Buterin: OpenAI's GPT-4 has passed the Turing test! Can deceive half of humanity

Ethereum co-founder Vitalik Buterin posted on the decentralized social platform Farcaster yesterday (16th) that the generative artificial intelligence (AI) model GPT-4 developed by OpenAI has Pass the Turing test.

What is the Turing test?

The Turing test is a fuzzy standard used to judge whether the dialogue model of an AI system reaches a "human-like" level. The term comes from the famous mathematician Alan Turing, who proposed this test method in 1950.

According to Alan Turing's point of view, if an AI system can generate text that makes people mistakenly think that they are talking to another person, then the system has the ability to "think".

Vitalik: Can deceive half of humans, AI has passed the Turing test

According to the latest research from the University of California, San Diego, "People cannot distinguish between GPT-4 and humans in the Turing test In "Performance," the researchers asked 500 human subjects to conduct a blind test to determine whether they were interacting with humans or AI. The results showed that the subjects had a 56% chance of mistaking GPT-4 for a human.

Based on the above results, Vitalik said that an AI system that can deceive more than half of humans has reached the standard of passing the Turing test. Vitalik added: This means that people's ability to identify whether it is a human or a robot is basically like tossing a coin. But he also pointed out that although the probability of a human being recognized as a human is 66%, the probability of a human being recognized as a human is 66%. The probability of robots is 54%, but the gap of 12% is very small, and it basically passes the test in any real situation.

In addition, Vitalik also said in the comment area that the Turing test is currently the most famous milestone recognized by society, but this important milestone has been surpassed by AI.

Passing the Turing Test does not equal AGI

On the other hand, many people confuse "passing the Turing Test" with "Artificial General Intelligence (AGI)". According to the standards of the Turing test, AI can deceive humans through dialogue and make people think it is a human. This kind of system with the ability to "think" can pass the Turing test.

However, the concept of AGI is more complex. In scientific terms, "general intelligence" should be able to complete any intelligence-based task, but no one can demonstrate "general" abilities within the scope of human intellectual activities. As a result, AGI will be able to think far beyond any known human being.

In other words, both AGI and AI can pass the Turing test and have the ability to "think", but AGI's thinking ability far exceeds the level of known humans.

The above is the detailed content of Buterin: OpenAI's GPT-4 has passed the Turing test! Can deceive half of humanity. For more information, please follow other related articles on the PHP Chinese website!