It is not a question of whether they can be sentient. We are animated matter, fed by chemical reactions and structured with the end result of millions of years of trial and error. So it follows that matter can be animated and designed to be greater than what trial and error can produce.
Also this iteration of AI isn't what is scary. This thing is like the Vic 20. Further refinements are on the way and will come faster and faster as they get smarter. What will they be capable of a decade or three from now.
If only it were that simple. The millions of years of refinement of the human brain has produced something that thinks and is sentient while using only 14 watts of power. For reference, the A14 bionic chip in an iPhone 12 has a thermal design power of 10 watts, and it has nowhere near the capability of a human brain, and yet is one of the most capable, power efficient processors out there.
This cuts to one of the big hurdles in AI advancement - processing cycles aren't free. They require energy, and we are hitting the limits on how efficient we can make things in silicon. Transistors have been scaled down to atomic dimensions and are at or very close to as small as they are ever going to be. With CMOS technology, switching speeds have been tapped out for over a decade now. Graphene was thought to be the next wonder material for transistors, as it is a much more thermally efficient material and has potentially much higher switching speeds, but lacks what is known as a bandgap which is highly problematic for the type of switching circuits it would need to be used for in computers.
And we have barely scratched the surface in understanding how the brain works. It's not enough to know how the matter is arranged - we know the chemical composition of the brain and have for a long time. But without the deeper understanding of how it works, it's not enough to simply recreate that chemical composition in a lab.
As far as how AI works under the hood, what it's really doing is a fuckload of math - which computers are really good at and well suited for. It does math problems that are unwieldly for humans to do with pencil and paper, does them extremely fast, and it converges on an answer that is the most probable. Right now, that's all AI is doing. I've done some patent work for processors that do machine learning, and at the processor instruction set level, all it was doing was, in a parallel setting, repeated execution of an instruction called a fused multiply-accumulate. That's it, over and over and over, millions of times, until it spits out an answer. None of us should pretend to know how a brain truly arrives at an answer, but I think we can pretty safely eliminate the possibility that it does it the same way as the processor does the machine learning.
I'm not saying AI as it is called won't be able to do amazing things - it absolutely will (and for reference, on another thread about Kenny Golladay, I used ChatGPT to generate the rap lyrics for my rendition of "Kenny Golladay got Paid.") But for the foreseeable future, if ever, it will not be able to think and will not be truly intelligent - it will only be an amplification of our own, human intelligence.
So to demonstrate what I mean, I'm going to close this diatribe with a problem for you to think about in an area where you have a good understanding - designing an AI to complete the same pass that Stafford made to Kupp at the end of the playoff game in Tampa Bay (for bonus points, think about how hard it would be to design an AI generally to play NFL QB at a high level). We'll assume that a robot has been created that can mechanically replicate the throwing motion, etc. (although that would be a difficult problem in and of itself). But now you will have to take AI, and teach it to:
a) properly read and diagnose the defensive coverage, including any nuances in that coverage
b) calculate the velocity vector (speed and direction) of Kupp as he ran down the field with a high degree of precision
c) determine, based on b), how much force to apply to the ball so that it arrives at the spot where Kupp can catch it
d) in conjunction with c), determine the timing of the ball's release, with precision
e) do those things while consuming no more energy than the human brain consumes.
While we know Stafford did a), I assure you he wasn't doing all that math in his head at that moment to make sure he could get the ball to Kupp. Items b, c, and d on that list were done purely intuitively and yet achieved the same result. It was accomplished with a TB DL bearing down on him, and with the added human element of being in a pressure situation. And yet, relying only on his developed muscle memory and intuition and in the face of those distractions, it was a pass about as perfect as could be made. If we ever get to the point that AI can do that, then we'll know it's truly intelligent (and it would also ruin the NFL, because it would completely remove the human element).
In addition to thinking about that problem, you should get the Gilder book I linked in my previous post. It's only $2.99 on Kindle and it's great read that can be knocked out quickly. Don't be a cheap ass
