A multi-institutional team of scientists led by Jack Hessel at the Allen Institute for AI tested artificial intelligence (AI) models' ability to understand humor.
The researchers tasked computer vision and description-led models to match jokes to 14 years' worth of cartoons from The New Yorker's Cartoon Caption contest by identifying winning captions and explaining why those captions are funny.
Humans outperformed the AIs across the board, with the best-performing model achieving only 62% match accuracy in multiple-choice questions compared to humans' 94%.
The preference for human-generated explanations of the jokes also surpassed that for AI-generated explanations by approximately two to one.
From Cornell Chronicle
View Full Article
Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA
No entries found