Testing the Robustness of the Gittins Index Through A Virtual Slot Machine Simulation

Abstract

This research delves into the assessment of the Gittins Index, a predictive metric pivotal in guiding artificial intelligence agents within multi-armed bandit scenarios. The study unfolds through two trials, one devoid of Gittins Index guidance and the other with the index informing decision-making in a virtual slot machine simulation. Results reveal a notable increase in credit accumulation in the guided trial, maintaining a comparable machine count across both trials. The Gittins Index consistently demonstrates high predictive accuracy. Conclusively, the Gittins Index emerges as a robust tool, offering valuable insights for optimizing decision processes in artificial intelligence training, exemplified by its efficacy in navigating search spaces with enhanced efficiency.