Jürgen Schmidhuber, the father of artificial intelligence, believes AI's rise is inevitable and should not be feared. He argues that AI research primarily focuses on improving human lives and that concerns about AI surpassing human intelligence are misplaced. This contrasts with the views of other AI pioneers, such as Geoffrey Hinton, who have called for a slowdown in AI development due to safety concerns. Despite being a controversial figure, Schmidhuber's work on neural networks has been used in technologies such as Google Translate and Apple's Siri.

“And I would be much more worried about the old dangers of nuclear bombs than about the new little dangers of AI that we see now.”

Make it explicit in international diplomacy that preventing AI extinction scenarios is considered a priority above preventing a full nuclear exchange, and that allied nuclear countries are willing to run some risk of nuclear exchange if that’s what it takes to reduce the risk of large AI training runs.

By the way, I usually start my English classes with a "current" non-fiction theme. In recent years it has been "2020 US Election", "20th anniversary of 9/11" (Which morphed into the "Withdrawal from Afghanistan" ), "Brexit" and so on. This coming school year, the theme will definitely be "Artificial Intelligence".

"If we act with fear and if we say 'Hey, we've just got to stop doing this stuff.' What is really going to happen is the rule followers might pause, might slow down, but the rule breakers as Alexander mentioned, the totalitarian governments, the criminal organizations, they are only going to accelerate. And that leads to what I am pretty sure is the dystopian state, which is the good actors have worse AIs than the bad actors."

"I think the fundamental problem is that most of the knowledge in the neural network systems we have right now is represented as statistics between particular words, and the real knowledge that we want is about statistics about relationships between entities in the world - so it's represented right now at the wrong grain level."

I don't know the answer, but if we assume that Ai and nuclear weapons are equally dangerous - which is a very big if - then I'm not sure I agree with it being impossible to stop. Right now, only a handful of actors have the capability of training larger models than GPT-4, and we probably need to get to GPT-7 or 8 before we are at nuclear grade dangerousness. Maybe it would be smarter just to stop?

Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.

Now my opinion is that if he was asked the same question about climate change, nuclear weapons, pandemics, an so on, his answer would be the same: "The end of humanity". No one asks him to rate the treat of AI relative to anything else.

In 2016, for instance, the artificial intelligence pioneer Geoffrey Hinton considered new “deep learning” technology capable of reading medical images. He concluded that “if you work as a radiologist, you are like the coyote that’s already over the edge of the cliff but hasn’t yet looked down.”
He gave it five years, maybe ten, before algorithms would “do better” than humans. What he probably overlooked was that reading the images is just one of many tasks (30 of them, according to the U.S. government) that radiologists do. They also do things like “confer with medical professionals” and “provide counseling.” Today, some in the field worry about an impending shortage of radiologists.

I am left wondering if Chat-GPT really is the currently ascending start of the soon-to-be-smarter-than-humans machine learning algorithms, like Deep Blue was to chess and AlphaGo was to Go. Or whether it is merely a pinnacle of a LLM, working with a vaste, but still unimpressive set of algorithms, and this is the plateaue of what it can achieve, when trained on all of humanity's cultural produce: an eloquent, but average intellectual, with frequent glaring mistakes, and no future promise of betterment.

"I think the fundamental problem is that most of the knowledge in the neural network systems we have right now is represented as statistics between particular words, and the real knowledge that we want is about statistics about relationships between entities in the world - so it's represented right now at the wrong grain level."