Researchers found: ChatGPT is getting better at fooling people

According to researchers from the University of California, ChatGPT 4.0 has advanced abilities to imitate a conversation with a human. OpenAI’s latest released artificial intelligence model has proven this through the Turing test. Scientists conducted a test in which the task of a human being is to perform text conversation with a human and a machineand then indicate who he just had a conversation with. ChatGPT 4.0 achieved a good result – it managed to mislead as many as 54% of respondents.

The Turing test is a test that checks to what extent a machine can imitate human behavior. The machine doesn’t have to answer correctly – it has to how to best imitate knowledge and errors made by humans, which often requires underestimating the capabilities of artificial intelligence. What’s more, the test result does not depend only on the machine itselfbut also from the evaluator: what knowledge he has in the field of technology and human behavior.

Chat GPT 4.0 has proven that it can communicate like a human

Researchers at the University of California, San Diego conducted a study involving 500 participants who were asked to talk to four different interlocutors for five minutes each. The interviewees included: real people, ELIZA – a robot from 1966 created by Joseph Weizenbaum, ChatGPT-3.5 and GPT-4. After each conversation, participants had to determine whether their interlocutor was a human or a machine.

The test results were as follows:

• 22% of participants confused ELIZA with a real person.
• 50% of participants mistook GPT-3.5 for a real person.
• 54% of participants mistook GPT-4 for a real person.
• 67% of participants recognized the conversation with a human.

Exceeding the effectiveness barrier in “deceiving people” at the level of 50% allowed us to conclude that ChatGPT 4.0 successfully passed the Turing test. The ELIZA machine from 1966, which was easily identified as a fake caller in this study, She passed the Turing test as the first person in history. This shows that society and its awareness in the context of communication are developing along with machines.

On the other hand, the success of ChatGPT 4.0 in the Turing test prompts researchers to reflect on the risks associated with the use of AI and the need to balance technology development with applicable ethical standards.