GPT-4.5 Excels in Triadic Turing Test

Researchers conducted a triadic Turing test involving four AI systems—ELIZA, GPT-4o, LLaMa-3.1-405B, and GPT-4.5. The latter achieved the highest score.

In a paper published on March 31, Cameron Jones and Benjamin Bergen from the Department of Cognitive Science at the University of California, San Diego, shared the results of the experiment.

They employed an original triadic version of the test, where participants engaged in five-minute conversations simultaneously with another person and one of the AI systems, then determined which interlocutor they believed to be human. This version is more challenging compared to the test where individuals interact solely with a machine.

In 73% of cases, participants identified GPT-4.5 as human. Other AI systems scored lower:

• LLaMa-3.1 — 56%;
• ELIZA — 23%;
• GPT-4o — 21%.

“The data obtained represents the first empirical evidence that an artificial system passes the standard triadic Turing test,” the researchers noted.

The Turing test is a conceptual test proposed by British mathematician Alan Turing in 1950 to determine a computer’s ability to exhibit intelligent behavior indistinguishable from that of a human.

The essence of the test:

1. A person engages in a text-based conversation with two interlocutors: another person and an artificial intelligence.
2. If the participant cannot confidently identify which is the machine, the computer is considered to have passed the test.

The Turing test has been conducted multiple times among popular AI models. In June 2024, people failed to distinguish ChatGPT from a human interlocutor in 54% of cases. ELIZA then scored 22%, GPT-3.5 — 50%, and humans — 67%.

In 2023, in a similar study by Jones, GPT-4 scored 41%, GPT-3.5 — 14%, ELIZA — 27%. Humans scored 63% at that time.

In February 2025, OpenAI released a new version of the chatbot, GPT-4.5, with advanced “emotional intelligence.”