Mastering the Dungeon: Grounded Language Learning by Mechanical Turker Descent
This work addresses the challenge of grounded language learning for AI agents in interactive environments, though it is incremental in applying human-in-the-loop methods to a specific domain.
The authors tackled the problem of training agents to execute natural language commands in a fantasy text adventure game by introducing Mechanical Turker Descent (MTD), an interactive learning procedure where Turkers compete and collaborate to train agents, resulting in a better quality teaching signal compared to static datasets.
Contrary to most natural language processing research, which makes use of static datasets, humans learn language interactively, grounded in an environment. In this work we propose an interactive learning procedure called Mechanical Turker Descent (MTD) and use it to train agents to execute natural language commands grounded in a fantasy text adventure game. In MTD, Turkers compete to train better agents in the short term, and collaborate by sharing their agents' skills in the long term. This results in a gamified, engaging experience for the Turkers and a better quality teaching signal for the agents compared to static datasets, as the Turkers naturally adapt the training data to the agent's abilities.