Oh my god, Ultraman’s discussion of GPT has jumped several generations to GPT-8 !
Recently, he had a conversation with David Deutsch, the founder of quantum computing , on a show. Regarding the issue of " whether AI can develop into conscious superintelligence ", on which the two had a disagreement, Altman brought out GPT-8 to try to convince Deutsch:
GPT-8 has figured out quantum gravity and can even tell you how it did it—for example, what questions it was thinking about and why it decided to research this area. But what it presents still looks like the output of a language model. But it has indeed solved the problem of quantum gravity. Will you then accept it?
This conversation has attracted a lot of discussion online. Some netizens believe that the conversation between the two shows that the definition of AGI (artificial general intelligence) is vague, the goals are constantly changing, and it is impossible to measure it:
One could argue that we’ll soon surpass that and set our sights on ASI (artificial superintelligence).
So, what did the two of them say?
Complete conversation summary
During the show, the host first asked Altman, " Is your favorite book really The Beginning of Infinity by David Deutsch? " Altman answered in the affirmative and then explained why he liked the book:
Even if you haven't read the entire book, the first forty or fifty pages are already very exciting. I think it explains with extreme optimism that even in a world with AI, humans will never lack things to do, ways to contribute, problems to solve, and unknowns to explore.
And it does a wonderful job of explaining how relatively simple developments in human history have brought us to this amazing place today.
After Altman shared his views on the book, David Deutsch joined the show online.
The host then changed the subject: "I remember you and him (David Deutsch) had a disagreement on one issue: Is it possible for artificial intelligence to develop into conscious superintelligence? He (David Deutsch) seemed to think it was impossible, while you (Altman) thought it was possible. Perhaps you could elaborate on your agreement and disagreement."
Ultraman quickly said, "I don't care about that issue, there is no disagreement."
David Deutsch then took the stage to explain why he believes AI will not develop into conscious superintelligence:
On my computer, I keep a list of developments that I thought were impossible but have since become reality. One of them, I'm embarrassed to admit, is the World Wide Web.
Another one is that I used to think that no computer program could have an open-ended conversation on a wide range of topics in natural language, unless it was an AGI. I prefer to call this ability " explanatory creativity ."
ChatGPT got me wrong.
It's not AGI, but it can have a conversation. This ability is a byproduct of knowledge.
Unlike the 1960s chatbot Eliza, which could only use the words and phrases you input, ChatGPT can chat about anything, drawing on a vast body of knowledge. This makes it incredibly useful. For some, it's "too useful," making them feel like they're talking to a person or an AGI. Just like Eliza back then, users treat it like a human.
This reminds me of a widespread myth - the Turing test .
In reality, Turing never proposed a test or benchmark for AGI. His "Imitation Game" wasn't a test, but a thought experiment designed to shatter the intuition that machines can't think. In reality, there's no such thing as a benchmark, because a truly general intelligence must be able to choose to remain silent .
This in itself shows that existing methods cannot create AGI.
Existing systems can and must be benchmarked.
Conversely, if something offers a completely new explanation, you have no way of testing whether it was created by itself or by humans—even if you were to do the test yourself. In Edison's words, intelligence consists of "inspiration," which only humans can achieve, and "perspiration," which machines can help us liberate.
So, since there is no test benchmark, how do we know that humans are general intelligence? The answer is through "telling stories about ourselves."
Human thinking isn't a mechanical process of converting motivations into actions or cues into outputs; it's primarily a process of actively selecting motivations . Just as scientific research isn't about extracting theories from data, but rather identifying problems, proposing explanatory hypotheses, and then critiquing and validating those hypotheses.
So how do you tell if something is doing this? Not always. Sometimes you're really just chatting with a robot.
But if there is no explanation that you yourself are a robot, or that humans as a whole are robots, then the reasonable assumption is that “we are not robots.”
Some people like to question whether Einstein actually invented the theory of relativity or simply cobbled it together mechanically from various existing ideas. We know he did, because we know his story—what problems he was solving and why he was solving them.
As we know, Sam Altman didn't need to write any code to turn ChatGPT from a technical concept into an actual product, becoming a phenomenal existence. He relied on intuition and the courage to "determine that this is the right direction for humanity to try next."
There's no program that can give a computer this kind of intuition, at least not yet.
Hearing this, Altman couldn't help but want to refute and brought up GPT-8:
You mentioned Einstein's theory of relativity, and I completely agree with you. I think relativity is one of the most ingenious things humanity has ever figured out, perhaps even the most ingenious. And Einstein has his own story; we know what he was studying.
Suppose, a few years from now, GPT-8 has figured out quantum gravity and can tell you how it did it—for example, what questions it was thinking about and why it decided to pursue this area of research. While the content it presents will still appear to be the output of a language model, it does indeed solve the problem of quantum gravity. Would you then accept it? I know you keep track of past misjudgments, and I do the same. So, would this scenario convince you?
David Deutsch:
I think so. Yes, the "process story" is crucial in this. I also think it is the key to judgment.
Quantum computing pioneer: David Deutsch
David Deutsch is a founding member of the Centre for Quantum Computing at the Clarendon Laboratory at the University of Oxford and a visiting professor of physics. He is also an honorary fellow of Wolfson College, Oxford, and a fellow of the Royal Society and the Institute of Physics. He is considered one of the founders of quantum computing and quantum information science .
He works on fundamental problems in physics, especially quantum computation and information theory, as well as constructor theory.
In 1985, he proposed the theoretical framework of quantum Turing machines . By introducing the network operation method of quantum logic gates, he proved that quantum computers can theoretically accurately simulate any physical system that conforms to the laws of quantum mechanics, laying the mathematical foundation for the development of quantum computers.
In 1992, he and Australian mathematician and theoretical physicist Richard Jozsa jointly proposed the Deutsch-Jozsa algorithm , which was the first to prove that quantum algorithms have an exponential acceleration effect compared to classical algorithms, laying the theoretical foundation for the subsequent development of Shor's algorithm and Grover's algorithm.
He is also the author of two well-known books for general readers: The Fabric of Reality and The Beginning of Infinity .
David Deutsch has won awards including the ICTP Dirac Prize, the Micius Quantum Prize, two Breakthrough Prizes in Physics, and the Isaac Newton Medal .
Reference Links:
[1]https://www.youtube.com/watch?v=WZ22AJmuKKQ
[2]https://x.com/rohanpaul_ai/status/1971884939667738838[3]https://www.daviddeutsch.org.uk/
This article comes from the WeChat public account "Quantum Bit" , author: Xifeng, and is authorized to be published by 36Kr.
