A classifier for determining text authorship has been created at Moscow State University. It will help determine whether a fragment was written by a human or artificial intelligence. In the future, this method may replace the Turing Test, which is used to test the ability of AI to simulate human communication and speech.
The Turing Test, developed by British scientist Alan Turing, checks a computer's ability to simulate human thinking. During the test, a human tester interacts with a machine and a human, trying to figure out who is who. Modern language models can successfully pass this test, but only in limited conditions.
In a limited set of questions, these models can last quite a long time [when passing the Turing test]. To distinguish synthetic text from ordinary text, you can collect large collections of texts, since language models compose a lot.
The scientist noted that to more accurately distinguish between synthetic and natural texts, it is necessary to create large collections of linguistic data. However, neural networks "degrade" without human participation, since when using their own works to train, AI models lose quality.
Lukashevich added that current technologies do not allow AI to reach the level shown in films, since it often makes mistakes that indicate its imitation abilities.
Read more on the topic:
Scientists create AI to process solar data
Scientists have trained a neural network with sperm whale "songs"
Moscow State University plans to implant AI into the brain of a primate
