>> “AI models that can pass the Turing test [where someone cannot tell in conversation that they are not speaking to another human] are the same warning for the kind of AI that you can lose control over. That’s why you get people like Geoffrey Hinton and Yoshua Bengio – and even a lot of tech CEOs, at least in private – freaking out now.”
I don't understand that. Tegmark -let alone Hinton and Bengio- should be enough of an expert on AI to reocgnise that "the Turing test" (which Turing didn't even mean as a test) is no test of artificial intelligence, let alone the equivalent of finding out that Fermi built a self-sustaining nuclear reaction.
There are much better "A-IQ tests", like the Winograd Schmea or Bongard Problems, or more recent ones like the Abstraction and Reasoning Corpus. While the latter two stand for now, Winograd Schmeas have been defeated I don't see Tegmark or any other on his side of the existential risk debate "freaking out" about that. And with good reason: those tests were beat by sheer force of data and compute and not because some AI system entered anything comparable to a self-sustaining reaction.
Anyway the AI world freaks out every few years whenever a deep learning researcher beats some benchmark. Then everyone forgets about it five years later when the Next Big Thing in AI comes along. It's all par for the course.
I don't understand that. Tegmark -let alone Hinton and Bengio- should be enough of an expert on AI to reocgnise that "the Turing test" (which Turing didn't even mean as a test) is no test of artificial intelligence, let alone the equivalent of finding out that Fermi built a self-sustaining nuclear reaction.
There are much better "A-IQ tests", like the Winograd Schmea or Bongard Problems, or more recent ones like the Abstraction and Reasoning Corpus. While the latter two stand for now, Winograd Schmeas have been defeated I don't see Tegmark or any other on his side of the existential risk debate "freaking out" about that. And with good reason: those tests were beat by sheer force of data and compute and not because some AI system entered anything comparable to a self-sustaining reaction.
Anyway the AI world freaks out every few years whenever a deep learning researcher beats some benchmark. Then everyone forgets about it five years later when the Next Big Thing in AI comes along. It's all par for the course.