A key challenge on the path to developing agents that learn complex human-like behavior is the need to quickly and accurately quantify human-likeness . While human assessments of such behavior can be highlyaccurate, speed and scalability are limited . We address these limitationsthrough a novel automated Navigation Turing Test (ANTT) that learns to predicthuman judgments of human-likes . We demonstrate the effectiveness of ourautomated NTT on a navigation task in a complex 3D environment. We investigatesix classification models to shed light on the types of architectures bestsuited to this task, and validate them against data collected through a humanNTT . Our best models achieve high accuracy when distinguishing true human andagent behavior.
Author(s) : Sam Devlin, Raluca Georgescu, Ida Momennejad, Jaroslaw Rzepecki, Evelyn Zuniga, Gavin Costello, Guy Leroy, Ali Shaw, Katja HofmannLinks : PDF - Abstract
Code :
Keywords : human - navigation - behavior - ntt - task -