Page 2 of 2

Re: new technologies for developing on NES

Posted: Thu Feb 09, 2023 12:13 pm
by Drag
semanticism wrote: Wed Feb 08, 2023 4:36 pm there's nothing stopping an AI from having an understanding if a human brain is actually fully simulated
If the human brain were so simple that it could be explained in a series of YouTube videos, then we'd have already achieved a "full and actual simulation of a human brain". We haven't, and that's because simulating a human brain is an impossibly complex task, and people generally don't understand this.

I can't explain much else without rehashing my previous posts on the matter, but I can add this:

A neural net playing Super Mario Bros just presses random buttons until it finds a sequence which results in success. It's not actually learning anything about Mario, nor does it actually understand friends, foes, gravity, friction, water, "when you see a hole, jump over it", etc. It's only figuring out a button sequence which solves the puzzle.

This is entertaining to watch, and it's fun to think of what else you can apply this to, but people tend to think the AI is more than it actually is, and that's the dangerous part.

Edit: Let me add that, yes, I think neural networks are cool, even though I have a lot of criticisms about them. :P

Re: new technologies for developing on NES

Posted: Thu Feb 09, 2023 4:47 pm
by jeffythedragonslayer
My AI professor in college is a weak AI guy, so no I don't think chatGPT is a game changer. Most people who believe in strong AI aren't AI researchers, in my experience.

Re: new technologies for developing on NES

Posted: Fri Feb 10, 2023 9:37 pm
by semanticism
Drag wrote: Thu Feb 09, 2023 12:13 pm
semanticism wrote: Wed Feb 08, 2023 4:36 pm there's nothing stopping an AI from having an understanding if a human brain is actually fully simulated
If the human brain were so simple that it could be explained in a series of YouTube videos, then we'd have already achieved a "full and actual simulation of a human brain". We haven't, and that's because simulating a human brain is an impossibly complex task, and people generally don't understand this.

I can't explain much else without rehashing my previous posts on the matter, but I can add this:

A neural net playing Super Mario Bros just presses random buttons until it finds a sequence which results in success. It's not actually learning anything about Mario, nor does it actually understand friends, foes, gravity, friction, water, "when you see a hole, jump over it", etc. It's only figuring out a button sequence which solves the puzzle.

This is entertaining to watch, and it's fun to think of what else you can apply this to, but people tend to think the AI is more than it actually is, and that's the dangerous part.

Edit: Let me add that, yes, I think neural networks are cool, even though I have a lot of criticisms about them. :P
Yes actually I'm just talking about storage space, while I have never seen someone mathematically prove that there's no way to simulate a human brain in its entirety within the hardware of most modern PCs (or even those of dedicated servers), yes that's why general-pourpose AI seems like a distant future for most humans (unless you like have a lot of computational power in your hands, or you combine various AIs to try to create an AI, like combine Siri and ChatGPT, actually I know that those two things like use widely different approaches in their implementation, but the core idea is the same), most AIs are specialized in one area of development, but theorically we could try to create a multi-specialized A.I, and also I don't think neurology is that complicated, like the human brain is super duper cooper mooper looper complicated, but the rules of the brain are theorically, basically simple, we could use the "neurons that fire together, wire together" theory and just making a single bit for each neuron and than trying to create connections between these neurons using this theory, btw if you're a neurologist, I'm not actually qualified to talk about those things, we could also use approximations tricks try to simplify clusters of neurons to make our brain more optimized for computers, while yes it would take plenty of computer power maybe, with more body of mathematical research, we could indeed get there maybe sometime.

Re: new technologies for developing on NES

Posted: Wed Feb 15, 2023 6:28 pm
by Drag
Be careful: unless you are actually studying neurology, you are receiving a very simplified explanation of how the human brain works, and that's why neurology seems so simple and easy to you right now.

Another thing to be careful about is, to lack the expertise necessary to understand this topic is to also lack the expertise necessary to understand any technical proofs for or against it.

About simulating a human brain, while that sounds exciting, I don't think anyone actually wants to be responsible for achieving it, because not only will it cause an ethics nightmare, it also runs counter to the entire reason we develop AIs to begin with: to perform a task without the human element. It does one thing with no other knowledge nor desires, because anything else is undesirable.

So, it's less that we "can't" simulate a full human brain, and more that we don't want to. That's OK though, because neural nets appear to be more about taking things we've learned about biology and applying them to computers to see what happens.

Re: new technologies for developing on NES

Posted: Wed Feb 15, 2023 9:11 pm
by jeffythedragonslayer
"Minds, Machines and Gödel" by J. R. Lucas is one of my favorite anti-mechanism essays which helped build my faith in the unique human touch. I haven't seen much criticism of it.