This proposal seems to test whether an AI has a human-like consciousness, while ignoring the possibility that an AI has an experience of its own self that is really different from our own.
The article seems to have forgotten that the Turing Test isn’t really a test, but a tautology: Turing was saying that if a machine can converse like a human, then it is intelligent, because the most fundamental aspect of human intelligence is our ability to communicate.
Exactly. You cannot create a convincing test for what you cannot define.
So, this proposed test is just a thought experiment demonstrating the pointlessness of the question. At least I hope that was the intent. If they really thought they had something, that would just be sad.
This test assumes a very specific type of AI, that is, one that is equal to humans but has been kept in the dark about parts of humanity. This assumption then must include its inability to teach itself new things, i.e. learn about those parts of humanity that were hidden from it, which then makes me skeptical of its abilities. Honestly I find this state of AI highly unlikely, but not impossible. More than that, even if you could create this type of AI and it is successful in this test, it still leaves room for doubt. It's interesting to consider that it would raise our Bayesian priors, but it proves nothing. I do think we should be striving to solve this issue but I'm skeptical we can without understanding consciousness in general.
4 comments
[ 24.4 ms ] story [ 60.0 ms ] threadThe article seems to have forgotten that the Turing Test isn’t really a test, but a tautology: Turing was saying that if a machine can converse like a human, then it is intelligent, because the most fundamental aspect of human intelligence is our ability to communicate.
So, this proposed test is just a thought experiment demonstrating the pointlessness of the question. At least I hope that was the intent. If they really thought they had something, that would just be sad.