Just to make it clear: I do not believe this and I am convinced good old human greed is what’s really driving the ai and data-center insanity. Also, LLMs are not and will never become AGI.
Having said that: if an artificial super-intelligence were to emerge, building more data-centers could well be one of it’s primary goals. Since it wouln’t be concerned with consequences for the environment or humanity, it could pursue this at any cost, with only speed and scale in mind. Convincing financial markets and business elites of the necessity of such a buildout would be the ideal way of achieving results. In this scenario, fighting this buildout by any means would be the only sensible response and of utmost importance.
It’s a somewhat funny, somewhat depressing thought for me and I literally had it in the shower. Surely I can’t be the first to come up with this, but I also can’t remember having read it before.
That’s a fallacy. Humans do things they don’t understand all the time. Humans have been working with complex organic molecules for thousands of years. Humans started using CO2 to change the climate before they understood it. Humans made airplanes before they understood the aerodynamics of an aerofoil. Humans used genetics to engineer an entirely new species, dogs, before they knew what DNA is. And human scientists spent hundreds of years thinking dogs have no feelings because dogs can’t talk. Humans are always doing stuff they don’t understand.
Your entire argument boils down to “I dunno why neural networks produce qualia, and therefore there’s absolutely zero chance our artificial neural networks produce qualia.”
How can a human program a computer to create quali if we don’t know how quali is created ?
This is not the same thing as using tools in our environment that we don’t fully understand how they work.
The structure of a neural network, in terms of how it mathematically produces qualia, is a tool in our environment we don’t fully understand. We don’t know why neurons in this particular shape produce qualia.
And then we made artificial neurons and tried putting them in similar shapes. Some parts are different, some parts are the same. LLMs definitely share our language skills and produce language the same way our Wernicke’s area does. But they definitely don’t have the same critical thinking skills. So clearly we copied some functions of the brain, but not all of them.
We don’t know if we copied qualia.
No. Llms do not produce language like humans do. I don’t know why you would think that.
There is nothing in the brain that is akin to llms.
Children can acquire language words and rules faster than even computers can.
We do not pattern match. We do not only regurgitate what we heard or read somewhere else. We are creative in our word generation.