How to tell if text was generated by a neural network

Content:

• The text is significantly longer than the assignment requires
• The text does not convey real emotion
• The text makes inappropriate use of lists
• There are no typos, spelling, or punctuation errors
• The text differs in style from the student's other written work
• There just feels like something is "off"

Distinguishing text generated by ChatGPT or other similar tools from human-written text is quite difficult. A good knowledge of the topic can help identify the "hallucinations" of artificial intelligence—false statements it includes in the text when there is insufficient information. However, ChatGPT can appear quite convincing, and when evaluating written works such as essays, term papers, or theses, one cannot say with complete certainty that they were written by students independently. Therefore, it's important to pay close attention to the originality and quality of texts, especially in an educational environment where sincerity and a personalized approach are crucial.

Specialized tools, such as AI detectors, can be used to identify generated texts. One such solution is the Antiplagiarism service, as well as a similar tool developed by OpenAI. However, it's important to keep in mind that these technologies are not perfect. They may fail to recognize texts generated by artificial intelligence and may also mistakenly identify human-written works as generated. Therefore, caution and critical thinking should be exercised when using these tools.

Eric Ofgang, a journalist and creative writing instructor at Western Connecticut State University, encountered the problem of recognizing generated texts while reviewing student work. In an article for Tech & Learning, he highlights the signs by which texts generated by artificial intelligence can be distinguished from those written by humans. Ofgang also cautions educators against prematurely accusing students of using neural networks and suggests viewing such cases as an opportunity to discuss the ethical and appropriate use of AI technologies in the educational process.

Although the features described by Ofgang apply to English-language texts, they are also relevant for Russian-speaking audiences. Analyzing these features can help create higher-quality content that will be better received by readers. Applying these recommendations when writing texts in Russian improves their structure and increases readability.

The text is significantly longer than the assignment requires.

In every study group, you can find students who are especially diligent and always complete assignments at a high level without the help of neural networks. According to Eric Ofgang, such students often ask teachers for clarification, submit assignments earlier than their classmates, and write competently. However, if a student who has not previously been a perfectionist suddenly submits a long text, this may raise suspicions. Such a sharp jump in the quality of work can be alarming and become a signal of possible dishonest practices.

Generated A long text generated by a neural network can have many shortcomings. Most often, it presents vague discussions on a given topic without providing clear answers to the questions posed. Such texts can be overloaded with information, making it difficult to perceive and understand key ideas. To improve content quality, it is important that the text is structured, logical, and as relevant as possible to user queries. Effective text generation should consider not only the volume but also the quality of the information presented to ensure usefulness and meet the needs of the audience.

The text does not convey true emotions

The author of the article notes that text generated by a neural network is reminiscent of background music in an elevator. It may sound cheerful and positive, but it does not convey genuine emotions. For example, if you ask students to write about significant personal memories or a favorite hobby, the generated essay will seem empty and lifeless compared to work done by a human. Human creativity is capable of conveying depth of feeling and unique experience, which makes it more valuable and rich.

Lists are used inappropriately in the text

Generative neural networks, such as ChatGPT, often represent responses to queries as lists. However, this format is less common in human communication and writing. When describing complex ideas, using a list can be inappropriate. Even when a list format is appropriate, many students prefer not to structure information this way, according to Ofgang's observations. Therefore, excessive use of lists may indicate that the text was generated by artificial intelligence. This highlights the importance of naturalness and flexibility in writing, which helps avoid formulaic writing and makes texts more readable and informative.

No typos, spelling, or punctuation errors

Human-generated texts often contain typos, missing commas, and poor phrasing. In contrast, according to the author, ChatGPT generates more fluid and structured texts that, on average, have fewer errors than student work. This highlights the importance of using modern technology to improve writing quality and increase the accuracy of texts.

The text differs in style from other student writing

Eric Ofgang teaches creative writing and reads many student papers. In his observations, if the style of a particular text differs significantly from a student's previous work, this may indicate plagiarism or the use of AI to generate the text. It is important to note that this criterion should only be applied if the student truly has their own unique stylistic techniques with which the instructor is familiar.

Just a feeling that something is "off"

Instructors note that in some student work, several signs indicating the use of artificial intelligence can be immediately identified, raising doubts about the integrity of the assignment. In other cases, it is difficult to identify specific indicators of AI generation, but when reading the text, there is a clear sense that its author is not human.

In such situations, Eric Ofgang often seeks a second opinion from a fellow instructor. If his suspicions are confirmed, this becomes the basis for a discussion with students about the ethics of using neural networks in education. The author of the article emphasizes that this is a conversation, not an accusation. This is especially important, as proving that a text was generated by artificial intelligence is virtually impossible. Discussing such topics helps students develop a responsible attitude toward technology and its use in the educational process.

In one of his posts, a teacher shared a case where his suspicions were confirmed not only by his own observations but also by expert opinions and the results of AI detectors. Before grading a student's work, he sent him an email inviting him to discuss the text and answer a few questions. At the same time, the teacher sent all other students reminders that the use of neural networks was prohibited by course rules. The student who submitted the suspicious text ignored this message and, as a result, did not receive points for his work. This case highlights the importance of honesty in the educational process and the consequences of violating academic integrity.

Learn more about education in our Telegram channel. Subscribe to updates and receive useful information about news, trends, and useful resources in education. Don't miss the opportunity to stay up-to-date on current events and useful tips.

Reading is an important part of our lives, contributing to the development of thinking and broadening horizons. It helps us not only acquire new information but also develop emotional intelligence. In today's information-saturated world, it's important to be able to select quality sources. Read books, articles, and research that enrich your knowledge and help you better understand the world around you. Don't forget the importance of reader recommendations and reviews, which can direct you to interesting and useful materials. Maintain a constant interest in reading, and you will notice how your perception of information and critical thinking will significantly improve.

• How the ChatGPT neural network turned the humanities education upside down
• Principles for assessing student work in the AI ​​era have been developed
• How to use ChatGPT and its analogues in education: a large overview from experts
• What AI assistants have learned to do in education