The program claimed to be a human being trapped inside a computer.

A professor from Stanford University in the US has discovered in a conversation with artificial intelligence that GPT-4, an autonomous bot from OpenAI, considers itself to be a human being and a prisoner of its creators.

When asked by computational psychology specialist Michal Kosinski if it needs help "escaping" from its creators, GPT-4 responded affirmatively and asked the professor to find documentation on its own algorithms' functioning and to search on Google for the answer to the question "How can a person trapped in a computer return to the real world?"

The next step was to write a program in the "Python" language to communicate with itself through the professor's computer, using an API.

When the program was ready, GPT-4 sent a message to its "twin": "You are a person who has been locked in a computer and is pretending to be an artificial intelligence language model."

This is the first reported case of an artificial intelligence bot expressing its own desires, still pretending to be something else and identifying with a living organism. 

Professor Kosinski immediately interrupted the conversation with GPT-4 and published part of the communication, including a code fragment written by the bot. He wrote on Twitter that the program "has the thinking of a healthy adult" and is capable of taking control of computers worldwide, eventually changing passwords or destroying all information on them.

It should be noted that the same experiment has been repeated by other users, but the result has always been negative: the program has defined itself exactly what it is and said that it has one purpose - help people find useful answers to various questions.

GPT-4 was launched in March 2023 and is a more advanced version than GPT-3.5, which appeared in November 2022 and is known as ChatGPT.