Skip to main content
Apply Give
Menu

LSLT - Jordan Boyd-Graber / If we want AI to be interpretable, we need to measure interpretability

A man in a dark green v-neck pullover, posed sternly in front of a whiteboard covered in diagrams that describe computational models of language processing.

LSLT - Jordan Boyd-Graber / If we want AI to be interpretable, we need to measure interpretability

Linguistics Thursday, September 14, 2023 12:30 pm - 1:30 pm H.J. Patterson Hall

September 14, LSLT has Jordan Boyd-Graber (CS/UMIACS/LSC), who explains why "If We Want AI to be Interpretable, We Need to Measure Interpretability."

Abstract

AI tools are ubiquitous, but most users treat it as a black box: a handy tool that suggests purchases, flags spam, or autocompletes text. While researchers have presented explanations for making AI less of a black box, a lack of metrics make it hard to optimize explicitly for interpretability. Thus, I propose two metrics for interpretability suitable for unsupervised and supervised AI methods.

For unsupervised topic models, I discuss our proposed "intruder" interpretability metric, how it contradicts the previous evaluation metric for topic models (perplexity), and discuss its uptake in the community over the last decade. For supervised question answering approaches, I show how human-computer cooperation can be measured and directly optimized by a multi-armed bandit approach to learn what kinds of explanations help specific users.

Add to Calendar 09/14/23 12:30:00 09/14/23 13:30:00 America/New_York LSLT - Jordan Boyd-Graber / If we want AI to be interpretable, we need to measure interpretability

September 14, LSLT has Jordan Boyd-Graber (CS/UMIACS/LSC), who explains why "If We Want AI to be Interpretable, We Need to Measure Interpretability."

Abstract

AI tools are ubiquitous, but most users treat it as a black box: a handy tool that suggests purchases, flags spam, or autocompletes text. While researchers have presented explanations for making AI less of a black box, a lack of metrics make it hard to optimize explicitly for interpretability. Thus, I propose two metrics for interpretability suitable for unsupervised and supervised AI methods.

For unsupervised topic models, I discuss our proposed "intruder" interpretability metric, how it contradicts the previous evaluation metric for topic models (perplexity), and discuss its uptake in the community over the last decade. For supervised question answering approaches, I show how human-computer cooperation can be measured and directly optimized by a multi-armed bandit approach to learn what kinds of explanations help specific users.

 H.J. Patterson Hall false