GPT-4 Based Chatbot Passes The Turing Test
A team of scientists from the University of California recently conducted the world-famous Turing test to find out whether modern artificial intelligence (AI) models can plausibly simulate human communication and how many people can be fooled by allowing them to communicate with AI.
The goal of the study was to determine whether participants could distinguish a synthetic interlocutor from a real person. The scientists’ study is a modern interpretation of the test, which was proposed by the famous mathematician Alan Turing back in 1950. The test is considered passed if the AI algorithm, in the process of communicating with a person, can make him think that another person is having a conversation with him.
The testing involved 500 people, who for some time alternately talked with four agents, one of which was a person, and three more were software products, such as the virtual interlocutor ELIZA written in the 60s of the last century and modern chat bots built on based on the large language models GPT-3.5 and GPT-4 (the latter is also the basis of the popular AI bot ChatGPT). As a result, it was found that 54% of test participants mistook GPT-4 for a person.
This is a huge step forward and an indication of how much AI has changed in the GPT era. This is a great challenge to the future interaction of man and machine, which could bring with it large-scale social and economic consequences for all humanity,” the scientists emphasize
