Scientific abstracts written by ChatGPT are misleading researchers

THEM ChatGPT It is a natural language processing tool with artificial intelligence (AI) and has gained a lot of popularity in recent weeks, especially for its ability to generate text similar to content written by humans. Now a study shows that the resemblance is so perfect that even scientists can’t detect fake texts.

In a preprint published in bioRxiv (the article has not yet been peer-reviewed), the researchers stated, artificial intelligence chatbot can write fake summaries of scientific papers that can persuade scientists, as many people cannot determine that it was written by an artificial intelligence. Professors from North American universities have also complained about the situation, reporting that their students are doing university studies using this tool.

The scientists asked the chatbot to write a summary of 50 medical articles based on real articles published in different scientific journals. Next, they scrutinized the content with plagiarism detectors, the use of artificial intelligence, and reading by medical researchers. None were detected as plagiarism, and 66% were marked as using artificial intelligence, but the researchers were able to detect only 68% of false abstracts — 32% of which were mistakenly identified as genuine.

Scientists tricked by artificial intelligence

Using a human-learning language model based on neural networks, ChatGPT can write believable texts on a variety of topics. Especially if the user asks the chatbot more critical questions, but in most cases the tool has a few flaws – for example, it can convincingly write content with a lot of non-existent information.

“It is unlikely that any serious scientist would use ChatGPT to generate a summary. The question is whether the tool can produce an accurate and convincing summary. Princeton University computer scientist Arvind Narayanan points out that ChatGPT is not useful to a scientist, and it cannot, and so the advantage of using ChatGPT is very small and the disadvantage is significant.

The ethical issue has been an issue since the inception of ChatGPT, which was released for free by OpenAI in November 2022. As AI manages to write text similar to human-generated text, scientists and education experts fear the problem will become much bigger in the future.