GPT‑4.5 nears human mimicry, passing the Turing test at 73%
Does GPT chat think in a similar way to us? The media has reported that artificial intelligence has passed the Turing test. Does this mean that AI might become indistinguishable from humans?
Researchers from the University of California, San Diego conducted a study which showed that the GPT-4.5 language model from OpenAI was mistaken for a human in 73% of cases. The Turing test, meant to confirm this, involved conducting a text conversation with both the chat and a human simultaneously. The study aimed to determine whether participants could differentiate between a human and a machine.
Other models, such as LLaMa-3.1-405B, were also tested but were mistaken for humans in only 56% of cases. Older systems, like ELIZA and GPT-4o, were much less effective, with only 23% and 21% incorrect identifications, respectively.
The Turing test has sparked many debates about consciousness, thinking, and the limitations of artificial intelligence. It remains one of the most well-known and controversial symbols of discussions concerning the relationship between humans and machines. It has been popularised as the ultimate indicator of machine intelligence, but the debate continues over what is truly being measured and whether it genuinely assesses machine intelligence.
The Turing test was introduced in 1948 by Alan Turing, a renowned computer scientist. Initially, the test involved three people: two players and an operator who were supposed to play chess with a hypothetical machine. This experiment aimed to simulate a game, and its creator believed it was an effective method to assess intelligent machine behaviour.
The study has sparked discussions about the limitations of artificial intelligence and its ability to emulate human thinking. Critics point out the short duration of the test and the fact that the AI adopted a specific personality, which could have influenced the results.
