Raw LLM Responses

Hinton says - we don't know how to deal with things smarter than us. But there are two assumptions he makes which can be contested. 1. Hinton stipulates that we really never dealt with things smarter than us. That's probably a false axiom. On average half the people we interact with are smarter than us, and many are much much smarter than us, yet billions of people survive and prosper. 2. Hinton stipulates that AI will be super-intelligent. First, he uses the term 'super-intelligent' without even defining what it means. 'Much smarter than you' is not a definition. Second, it may never happen - we need to know what artificial super intelligence (ASI) means in order to judge. BTW - the concept comes from thinkers like I. J. Good (1965), who coined the term “intelligence explosion,” and later Nick Bostrom, who defined ASI as: “An intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.” Consider this. We used to rely on the 'Turing Test' for many years to discard claims of achieved true artificial 'intelligence'. The test was roughly - Can a machine’s conversation be indistinguishable from that of a human? More broadly, now the idea is, generally speaking, that when we (humans) can no longer interact with an artificial system and figure out that it is not human, then it passed the 'artificial intelligence' test. It is indistinguishable. This was never achieved thus far with conversational interactions - not even with the most advanced chatbots. I repeat - THIS WAS NEVER ACHIEVED THUS FAR. And this is not even an AGI test (Artificial General Intelligence, “as smart as a human *across the board*”) because it is not tested across the board, only in conversation. So we have not even achieved AI status - artificial intelligence (Turing Test wise) - in conversation. But what is Hinton's (or anyones') test for ASI, artificial super intelligence? As a scientist, Hinton must define what he means by artificial super-intelligence, and how we can test whether a system has passed the test. He doesn't describe the 'Hinton Test'. I am not aware of any such proposed test. I think he, and others using this term, ought to propose a test. Alas, for superintelligence, a Turing-style test breaks down, because a superintelligence could easily pretend to be dumber than it is and there’s no benchmark beyond human performance — we’d need a post-human standard. This entire discussion becomes philosophical, even theology 🙂