American researchers claim that ChatGPT-4 is the first neural network to pass the Turing Test, which evaluates an AI’s ability to mimic human intelligence in text conversations. In the study, 500 participants interacted with ChatGPT-4, ChatGPT-3.5, and the 1960s chatbot ELIZA.
The more often participants identified the machine as their conversation partner, the lower the AI’s score. Here are the results:
- ELIZA: 22%
- ChatGPT-3.5: 50%
- ChatGPT-4: 54%
Researchers believe that the latest version of OpenAI’s chatbot successfully passed the Turing Test. However, they emphasize that these results reflect the AI’s ability to mimic or deceive humans, rather than demonstrating actual intelligent behavior.