Mustafa Abdulkhader Jassim

Abstract :

This study investigated the impact of an explainable artificial intelligence (XAI) adaptive simulation on building sustainable judgment and trust calibration in algorithmic decisions in varying environmental scenarios with school students. The study used a two-group pretest–posttest design with 160 students (80 students in XAI-adaptive and 80 students in control group) from 8th to 11th grade. Three decision scenarios (stable conditions, drought shift and extreme weather) were administered to all students, and they were evaluated pre- and post-tests on sustainable judgment, trust calibration, overtrust, undertrust, AI literacy and explanation understanding. The XAI-adaptive group was the only one that was explained and adapted during the simulation. Paired and Welch t tests, Cohen’s d, and analysis of covariance were used to analyze the data. The XAI-adaptive group showed a 18.29 point increase in sustainable judgment compared to the control group (d = 2.07), and a 21.65 point increase in trust calibration compared to the control group (d = 2.82). Overtrust decreased by 11.54 points compared to 2.33, and undertrust decreased by 9.68 points compared to 1.01. Improvements were also greater in the areas of AI literacy (20.35 vs 5.63) and explanation understanding (25.52 vs 4.69). The accuracy of the mean scenario was higher in the XAI-adaptive group (74.66) than in the control group (54.84), with the difference being the most pronounced in extreme weather (21.51 points). All differences were significant at p < .001 and were not due to pre test scores. The results suggest that explanation-supported adaptive simulation is effective to support students’ appropriate judgment of algorithmic advice when conditions change, but the large effects warrant further replication with follow-up.