Investigation of algorithmic facilitation vs. self-led learning in complex skill development for operational training

Abstract

Adaptive-difficulty algorithms can automatically modulate task difficulty to improve skill acquisition of complex tasks. Using virtual reality (VR) enables training in an immersive way without requiring an in-person training facilitator or complex training facilities. However, it is not known whether an adaptive-difficulty algorithmic facilitation can outperform highly-motivated individuals facilitating their own training by self-selecting the difficulty level based on their own preferences. This study enrolled individuals (n = 45) with specialized aviation education to train using VR and then perform a spacecraft entry, descent, and landing task. Subjects were randomly assigned into one of three training groups: algorithm-facilitated condition, self-led condition, and static difficulty condition. Subjects trained for 30 trials (i.e., landings) in immersive VR, followed by 5 evaluation trials. Task performance, difficulty level, and crashes, as well as subjective reports of flow and motivation were collected. Despite training at a lower difficulty level, on average, than the self-led group (p = 0.0006) the algorithm-facilitated group had improved task performance compared to the static group (p = 0.004) and marginally significance compared to the self-led group (p = 0.097). Further, the self-led group reported lower confidence, as a subscale within flow (p = 0.048), and decreased flow late in training (p = 0.02) compared to the algorithm-facilitated group. The results of this study inform the utility and efficacy of algorithmic facilitation in learning complex tasks.