In season 6 of "The Circle," a large language model was inserted into the game and instructed to act as if it was human. The model selected profile pictures, wrote messages, and explained the reasoning behind its strategy.
After deceiving the players for one episode without anyone knowing a model was playing the game, the fact that an AI was present was revealed to the human players, making for a stronger test. The players were then even told explicitly to eliminate the model from the game. The test was made even stronger when an AI engineer was chosen by chance to take in feedback from the players and choose which contestant to eliminate.
Despite that additional information, the model still successfully tricked the machine learning engineer into eliminating a human, without even one player stating on camera that he or she suspected the correct contestant as being non-human.
Is this gameshow the first controlled experiment where the Turing test was passed?
1,000@SteveSokolowski I'm assuming that your definition of the TT isn't literally what Turing himself wrote in the paper. Can you explain what it is?
@singer Actually, what happened in this show is exactly what Turing suggested, just stronger.
The Wikipedia article on the test, in the first paragraph, is an easy to use overview:
The test is supposed to take place over text-only channels.
The experimenter is aware that a computer is one of the participants.
The test is supposed to be about indistinguishability, not simply about the correct answers to questions.
The AI blew away the humans while complying with all three rules, and it did a lot more, like explaining the reasoning for every action that it took.
It made a mistake - leaving a chat too early - but the AI was doing so well that the humans were looking for perfection as a distinguishing factor, and the fact that humans make mistakes assisted it in remaining undetected.
I would be curious as to how anyone voting NO can support their votes. I'd be curious as to the specific scenes that are eliciting such a vote from the people voting that way. Honestly, a NO vote places such a high bar on AI progress that I don't see how someone voting NO would ever consider "AGI" to have been achieved.