Certainly LLMs are not AGI, and AGI has not yet been built. But what's your knock-down argument for AGI being "science fiction"?
There’s a great deal of science on figuring out what intelligence is, what sentience is, how learning works. A good deal of it is inconclusive and not fully understood.
AI is a misnomer and AGI is based on a false premise. These are algorithms and systems in the family of Machine Learning. Impressive stuff but they’re still programs that run on fancy calculators and no amount of reductive analogies are going to change that.
I assert that machine learning is learning and machine intelligence is intelligence. We don't say that airplanes don't really fly because they don't have feathers or flap their wings. We don't say that mRNA vaccines aren't real vaccines because we created them with CRISPR instead of by isolating dead or weakened viruses.
What matters, I believe, is what LLMs can do, and they're scarily close to being able to do as much, or more, than any human can do in terms of reasoning, despite the limitations of not having much working memory, and being based on a very simple architecture that is only designed to predict tokens. Imagine what other models might be capable of if we stumbled onto a more efficient architecture, one that doesn't spend most of its parameter weights memorizing the internet, and instead ends up putting them to use representing concepts. A model that forgets more easily, but generalizes better.
That’s what makes my day: AGI folks have to rely on myth-making and speculation.
The here and now incarnation of GPT-4 is what it is.
I’m not saying it isn’t useful, powerful, or interesting. I’m saying that needless speculation isn’t helping to inform people of the real dangers that such proselytizing is causing.
But there's speculation of the fantasy sort, and then there's speculation of the well-grounded "this conclusion follows from that one" sort, and the AGI folks seem to be mostly in the second camp.
And yeah, unlike GPT-4, AGI isn't here-and-now. But GPT-2 was an amusing toy in 2019, and GPT-3 was an intriguing curiosity in 2020. In 2014, "computers [didn't] stand a chance against humans" at Go[2], but two years later, it was humans who no longer stood a chance against computers. The here-and-now is changing fast these days. Don't you think it's worth looking even a little bit up the road ahead?
Isn't there some role here for speculation?
[1] https://www.vice.com/en/article/kbzd3a/the-new-york-times-19...
I like how Robert Miles puts it (he's speaking about Safety in the sense of AI not taking over and/or killing everyone):
"I guess my question for people who don't think AI Safety research should be prioritised is: What observation would convince you that this is a major problem, which wouldn't also be too late if it in fact was a major problem?"
What is an example of a task that demonstrates either intelligence, sentience, or learning, and which you don't think computer scientists will be able to get a computer to do within, say... the next 10 years?
You don't have to understand how something works to build it.
An algorithm may perform a task such as building models that allow it to solve complex problems without supervision in training. That doesn’t mean it’s intelligent.
If autopoiesis turns out to be necessary for AGI and we embody these systems and embed them in the real world, are they still going to be fancy calculators?