Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Machine Explanations and Human Understanding

Authors: Chacha Chen, Shi Feng, Amit Sharma, Chenhao Tan

TMLR 2023 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To validate our theoretical claims, we conduct human subject studies to show the importance of human intuitions.
Researcher Affiliation Collaboration Chacha Chen* EMAIL Department of Computer Science University of Chicago, Shi Feng* EMAIL Department of Computer Science University of Chicago, Amit Sharma EMAIL Microsoft Research, Chenhao Tan EMAIL Department of Computer Science University of Chicago
Pseudocode No The paper uses causal diagrams and theoretical frameworks, but there are no sections or figures explicitly labeled as 'Pseudocode' or 'Algorithm', nor are there structured code-like procedures presented.
Open Source Code Yes Also available at https://github.com/Chacha-Chen/Explanations-Human-Studies.
Open Datasets Yes Inspired by the Adult Income dataset (Blake, 1998), we choose the task of predicting a person’s annual income based on their profile because people generally have intuitions about what factors determine income but are unlikely to know every person’s income (hence a discovery task).
Dataset Splits No The paper describes the construction of synthetic data instances (Table 3) and their grouping (A-H) for human subject studies, but it does not specify traditional training/test/validation dataset splits for machine learning model development or evaluation, as the core of their experiments involves human interaction with a simulated model rather than training a new model.
Hardware Specification No The paper does not mention any specific hardware used for conducting the experiments or running the synthetic model.
Software Dependencies No The paper does not specify any particular software libraries or versions used for implementing their methodology or conducting their experiments.
Experiment Setup No The paper describes the design of a human subject study, including how synthetic data was generated and presented, how human intuitions were measured, and the evaluation metrics. However, it does not provide hyperparameters or system-level training settings for a machine learning model, as the 'model' used in their experiment is synthetic and its behavior is dictated by the experimental design to test human interaction, not a trained ML model with tunable parameters.