Preface

This reading list contains some materials for a seminar. The audience include graduate students from CompSci and other disciplines.

Prerequisite: An NLP lecture would be nice to have.

Psychology of reasoning and explanations

• R. C. Cummins (2000) "How does it work" versus "what are the laws?": Two conceptions of psychological explanation. https://philpapers.org/rec/CUMHDI
• Lombrozo, Holyoak, Morrison (2012) Explanation and abductive inference. https://cognition.princeton.edu/publications/explanation-and-abductive-inference
  • This is a chapter of the Oxford Handbook of Thinking and Reasoning, reviewing evidence concerning the structure and function of explanations.
• Keil (2011) Explanation and understanding. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3034737/
• Chapter 1 of the XAI-HCEE tutorial: https://xai-hcee.github.io/chapter_1.html
• Miller (2019) Explanation in artificial intelligence: Insights from the social sciences https://doi.org/10.1016/j.artint.2018.07.007
  • Many explanations are contrastive. They are selected. Explanations are social.

Quality of explanations

• Lombrozo (2016) Explanatory preferences shape learning and inference https://doi.org/10.1016/j.tics.2016.08.001
  • Simplicity and breadth.
• Vrantsidis and Lombrozo (2022) Simplicity as a cue to probability: multiple roles for simplicity in evaluating explanations. https://cognition.princeton.edu/publications/simplicity-cue-probability-multiple-roles-simplicity-evaluating-explanations
• Vasil, Blanchard, Lombrozo (2018) Stable causal relationships are better causal relationships. https://cognition.princeton.edu/publications/stable-causal-relationships-are-better-causal-relationships
• ERASER (2020) (sufficiency and comprehensiveness): https://aclanthology.org/2020.acl-main.408.pdf
  • Fidelity curve (sufficiency vs occlusion, comprehensiveness vs occlusion) https://arxiv.org/pdf/2010.04736.pdf
• Chen et al (2021) What makes a good explanation? A harmonized view of properties of explanations https://arxiv.org/abs/2211.05667
  • Sensitivity, faithfulness, complexity, homogeneity.
• Liao et al., (2022) Connecting algorithmic research and usage contexts: a perspective of contextualized evaluation for explainable AI https://arxiv.org/abs/2206.10847