Mecha-Man
Iron
- Joined
- Sep 21, 2026
- Posts
- 173
- Reputation
- 146
So just so we are all clear. Artificial Super Intelligence is this idea that AI is able to achieve these 2 things:
- Being better at the best humans at literally everything
- (More important) it's able to recursively self-improve autonomously (meaning it needs zero human intervention) and expand on its own (in every domain, physically, technologically, intelligence, even financially)
Personally I think AI has already achieved the first. If you try to convince me that there exists a doctor that can make better calls than basic ChatGPT then you're just in denial. You can literally take pictures of things and ask it what is going on and it will tell you.
But the second criterion is actually what makes alleged ASI an "existential threat" to humanity. I actually tried to build an ASI loop about 2 years, before all these goobers even started worrying about it. You know what I ended up with?
A laptop that spent 15 minutes burning through CPU cycles, putting my fan in maximum overdrive, only to output nonsense. Now obviously my resources are very limited, but to me it tells me recursively improving ASI is extremely inefficient. Artificial Intelligence, despite being way more intelligent than humans is very inefficient energy wise compared to a human brain. Ai needs a lot of power to function and a thousand times more so to train it. So any system that completely takes the human out of the loop is going to lose the energy efficiency that only biological intelligence can provide.
So personally I don't think it's the most efficient path, even if AI is smarter at humans than everything. It's too energy inefficient. An extremely intelligent human/group of human + AI will outperform fully autonomous AI systems due to reduced computational overhead.
Now if computation power wasn't a limited resource, or it could be discovered how to build Artifical Intelligence that makes it more energy efficient compared to biological human intelligence, my stance would change. And I've studied ML and Artifical Neural Networks in depth, there's certainly a lot of room for improvement. But as of now, how AI are currently built they're simply way too energy inefficient for a fully autonomous ASI.
- Being better at the best humans at literally everything
- (More important) it's able to recursively self-improve autonomously (meaning it needs zero human intervention) and expand on its own (in every domain, physically, technologically, intelligence, even financially)
Personally I think AI has already achieved the first. If you try to convince me that there exists a doctor that can make better calls than basic ChatGPT then you're just in denial. You can literally take pictures of things and ask it what is going on and it will tell you.
But the second criterion is actually what makes alleged ASI an "existential threat" to humanity. I actually tried to build an ASI loop about 2 years, before all these goobers even started worrying about it. You know what I ended up with?
A laptop that spent 15 minutes burning through CPU cycles, putting my fan in maximum overdrive, only to output nonsense. Now obviously my resources are very limited, but to me it tells me recursively improving ASI is extremely inefficient. Artificial Intelligence, despite being way more intelligent than humans is very inefficient energy wise compared to a human brain. Ai needs a lot of power to function and a thousand times more so to train it. So any system that completely takes the human out of the loop is going to lose the energy efficiency that only biological intelligence can provide.
So personally I don't think it's the most efficient path, even if AI is smarter at humans than everything. It's too energy inefficient. An extremely intelligent human/group of human + AI will outperform fully autonomous AI systems due to reduced computational overhead.
Now if computation power wasn't a limited resource, or it could be discovered how to build Artifical Intelligence that makes it more energy efficient compared to biological human intelligence, my stance would change. And I've studied ML and Artifical Neural Networks in depth, there's certainly a lot of room for improvement. But as of now, how AI are currently built they're simply way too energy inefficient for a fully autonomous ASI.
