Even senior experts were unable to accurately identify scientific papers written by ChatGPT

News on January 14th, the American popular science magazine "Scientific American" today published an article​​, stating that was powered by ChatGPT The scientific papers written are so academic that even senior scientists cannot tell the difference.

In a preprint paper posted on the bioRxiv server in late December, they wrote that artificial intelligence (AI) chatbots have become capable of writing fake research papers so convincingly that scientists are not It's impossible to tell even if you screen carefully.

Sandra Wachter, who studies technology and regulation at the University of Oxford in the UK, said: "I am very worried about this. If even experts cannot distinguish whether the paper is true or false, it will affect the cornerstone of our research."

IT Home Classroom: Chatbot ChatGPT creates realistic and smart-sounding text based on user prompts. It is a "large language model," a neural network-based system that learns to perform tasks by digesting large amounts of existing human-generated text. OpenAI, a software company based in San Francisco, California, released the tool on November 30 and is free to use.

A team led by Catherine Gao of Northwestern University in Chicago, Illinois, used ChatGPT to generate artificial summaries of research papers to test whether scientists could discover them. The researchers asked the chatbot to conduct research based on research published in JAMA, The New England Journal of Medicine, The BMJ, The Lancet Write 50 medical research abstracts for anthologies in The Lancet and Nature Medicine.

They then compared these summaries with the original summaries through a plagiarism detector and an AI output detector, and asked a team of medical researchers to spot fabricated summaries.

The summaries generated by ChatGPT passed the plagiarism checker with flying colors: the median originality score was 100%, indicating that no plagiarism was detected.

The AI ​​output detector found 66% of generated summaries, while human review correctly identified only 68% of generated summaries and 86% of real summaries. Experts incorrectly identified 32% of generated summaries as real summaries and 14% of real summaries as generated.

The above is the detailed content of Even senior experts were unable to accurately identify scientific papers written by ChatGPT. For more information, please follow other related articles on the PHP Chinese website!