ChatGPT: How Can You Tell If a Text Was Written by AI?

How can I tell whether a text was written by an AI like ChatGPT?

OpenAI's AI Text Classifier

Following the success and widespread use of the text AI ChatGPT, which can handle a wide range of tasks in writing, research and translation, OpenAI has developed another AI designed to identify text of this kind.

The AI Text Classifier is based on supervised learning. This means it is trained on sample data that has already been sorted into specific categories. The algorithm uses this data as a reference and learns from it to classify texts on its own. The more sample data is available, the more precise and reliable the AI Text Classifier becomes.

However, if this data is incomplete or biased, the AI Classifier's decisions will also be wrong or inaccurate.

🤖 Try OpenAI's AI Text Classifier here:
➤ https://platform.openai.com/ai-text-classifier.

Current limitations

OpenAI's AI Text Classifier needs a minimum of 1,000 characters (roughly 150-250 words) and detects AI-written text with an accuracy of 26%. A common source of error is analysing text in any language other than English, as the AI Text Classifier was trained mainly in English. In addition, the AI produces false positives in 9% of cases, meaning human-written texts that are identified as AI-written. This often happens with texts written by children, as they tend to use simpler sentence structures and can therefore trigger false alarms in an AI trained on the work of adults.

Although OpenAI's AI Text Classifier has already made considerable progress, the company continues to work on improving its capabilities and overcoming these limitations. Here are some of the ongoing improvements:

1. Expanding the training data: OpenAI is working to broaden the data the AI Text Classifier is trained on. In particular, the company aims to collect more texts in different languages to increase the model's accuracy and its ability to classify texts in a wide range of languages.

2. Integrating new technologies: the company is looking at various technologies to further improve the accuracy of the AI Text Classifier. For example, technologies such as transfer learning or active learning could be used to make the AI Text Classifier more effective at identifying AI-generated text.

3.  Human review: another approach OpenAI is pursuing is integrating human review into the classification process. The company plans to use human reviewers to identify texts that the AI Text Classifier has misclassified. This is intended to improve the model's accuracy.

Overall, OpenAI is committed to further improving the accuracy and effectiveness of its AI Text Classifier. The ongoing improvements described here are just a few examples of how the company intends to reach these goals. It remains to be seen what progress will be made in the future.