AI-generated content detection tools have been put to the test as the popularity of OpenAI’s Generative Pre-Trained Transformer (ChatGPT) continues to soar. While the AI chatbot has gained millions of users within days of its launch, its ability to produce responses that closely resemble those of humans has raised concerns in industries such as education, content marketing, publishing, journalism, and law, where human-generated content is highly valued. The challenge for these industries is to distinguish between human and AI-generated text and to detect AI-generated content.
ChatGPT utilizes a deep learning algorithm called a transformer, which is highly effective for natural language processing tasks. The model has been trained on vast amounts of text data from the internet, enabling it to learn patterns and relationships between words and phrases. ChatGPT uses unsupervised learning, which means that it does not require explicit instructions or labels to generate responses. Its ability to produce coherent and realistic responses to complex prompts has made it a valuable tool for various applications, including chatbots and virtual assistants.
However, detecting AI-generated content is still challenging. One approach to detect AI-generated content is to examine five key characteristics of the text: consistency, coherence, originality, errors, and context. AI-generated text is typically consistent in style, tone, and vocabulary, while human-written text may exhibit more variation and nuances. AI-generated text may sometimes lack coherence, especially when responding to complex or nuanced prompts. It may also contain repetitive or formulaic phrases or patterns, while human-written text is more likely to be original and creative. AI-generated content is more prone to errors than human-written text, particularly in areas where the model has not been extensively trained. Finally, the platform may sometimes struggle to understand the context of a given prompt, leading to inappropriate or irrelevant responses, whereas human-written text is typically tailored to a specific context and audience.
This article discusses the emergence of automated tools that can identify text written by artificial intelligence (AI) software, and examines some of the top AI-content detection tools available. OpenAI recently shared a link to their new classifier tool that can differentiate between text created by humans and that generated by various AI systems. However, the tool is not entirely reliable and may falsely identify human-written text as likely generated by AI tools. It is recommended for use only with English text and should be used in conjunction with other methods to determine the source of a piece of text. Content at Scale launched the “AI Detector” claiming to detect robotic sounding content and position itself as a first step towards buying their flagship content generator that claims to produce “undetectable” AI-generated content. However, the results of the AI Detector were unsatisfactory and it pointed out several instances that could be AI-generated, which in this case was not accurate.
In the second experiment, the AI detector was given an article authored by ChatGPT, but it did not perform well. Despite the article being entirely generated by AI, the platform gave it an 83% human content score. To test the platform’s performance further, the article was paraphrased using another AI tool (QuillBot), but the results remained unchanged. On the bright side, the human content score dropped to 75%, suggesting AI intervention.
Copyleaks AI is an anti-plagiarism software company that has recently launched an enterprise solution capable of detecting whether digital content was created by a human or generated by AI, including ChatGPT. The platform boasts an accuracy rate of 99.12%, along with enterprise-level LMS and API integration features that enable educational institutions or businesses to integrate the AI Content Detector into their native platforms. Multi-language detection is another critical component of the platform, with support for English, German, Spanish, French, and Portuguese. Additionally, the company provides an AI Content Detector Chrome extension that allows users to verify content across the internet, including social media, news articles, and consumer reviews.
Among the test candidates, Copyleaks AI demonstrated the highest accuracy. It accurately identified human-written text in the human content test. Similarly, the platform identified text generated by ChatGPT as AI-generated content with a 99.7% probability. Even in the last test that included a paraphrase of AI-generated text, the platform indicated that the content had a 99.9% chance of being written by AI.
As technology progresses, AI-content assisted production will inevitably become more prevalent, and with it, AI content detectors will improve.