Empirical studies from 2023 show 22% of faculty and 49% of students were using GenAI in higher education (Link). Similar trends have been seen in 2024 with 41% of students reporting AI use for fun and 35% reporting using it for coursework (Link). Surveys show that the most common ways students are using AI in the classroom include: creating written content, making artwork, solving mathematical calculations, translating to a new language, and generating new programming codes (Link).
This proliferation has created concerns among educators regarding potential breaches of academic integrity and the potential attenuation of critical skill development among students (Link). If AI Detectors, like Turnitin, are to be believed upwards of 1 out of 10 assignments turned in the last year contained AI written work (Link). Despite this unfortunate statistic, research shows that faculty who set parameters on ethical AI use and hold students accountable are less likely to see the misuse of AI in their classrooms (Link). See below for additional information on AI Detectors. For more information on the ethical use of AI see this page and for creating syllabus statements outlining AI use parameters see this resource.
AI detectors in higher education have become popular as institutions grapple with the rise of AI-generated content in student work. These tools aim to distinguish between human-written and AI-generated text, but their efficacy and reliability remain subjects of ongoing debate and research.
AI detectors typically use machine learning algorithms trained on large human-written and AI-generated text datasets. They analyze various linguistic features, such as sentence structure, vocabulary usage, and stylistic patterns, to make predictions about the origin of a given piece of text. Most AI detectors employ natural language processing (NLP) techniques to analyze text and may look at factors such as:
Perplexity: How predictable the text is to the model
Burstiness: The variation in complexity across sentences
Entropy: The randomness or unpredictability of word choices
Syntactic patterns: The structure and arrangement of sentences
Semantic coherence: The logical flow and consistency of ideas
Some popular AI detectors used in higher education include GPTZero, Turnitin's AI writing detection feature and Copyleaks AI Content Detector.
Chaka (2024), Comparing 30 AI Detectors
Walters (2023), Comparison of 16 AI Detectors
Yeadon et al. (2024), AI and Human Authorship Evaluating
Kumar (2024), Who Wrote This?
Pugalia et al. (2024), AI Detectors and Harm
Given these limitations, the use of AI detectors in higher education raises several concerns:
Inaccuracy and Trust Issues: The high rate of false positives can unjustly penalize students, while false negatives allow AI-generated work to go undetected, compromising academic integrity.
Student Distrust: Over-reliance on flawed AI detectors can erode trust between students and faculty, as students may feel unfairly accused or wrongly judged.
Need for Complementary Methods: Relying solely on AI detectors is insufficient. Combining these tools with traditional plagiarism detection methods and manual review processes is essential to ensure fairness and accuracy. For more information on using Academic Judgement and AI Detectors see this article by Perkins et al. (2023).
AI detectors in their current state are not reliably effective for identifying AI-generated content in higher education, therefore the BCTE and DLA do not recommend their use. Significant limitations in accuracy, high rates of false positives and negatives, and the rapid evolution of AI technologies challenge their efficacy. Educators must recognize these shortcomings and adopt a balanced approach, integrating multiple methods to maintain academic integrity and foster an equitable educational environment.
Identifying AI Generated Work
The proliferation of AI tools in academic settings has prompted faculty to seek methods for identifying AI-generated content in student assignments. While AI detection software exists, educators can also rely on certain characteristics commonly present in AI-produced work. These may include inconsistent writing style, generic or repetitive language, lack of nuanced argumentation, and impersonal tone. Additionally, AI-generated content often exhibits limited engagement with course-specific materials or recent events. However, it is crucial to recognize that these indicators are not foolproof. As AI technology rapidly evolves, its output becomes increasingly sophisticated and harder to distinguish from human-authored work (link). Faculty should approach the identification process with caution, using these characteristics as potential red flags rather than definitive proof of AI use. See below articles on specific trends associated with AI-generated work:
Steer (2024), Anatomy of an AI Essay
Wunsch (2024), How to Identify AI-Written Content & 20+ Red Flag Warnings
Barron (2023), Detecting AI In Student Work: Tips For Educators
Disclaimer: The content within this compendium was co-created using AI programs ChatGTP and Claude Sonnet. For more information on the co-construction of knowledge using AI, please see this resource by Robertson et al. 2024 and the AI uses in Education Page.