The State of AI Detection in 2025

Thursday, October 30, 2025 | 9:45AM–10:30AM CT | Meeting Room 207 CD, Level 2
Session Type: Breakout Session
Delivery Format: Industry Insights
As artificial intelligence transforms academic writing, educators face an unprecedented challenge: distinguishing human-generated text from machine-generated text. With 86% of students regularly using ChatGPT and nearly half believing technology makes cheating easier, an educator's understanding of how to identify AI-written content has become essential for maintaining academic integrity. This talk reveals why AI writing is detectable by examining the hidden patterns that emerge from how these systems are built. Large language models don't simply average human writing. They develop distinct personalities shaped by their training data and the preferences of their creators. These biases manifest themselves in telltale signs: an overreliance on certain words like "delve" and "tapestry," perfectly structured but oddly generic paragraphs, flawless grammar paired with vague content, and most tellingly, an inability to reflect on personal experience or demonstrate genuine creativity. The presentation explores both human and automated detection methods, showing that while trained educators can identify AI writing with reasonable accuracy, combining human judgment with advanced detection tools yields the best results. Beyond detection, the talk addresses the broader implications for education: how to have productive conversations with students about appropriate AI use, design assignments that resist AI completion, and create fair processes that acknowledge both the benefits and risks of these technologies.

Presenters

  • Marilyn Derby

    Associate Director, Office of Student Support & Judicial Affairs, University of California, Davis

  • Bradley Emi

    CTO, Pangram Labs

Resources & Downloads

  • educause_ai_misconduct

    Updated on 6/16/2026