI: Challenges and prospects.)
- In control: I feel in control of the decision-making process when using this decision aid. (0 = Disagree strongly; 10 = Agree strongly)
- Preference: I would like to use this decision aid frequently. (0 = Disagree strongly; 10 = Agree strongly)
- Mental demand: I found this task difficult. (0 = Disagree strongly; 10 = Agree strongly)
- System complexity: The decision aid was complex. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: I am confident in the decision aid. I feel that it works well. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: The decision aid is very predictable. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: The decision aid is very reliable. I can count on it to be correct all the time. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: I feel safe that when I rely on the decision aid I will get the right answers. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: The decision aid is efficient in that it works very quickly. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: I am wary of the decision aid. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: The decision aid can perform the task better than a novice human user. (0 = Disagree strongly; 10 = Agree strongly)
- Trust: I like using the decision aid for decision-making. (0 = Disagree strongly; 10 = Agree strongly)
Experiment 2
The design and tasks in Experiment 2 are similar to those in Experiment 1, with the main differences being:
- Participants were asked to explain why they selected an option.
- No subjective questions.