-
Princeton University
October 1, 2026
Location: Crick Conference
Evaluating Explanations from Humans and Machines
Until very recently, verbal and written explanations were a uniquely human form of communication—a way for one human mind to transmit understanding to another. In the last few years, however, we increasingly interact with explanations generated not by humans, but by artificial intelligence (AI) systems, such as large language models (LLMs). Though trained on explanations generated by humans, such systems produce explanations with distinctive features. In this talk I’ll present recent work on how explanation evaluation and inference are influenced by the characteristics of LLM explanations, including their content, linguistic properties, and perceived source (human or AI). This work reveals some tremendous opportunities: LLM-generated or rewritten explanations are judged better than human-generated explanations along a variety of dimensions. But it also reveals some risks: LLM-generated explanations can introduce cues to explanation accuracy and source reliability without actually improving explanatory content. Beyond these implications for living in an increasingly AI-mediated explanation environment, the contrast between human- and machine-generated explanations offers new tools for investigating the basic mechanisms that drive our evaluation and use of explanations.
-
Yale University
November 5, 2026
Location: Crick Conference
-
USC
December 3, 2026
Location: Crick Conference
-
Boston University
March 4, 2027
Location: Crick Conference
-
UC Davis
April 8, 2027
Location: Crick Conference
Anderson Colloquium Speaker Series
Thursdays, 11:00 am
Contact: lnngo@ucsd.edu
Past Speakers: Colloquium Archive
Upcoming Speakers