The Utility of Explainable AI in Ad Hoc Human-Machine Teaming

Authors: Rohan Paleja, Muyleng Ghuy, Nadun Ranawaka Arachchige, Reed Jensen, Matthew Gombolay

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we present two novel human-subject experiments quantifying the benefits of deploying x AI techniques within a human-machine teaming scenario.
Researcher Affiliation Collaboration Rohan Paleja1, Muyleng Ghuy1, Nadun R. Arachchige1, Reed Jensen2, Matthew Gombolay1 1Georgia Institute of Technology, 2MIT Lincoln Laboratory 1Atlanta, GA 30332, 2Lexington, MA 02420
Pseudocode No The paper describes the cobot's policy as 'decision tree-based policies' but does not provide any pseudocode or algorithm blocks.
Open Source Code Yes We provide a codebase with our experiment domain at https://github.com/CORE-Robotics-Lab/Utility-of-Explainable-AINeur IPS2021.
Open Datasets No The paper describes human-subjects studies involving participants playing Minecraft in a custom environment. It does not provide access information (link, DOI, citation) for a publicly available dataset of the collected human-subject data or the generated in-game environment data.
Dataset Splits No The paper describes human-subjects experiments with different conditions and phases, but it does not refer to dataset splits like 'training', 'validation', or 'test' sets in the context of machine learning model development.
Hardware Specification No The paper states '[N/A]' for hardware specifications when asked if it included 'the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)?'.
Software Dependencies No The paper mentions software components like 'Microsoft Malmo Minecraft AI Project' and 'Pygame interface', but it does not provide specific version numbers for any of these software dependencies.
Experiment Setup Yes We utilize a 1 × 3 within-subjects design varying across three abstractions: 1) No explanation of the robot’s hierarchical policy, 2) A status explanation of the cobot’s hierarchical policy, and 3) A decision-tree explanation of cobot’s hierarchical policy. ... Both components of the hierarchical policy are decision tree-based policies of depth two and with four leaf nodes.