Tips & Resources

Tools for Detecting AI-Assisted Writing and Research

Al-Fanar Media has been looking at some of the smart tools university faculty members can use for detecting AI-assisted writing and research.

Since the U.S. company OpenAI released ChatGPT last November, an increasing number of students are using artificial intelligence-powered text-generating tools to write or assist them in writing research papers and other university assignments, or even to do their exams.

While some academics argue for embracing the use of AI in the classroom, others are concerned that many students will misuse AI chatbots to cheat.

This poses a major challenge for academics who want to make sure students are doing their own work and not turning in assignments written by a robot. They want tools to help them detect AI-generated content. At the same time, they want to avoid falsely accusing students of cheating.

Humans vs. Robots

Insights into how human writing differs from what a machine can produce might aid in the detection of AI-generated content.

The growing use of AI chatbots poses a major challenge for academics who want to make sure students are doing their own work and not turning in assignments written by a robot.

So far human beings write in a distinctive way which artificial intelligence has not yet mastered. Another distinctive feature of humans’ writing is in the text structure, dividing text into paragraphs related to a single concept. AI, on the other hand, tends to divide text into paragraphs by number of words rather than meaning.

Another way in which AI-generated texts might be revealed is the presence of phrases that indicate that they are an intelligence system’s answers to questions a person has asked about a specific topic: phrases like “Here are the steps”, for example. Students who relied on AI to create their text may overlook these, so finding one is a strong hint that the writing was machine-generated.

Detection Tools

There are a variety of smart tools on the market that allow academics to scan research papers and group assignments for signs of AI use, as well as check the sources of quotations cited in students’ papers.

Such tools, however, are themselves controversial. Some of the earliest ones proved unreliable and were withdrawn by their developers or disabled by universities that were using them. Studies have also found evidence that the detection tools are biased against non-native English writers, which could put foreign students’ essays in applications for scholarships or admission to Western universities at a disadvantage.

Moreover, some students may try to outsmart AI-detection tools by modifying the prompts they use to generate a text, or paraphrasing a machine-composed text to make it sound more like human writing.

Software companies say they are constantly refining their products to overcome these problems and make the tools more effective in detecting AI-assisted writing and research.

In the end, professors must consider AI-detection tools merely as aids to flag potential violations in academic papers. The final judgment must depend on their own vision, based on their experience, and follow-up of students’ performance and commitment.

With those caveats in mind, here is a short list of some of the AI-detection tools professors are using now.

Originality.AI

Originality.AI is an AI-detection tool developed by a software company based in Ontario, Canada. It tries to ascertain how human the writing style is and whether content has been used without quoting the original sources. Originality.AI claims to be able to recognise content created by all modern AI text-generation tools with 94 percent or greater accuracy.

In the end, professors must consider AI-detection tools merely as aids to flag potential violations in academic papers. The final judgment must depend on their own vision.

Originality.AI offers a free trial period, after which it charges by the number of words scanned at a rate of $0.01 (U.S.) per 100 words. Users can pay as they go, starting with an initial purchase of $30, or subscribe for $14.95 a month.

Winston AI 

Winston AI, another system from a Canadian-based firm, is dedicated to finding AI-generated content and staying ahead of artificial intelligence language models.

It provides academics and teachers with a set of features such as optical character recognition technology to review research papers and spot signs that students have used generative AI. It also identifies quotations that student papers have cited without crediting their sources.

Winston AI provides a seven-day free trial capped at 2,000 words without the plagiarism detection feature. Subscriptions cost $12 (U.S.) a month for scanning up to 80,000 words and $19 a month for up to 200,000 words.

Passed.AI 

Passed.AI, also from a company based in Ontario, works through a dedicated add-on for the Chrome browser that allows teachers to audit how a research paper written as a Google Doc was created and gauge the probability that it was generated by artificial intelligence. On group projects, teachers can also see how much each student contributed to the text.

Passed.AI uses Originality.AI’s AI detector, which is trained on different artificial intelligence models, such as OpenAI’s GPT-4 and open-source alternatives like GPT-J and GPT-Neo.

Passed.AI also claims to have secondary tests to avoid false positives and false negatives.

Passed.AI offers a five-day free trial and its standard plan costs $9.99 (U.S.) a month.

GPTZero 

GPTZero, first released in January by Edward Tian, then a 22-year-old computer science major at Princeton University, instantly became one of the most popular AI-detection tools among academics. Tian now heads GPTZero as a full-fledged startup with a team of machine-learning experts who are striving to keep up with developments in artificial intelligence and improve the accuracy of the tool.

GPTZero calls itself “the gold standard” in AI recognition, saying its tool is trained to detect text created by ChatGPT, GPT-4, Bard, and other AI models. It looks for randomness in a text and changes of enthusiasm in the wording. It shades questionable parts for professors and teachers to evaluate.

GPTZero offers some free services. You can check short texts on its website to see how likely it is that they were written by AI. You can also check for AI-generated content through its Origin Chrome extension. An Educator Plan that costs $9.99 (U.S.) a month allows teachers to upload unlimited files for AI detection.

ScholarshipsFacebookNewsletter

Countries

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button