"Make a mistake on purpose to look like a human being"...Something unimaginable happened

Services that disguise contents created by artificial intelligence (AI) as if they were created by humans are pouring in.

According to the IT industry on Wednesday, U.S. AI company WordAI is providing AI detection avoidance service with the same name as its name for 9 dollars to 57 dollars per month. The company promotes that AI's writings are not filtered by AI detection technologies such as OpenAI, Originality Dot AI, and Huggling Face after reprocessing. AISO and Frazli Dot AI are also operating similar avoidance services.

A service that enables users to escape from AI detection is also called "AI humanizer," which means "humanization," in the sense that it makes writing style and expression as natural as written by humans. Customers of this service use spam mails and Google searches to advertise their products. On the Internet, it is easy to find ways to disguise AI-written articles as if they were written by humans. It is a common practice to adjust the length of sentences or mix everyday vocabulary. Some advise people to intentionally add grammatically incorrect expressions to reveal mistakes that people can make.

Big Tech is having difficulty in selecting AI contents. On the 1st, Meta decided to replace the "made with AI" mark on SNS such as Instagram, Facebook, and Thread with "AI information," which means that it is believed to have received AI help in some production processes. In the process of selecting images made by AI, actual photos taken by photographers were labeled with AI production. Industry sources estimate that applying Adobe's photo editing tool will cause AI labeling errors.

Some analysts say that it has become more difficult to identify AI content. A research team at Reading University in the U.K. disclosed to the scientific media "Plos One" last month the results of its study, which produced 33 answers with ChatGPT-4 on five short answer and essay tests required to obtain a bachelor's degree in psychology. The study found that 94 percent of AI-written answers were not detected by scorers. When comparing AI-written answers with those written by actual students, the probability that AI scores were higher reached 83 percent.