To avoid the AI exposing itself when asked, the researchers used three different language models, sometimes switching between them in a single conversation, and randomly prompting the AI to play someone with specific intentions and goals. They also added input latency for all players to mask the model's fast calculation time. The results of 1.5 million conversations showed that in 68% of cases, people could only tell whether they were dealing with another person or AI. Such accuracy is not much better than tossing a random coin. This result raises the question of whether a new generation of AI must self-declare itself in conversation. Computer scientist Alan Turing first proposed tests for machine intelligence in 1950. The initial test involves a person communicating with both man and machine through text, and then identifying who is a machine and who is human, and those machines that are considered human successfully pass the test. Later, people called such tests "Turing test". The recognition rate of machine intelligence is so low, that is, machines can pass the Turing test most of the time and fool humans, so should users of artificial intelligence have an obligation to inform? Should mandatory requirements be enacted in the future? This is a question that needs to be answered. #AI# #chatGTP AI# #理想的人工智能是什么样的#
