As students and professionals from various professions ask the well-known OpenAI ChatGPT chatbot for assistance, researchers have developed a method to identify AI-generated theses and writing draughts.
When ChatGPT was recently made available to the general public, it created quite a stir thanks to its capacity to produce text and dialogues that resemble human speech.
However, a study that was released on Wednesday in the scientific publication Cell Reports Physical Science claims that there are a number of telltale signs that can assist us in differentiating between AI chatbots and actual people.
The researchers exploited these factors to develop a system that, more often than not (almost 99% of the time), can recognise academic science papers written by AI.
As text is collected from numerous sources without any kind of quality check. At the moment, writing with AI is comparable to playing the game Two Truths and a Lie and has several rather clear limitations.
Despite the fact that there are numerous trustworthy AI text detectors online. They were not all created with educational content in mind. To achieve the same goal, the team set out to develop a strong tool for narrowing the gap.
The research placed a lot of focus on perspective articles. Which, as their name suggests, are often authored by scientists and provide an in-depth description of a specific scientific topic.
In order to train the model, the group chose 64 points of view. And produced 128 articles using ChatGPT that addressed the same study themes. They found predictability in the writing when they compared the pieces, which is a sign of AI writing.
Human writing has longer sentences, more words per phrase, and more intricate paragraph patterns. Despite ChatGPT’s strong similarity to human writing, it has higher word counts than AI.
These patterns are also influenced by word choice and punctuation preferences. Using a 20-feature system, researchers were able to differentiate between people and AI-generated articles with a 92% success rate. The research team thus developed a superior AI text detector in comparison to a commercial one.
They also intend to assess its suitability for bigger datasets and more formal scientific writing. With a robustness focus as AI chatbots get more sophisticated. The model makes a distinction between artificial intelligence (AI) and humans. But the research team noted that others may readily copy and modify their approach.
Visit CxO Global FORUM or CxO News Live for all the latest updates.
