Co-authored comprehensive research analysis examining a novel behavioral approach to AI detection that focuses on user input patterns rather than content analysis. The work addresses critical challenges in maintaining data quality for AI training by developing methods to identify when crowdsourced contributors use AI tools during content generation tasks. This research provides practical solutions for organizations struggling with AI contamination in human-labeled training data.
Key Features
• Behavioral Pattern Analysis: Developed methodology to analyze keyboard and mouse movement patterns to distinguish between human-only and AI-assisted content creation • Research Publication: Co-authored peer-reviewed study "Clicks Don't Lie: Inferring Generative AI Usage in Crowdsourcing through User Input Events" presented at HCOMP 2023 • Industry Problem Solving: Addressed critical data quality challenges facing AI practitioners in crowdsourcing and model training workflows • Alternative Detection Approach: Pioneered user behavior analysis as an alternative to traditional content-based AI detection methods like watermarking and style analysis
Technical Implementation
The research demonstrates advanced understanding of human-computer interaction patterns and their application to AI detection challenges. The behavioral approach leverages user input event analysis to create reliable detection mechanisms that work independently of content quality or sophistication. This methodology provides a robust framework for maintaining data integrity in crowdsourced AI training environments where traditional detection methods may fail.