January 2 news, today, people just tap the keyboard, can let AI writing tools to help write papers, create rap songs or write scripts. For the time being, however, it cannot be expected to have "Shakespearean originality". According to Science, a new study shows that these AI works are still clearly derivative, at least for now.

To test this idea, the researchers developed a program that could measure AI creativity. Mirco Musolesi, a computer scientist at University College London who studies creativity in AI, notes that assessing creativity is "a complex and interesting challenge" and believes that the new approach does a good job of solving the problem of linguistic novelty.

Since the advent of generative AI and large language models, there have been questions in the scientific community about their creative abilities. Although these ais can quickly generate what appears to be human-created text, some scholars believe that they are not truly innovative, but simply reassemble the content of the training corpus. Critics compare IT to "random parrots"; (the blind repetition of known texts).

Yet quantifying this creativity is not easy. Scientists usually take two approaches: one is to use computers to detect signs of plagiarism - but "no plagiarism" does not mean "originality"; The second is to have human scores assess fluency and originality - but this process is subjective and time-consuming.

In response, Lu Ximing, a computer scientist at the University of Washington, and his team developed a tool called DJ Search. It combines objectivity with the ability to capture nuances. The tool extracts snippets of phrases from AI output text and looks for similar content in a database. It not only looks for an exact match, but also identifies semantically similar expressions. Meaning is analyzed by an AI algorithm, which identifies synonyms by embedding vectors. Finally, it evaluates the novelty of the AI output by calculating the proportion of unmatched parts in the text.

The study found that humans significantly outperformed AI in poetry, fiction, and speech by 80 percent, 100 percent, and 150 percent, respectively.

DJ Search can also compare human works, such as The Hunger Games, with 35% more linguistic originality than Twilight. Lu, the research leader, likens AI to DJS: "They stitch together text like a DJ mixes music, but while wonderful, they can't replace a composer.";

Nanyun Violet Peng, a computer scientist at the University of California, Los Angeles, suggests that in the future, the originality of the overall narrative should be evaluated, not just the language level.