Could someone falsely accuse you of using AI-generated text? This might be the reason

artificial intelligence (AI) has grown in popularity in recent months, but remains controversial in some areas, such as education. Much of this concern focuses on student availability for school. Generation AI tools etc. Chat GPTto do their job, writing an essay again Create code.

Some professors allow the use of technology in the classroom, but others. forbid it, and other students, at their sole discretion, including using the GPT detector to scrutinize all student work. recently published, Peer-reviewed papers According to Patterns, a program built to detect whether text is AI-generated or human-generated is AI-generated if it’s written by a non-native English-speaking writer. It has been shown that researchers have found that it is more likely to be mislabeled with.

again: This new AI tool transforms your doodles into high-quality images

In the study, researchers used 91 essays written by Chinese speakers for the Test of English as a Foreign Language (TOEFL) and 88 essays written by eighth graders in the United States, using the widely used We tested the performance of seven GPT detectors. Hewlett Foundation Automatic Student Assessment Award (ASAP).

The GPT detector correctly classified all US student essays, but an average of 61% of TOEFL essays were incorrectly labeled as AI-generated. One of our detectors incorrectly flagged 97.8% of TOEFL essays as AI-generated.

again: These are my 4 favorite AI chatbot apps for Android

Research also found that these GPT detectors were not as effective in detecting plagiarism as users thought. Many detectors advertise 99% accuracy without any evidence to support their claims.

The researchers used ChatGPT to create essays and 70% were detected as being AI-generated by the GPT detector. However, a simple prompt, such as asking ChatGPT to “use written language to improve the provided text,” improved the text enough to reduce that figure to 3% of his. This means that the GPT detector incorrectly determined that 97% of the essays were written by humans. time.

“Our current recommendation is that we should be very careful in using these detectors, and perhaps try to avoid their use whenever possible,” says lead author James of Stanford University. Said the elephant.

again: Real-Time Deepfake Detection: How Intel Labs Uses AI to Combat Disinformation

The authors attributed the error to the GPT detector prioritizing complex language and penalizing simple word choices commonly used by non-native English writers. They found low text complexity in the TOEFL essays, which “wowed” the AI ​​model. If the next word in the essay is difficult for her GPT detector to predict, it is more likely to assume that a human wrote the text. In the opposite case, it is considered to be created by AI.

“Using common English words causes the detector to give a lower confusion score, which means my essay is more likely to be flagged as AI-generated. Using ugly words makes the essay more likely to be classified as written by a human,” Zou explained.

again: This is how generative AI changes the gig economy for the better

Detecting AI-generated content in general may be difficultAs such, detection methods in the form of third-party computer programs are prevalent. However, research suggests that these tools may marginalize non-native English-speaking writers in assessment and teaching settings.

“If these detectors are used to screen job applications, college admissions essays, high school assignments, etc., it could have serious consequences,” Zou explained.

Paradoxically, this study points out that GPT detectors can guide non-native English speakers to English. Use more generative AI tools It’s an effort to avoid detection and improve your language skills. By doing so, you can avoid the harassment and restricted visibility that can result from being discriminated against.