Why AI Detection Alone Shouldn't Be Used To Establish Academic Standing

Jonathan Gillham is the Founder and CEO of Originality.ai, a software specializing in AI Content and Plagiarism detection.

AI has become an everyday part of life. Yet, with continuous developments in AI and innovations in natural language processing (NLP), it’s difficult for humans to detect AI text on their own.

In the context of education, AI has become a frequent topic of debate. In a study shared by Oxford University Press, they noted that 69% of English language teachers in Europe took a hopeful but cautious approach to AI in education.

However, that same study also noted that 20% of teachers in the UK were concerned about what AI meant for education.

So, with teachers demonstrating a combination of optimism, concern, and caution about the role AI plays in education—where does AI detection fit into the conversation?

As an AI detection company, we believe that AI detection has a role in education—however, it shouldn’t be used as the only measure of academic standing or discipline.

False Positives Do Occur in AI Detection

To clarify, false positives are an important consideration when it comes to AI detection.

What Are False Positives?

In the context of AI detection, false positives happen if human-written text is identified as AI-generated.

To provide a benchmark for false positive rates, we can best speak to the rates at our company.

At Originality.ai, we clearly note that the false positive rate for our Lite model (which allows light AI editing) is 1% or lower and our Turbo model (which has a 0 tolerance for AI) is 3% or less.

However, false positive rates can vary by AI detection model and company.

So, to make an informed decision when selecting an AI detector, it’s important to check the particular company and AI detector model’s false positive rate.

How to Avoid False Positives

To avoid false positives, some of the best practices include:

Maintaining Clear AI Policies: Whether hiring a freelance writer or assigning a project to students, clearly identify whether AI is or is not permitted, and to what degree.

Minimize The Use Of AI Tools: When preparing content, use AI tools at a minimum, such as for light research (though make sure to fact check to avoid AI hallucinations, which are false information that AI may present as fact) or for editing to catch spelling and grammar errors.

Create Articles In A Platform Where Revision History Is Visible: If an AI detector incorrectly flags human-written content as AI, being able to demonstrate the complete creation process establishes transparency.

AI Detection In Education

AI is becoming deeply integrated into multiple aspects of society, and academia is no different. So, what does that mean for AI detection?

Should Educators Incorporate AI Detection?

A study available through Science Direct found that pre-service teachers only correctly spot AI text 45.1% of the time, while experienced teachers identify AI correctly 37.8% of the time.

As AI models continue to advance, an inability to identify AI text poses challenges, such as evaluating the authenticity of student work and measuring student comprehension.

If educators aren’t able to accurately identify AI text from authentic student work, it could lead to AI anxiety in the classroom. Further, it makes it difficult for educators to effectively assess student comprehension and whether they are grasping key concepts in the curriculum.

So, AI detection does have a place in establishing transparency in education—however, it shouldn’t be the only measure for determining educational standing.

How Does Academia Use AI Detection?

AI detection is present in education and academia, as noted by my company’s research into whether colleges are using AI detectors.

Some academic institutions directly state they may use tools to monitor the presence of AI in student work. However, it’s important to note that AI detection in education hasn’t yet been adopted across the board for evaluating assignments.

Where we do see a particularly prominent presence of AI detection in academia is research.

Several research studies have been conducted on AI detection. Many of these studies include research conducted by or in collaboration with authors affiliated with academic institutions.

For instance, the study RAID, which benchmarked machine-generated text detectors, credited authors associated with the University of Pennsylvania, University College London, King’s College London and Carnegie Mellon University.

Best Practices For Incorporating AI Detection In Education

The question now arises, if educators struggle to identify AI manually and AI is on the rise, how can AI detection be integrated into education?

The answer? A hybrid approach.

AI detectors are effective at detecting AI (although accuracy can vary by company and AI detection model), so they are an excellent choice for running an initial scan on educational material or assignments that may include AI-generated text.

Then, the results should be carefully reviewed by an educator before any further actions are taken.

While AI is creating complexities for the educational system, educators can:

• Compare the submitted work to previous assignments to review writing style.

• Cross-reference research as well as cited sources.

• Review document revision histories.

• Hold an open discussion with the entire class to clarify AI policies.

This brings a multi-faceted approach to identifying AI in student work that doesn’t just rely on an AI detection tool or a human to establish academic standing. Rather, it incorporates the innovations of AI detectors with educators’ human expertise.

Final Thoughts

AI detection does have a place in education—it just shouldn’t be used as the standalone decision for whether an assignment is or is not AI-generated, considering its impact on student educational standing.

Instead, it’s essential to balance human expertise and the unique experience of educators with AI detection tools, in order to make informed decisions about AI in an educational setting.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?