Comparing ai detectors: evaluating performance and efficiency

Abstract

The widespread utilization of AI tools such as ChatGPT has become increasingly prevalent among learners, posing a threat to academic integrity. This study seeks to evaluate capability and efficiency of AI detection tools in distinguishing between human-authored and AI-generated works. Three-paragraph works on “AutoCAD and Architecture” were generated through ChatGPT, and three human-written works were subjected to evaluation. AI detection tools such as GPTZero, Copyleaks and Writer AI were used to evaluate these paragraphs. Parameters such as “Human/Human Text/Human Generated Text” and “AI/AI Content Detected” were used to evaluate the performance of the three AI detection tools in evaluating outputs. Findings indicate that GPT Zero and Copyleaks have higher reliability in determining human-authored work and AI generated work while Writer AI showed a notable content classification of “Human Generated Content” on all tested outputs showing less sensitivity on determining human-authored work and AI generated work. Findings indicate that the use of Artificial Intelligence as an AI detection tool should be accompanied with thorough validation and cross-referencing of results.