GenAI Detection Tools, Adversarial Techniques and Implications for Inclusivity in Higher Education
Project Overview
The document explores the role of generative AI (GenAI) in higher education, emphasizing its potential benefits and challenges. It highlights the effectiveness of GenAI text detectors, revealing significant limitations in accuracy and a propensity to disadvantage non-native English speakers, which raises concerns about equity and inclusivity in educational assessments. While GenAI tools can enhance learning experiences, their misuse poses risks that educators must navigate carefully. The study underscores the necessity for a critical approach to the implementation of GenAI technologies, advocating for strategies that ensure fair assessments and minimize biases. Ultimately, the findings call for a balanced perspective on the integration of GenAI in education, recognizing both its innovative applications and the imperative to address its shortcomings to foster an equitable learning environment.
Key Applications
GenAI text detectors
Context: Higher education, targeting educators and students
Implementation: Evaluated six major GenAI text detectors against manipulated content to determine their efficacy.
Outcomes: Detected a significant drop in accuracy (from 39.5% to 17.4%) when faced with adversarially manipulated AI-generated texts.
Challenges: High false positive rates, especially for non-native English speakers, raising concerns about inclusivity in assessment.
Implementation Barriers
Technological
GenAI text detectors display low accuracy rates and high false positive rates, particularly for non-native English speakers.
Proposed Solutions: A combined approach to assessment strategies that incorporates understanding of GenAI usage, with non-punitive guidelines for students.
Access
Financial barriers and digital poverty limit access to GenAI tools for some students, exacerbating educational inequities.
Proposed Solutions: Developing institutional policies that provide equitable access to GenAI technologies and resources.
Project Team
Mike Perkins
Researcher
Jasper Roe
Researcher
Binh H. Vu
Researcher
Darius Postma
Researcher
Don Hickerson
Researcher
James McGaughran
Researcher
Huy Q. Khuat
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Mike Perkins, Jasper Roe, Binh H. Vu, Darius Postma, Don Hickerson, James McGaughran, Huy Q. Khuat
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai