GENAI DETECTION TOOLS, ADVERSARIAL TECHNIQUES AND IMPLICATIONS FOR INCLUSIVITY IN HIGHER EDUCATION

This paper “GenAI Detection Tools, Adversarial Techniques and Implications for Inclusivity in Higher Education” reveals that GenAI detection tools have significant limitations and are not reliable in detecting machine-generated content. The low accuracy rates and the potential for false accusations raise concerns about fairness, inclusivity, and the negative impact on certain groups such as non-native English speakers. The paper emphasizes the importance of considering the unintended consequences of these tools and the need for a cautious approach in their implementation.

Here is a list of key takeaways from the paper:

1. The study investigates the effectiveness of Generative AI (GenAI) text detectors when faced with machine-generated content that has been modified to evade detection.
2. The accuracy rates of the detectors are already low (39.5%), and they show significant reductions in accuracy (17.4%) when confronted with manipulated content.
3. The limitations in accuracy and the potential for false accusations demonstrate that these tools are not currently recommended for determining violations of academic integrity.
4. Inclusive and fair assessment practices in higher education face challenges in maintaining academic integrity due to the limitations of GenAI detection tools.
5. GenAI text detectors may have a role in supporting student learning and maintaining academic integrity when used in a non-punitive manner.
6. The study highlights the need for a combined approach to address the challenges posed by GenAI in academia, promoting responsible and equitable use of these technologies.
7. The current limitations of AI text detectors require a critical approach before implementing them in higher education, and alternative assessment strategies should be considered.
8. GenAI tools have the potential to disadvantage certain groups of students, such as non-native English speakers, due to linguistic biases and the digital divide.
9. Non-native English speakers may face false accusations by GenAI text detectors due to the higher level of perplexity and lower coherence often found in their writing.
10. Adversarial techniques designed to evade detection pose further challenges to the efficacy and reliability of AI text detectors.
11. The objective of the research is to assess the ability of GenAI tools and AI text detectors to foster inclusivity in education and provide equal opportunities for all students.
12. The study aims to measure the susceptibility of existing AI text detectors to adversarial techniques and linguistic changes to draw conclusions about their suitability for higher education.

Source:2403.19148.pdf (arxiv.org)