Explanations in the wild.

By Justin Sulik, Jeroen van Paridon, Gary Lupyan. 2021

16 citations recorded by the source corpus.

Why do some explanations strike people as highly satisfying while others, seemingly equally accurate, satisfy them less? We asked laypeople to generate and rate thousands of open-ended explanations in response to 'Why?' questions spanning multiple domains, and analyzed the properties of these explanations to discover (1) what kinds of features are associated with greater explanation quality; (2) whether people can tell how good their explanations are; and (3) which cognitive traits predict the ability to generate good explanations. Our results support a pluralistic view of explanation, where satisfaction is best predicted by either functional or mechanistic content. Respondents were better able to judge how accurate their explanations were than how satisfying they were to others. Insight problem solving ability was the cognitive ability most strongly associated with the generation of satisfying explanations.

Subfields: Artificial Intelligence, Cognitive Neuroscience, General Decision Sciences, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Cognitive Neuroscience

In Safety, Robustness & Interpretability

Citations in the corpus

Reframing Human-AI Collaboration for Generating Free-Text Explanations (2021, methodology): “…truthfulness be ascertained or is more information needed?) and factuality, which aligns with “compatibility with receiver’s existing be-liefs” and that the best explanation is the “most explanation from the receiver/user’s perspective (Lombrozo, 2007; Zemla et al., 2017; Sulik et al., 2021).”