All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Text

TL;DR

This study assesses how well non-expert human evaluators can distinguish between human and AI-generated text across different domains, revealing limitations and proposing improvements for evaluation methods.

Contribution

It demonstrates that untrained evaluators perform at chance level in identifying AI text and explores training methods to improve their accuracy, highlighting the need for better evaluation practices.

Findings

Evaluators' accuracy improved up to 55% with training.

Untrained evaluators perform at chance level.

Training methods showed inconsistent improvements across domains.

Abstract

Human evaluations are typically considered the gold standard in natural language generation, but as models' fluency improves, how well can evaluators detect and judge machine-generated text? We run a study assessing non-experts' ability to distinguish between human- and machine-authored text (GPT2 and GPT3) in three domains (stories, news articles, and recipes). We find that, without training, evaluators distinguished between GPT3- and human-authored text at random chance level. We explore three approaches for quickly training evaluators to better identify GPT3-authored text (detailed instructions, annotated examples, and paired examples) and find that while evaluators' accuracy improved up to 55%, it did not significantly improve across the three domains. Given the inconsistent results across text domains and the often contradictory reasons evaluators gave for their judgments, we examine the role untrained human evaluations play in NLG evaluation and provide recommendations to NLG researchers for improving human evaluations of text generated from state-of-the-art models.