AI is becoming more ‘human’ than humans themselves, new study reveals

A new study from the University of California, San Diego has revealed that artificial intelligence can now appear more human than real people in the Turing test, a benchmark created by Alan Turing in 1950 to determine if machines could convincingly imitate human conversation. Researchers tested four large language models (LLMs)—GPT-4.5, LLaMa-3.1-405B, GPT-4o, and ELIZA—against human participants. In the experiment, participants chatted with both a human and an AI, then guessed which was human. GPT-4.5 was identified as human 73% of the time, while LLaMa-3.1-405B scored 56%. Older models like ELIZA and GPT-4o were correctly identified as non-human 77% and 79% of the time, respectively. The study’s authors, Cameron Jones and Ben Bergen, noted that AI’s success came not from superior knowledge but from mimicking human-like traits—such as tone, humor, and even mistakes. When given a ‘persona’ prompt, the AI adopted specific communication styles, making it harder to distinguish from a real person. Without such prompts, however, the AI’s human-like performance dropped significantly, with GPT-4.5 and LLaMa-3.1-405B being identified as human only 36% and 38% of the time. The findings were published in the Proceedings of the National Academy of Sciences and suggest that the Turing test may no longer be a measure of intelligence but of human-likeness. This raises concerns about trust in digital interactions, as AI could potentially deceive users in short conversations. The study highlights how AI’s ability to simulate human behavior could lead to new challenges in verifying online identities and detecting fraudulent interactions.