ChatGPT-4 Passes Turing Test, Shows Human Equivalent Intelligence

JAKARTA - Scientists from UC San Diego claim that ChatGPT-4 is the first artificial intelligence (AI) to make it through the Turing Test, a well-known test that assesses AI's ability to mimic human intelligence.

In a recent study published on pre-print platforms, they revealed that ChatGPT-4 managed to deceive a human test subject as much as 54 percent of the time given.

The Turing Test, proposed by Alan Turing in 1950, assesses whether a computer can be considered intelligent if people can't tell the difference between whether they talk to humans or machines through a text interface.

In this experiment, 500 participants were invited to talk to four different agents: three AIs and one human. The AI tested were ChatGPT-4, ChatGPT-3.5, and a 1960s-era chat program called ELIZA.

The results showed that while ELIZA was only able to deceive participants as much as 22 percent of the time, ChatGPT-3.5 managed to escape undetected in 50 percent of conversations. ChatGPT-4 is even more superior with a 54 percent success rate, meaning this AI is capable of pretending to be human more often than expected from random opportunities.

The researchers note that these results provide the first empirical demonstration that artificial systems have successfully passed the two-player interactive Turing Test. However, they also emphasize that these tests may reflect AI's ability to emulate or deceive humans rather than show true intelligence.

Nell Watson, an AI researcher at the Institute of Electrical and Electronics Engineers (IEEE), stated that AIs such as ChatGPT-4 show human-like characteristics, which make them more sophisticated than previous AI models that have only an existing list of responses.

The study also notes that shifting public perceptions of AI may affect Turing Test results. In the past, without the presence of advanced AI, anything that sounds human-like is considered human. However, with increased awareness of AI, people became more likely to misidentify humans as AI.

Even so, AI's ability to imitate humans still has great economic and social consequences, making Turing Test an important evaluation tool in the development of future AI.