Early Chatbot ELIZA Beats OpenAI’s GPT-3.5 at Turing Test

An early computer program from the 1960s has triumphed over ChatGPT, a popular AI chatbot, in a recent Turing test study. UC San Diego researchers pitted ELIZA, a chatbot developed by MIT scientist Joseph Weizenbaum, against modern versions of AI technology. The Turing test, designed by Alan Turing in 1950, aims to distinguish between human and artificial intelligence. In the study, 652 human participants interacted with ELIZA and OpenAI’s GPT-3.5 AI to determine if they were conversing with a human or a chatbot. Surprisingly, ELIZA performed better than GPT-3.5, with a success rate of 27% compared to 14%. Experts have described ELIZA’s success as embarrassing for modern tech companies working on AI chatbots. However, some argue that ChatGPT was not designed to excel in the Turing test as it focuses on mitigating impersonation risks. The study suggests that ELIZA’s occasional uncooperative behavior actually helped its human-like deception. As the AI industry continues to evolve, the study raises questions about the capability and limitations of modern AI chatbots.