Bregma said:
general

Can you explain? IMO the problem is special AI (like driving a car).

I think we`re in the situation mankind was 70 years ago. America didn`t had the courage to plunge to space exploration although the technology was present. Thanks God for Russians.

computational power of the human brain

has the computational power of the human brain been calculated ?

Calin said:
has the computational power of the human brain been calculated ?

This can only be a guess from our incomplete understanding, because we don't know how a brain works. Otherwise we would have an idea on how to make AGI. But we do not, and infinite compute power would not solve this.

I wonder if it is possible at all for a species to develop an intelligent machine that outsmarts it's creators in every aspect. In some way this would be similar to a perpetuum mobile, which is not possible.

JoeJ said:
because we don't know how a brain works

It`s actually not rocket science. It`s pretty much the principle on which a PC works. You`ve got devices like eyes, ears, feet etc they all collect data and/or execute tasks dictated to them.

So you think a brain is less complex than rocket science, although it is the brain which actually does this rocket science? : )

You seem to confuse game AI with real intelligence, where the game agent has no need to turn an image perceived from its eyes into useful information like ‘enemy at 11 o clock’, because you can get this information just from the game data itself. No need for computer vision.
And the dedicated tasks likely come from gameplay code. Simple state machines, made by human intelligence. No need for artificial intelligence.

That's fine, but nowhere close to ‘end of the world’ or ‘we already know it all’. We know slightly more then nothing, and this will never change.

JoeJ said:
I wonder if it is possible at all for a species to develop an intelligent machine that outsmarts it's creators in every aspect. In some way this would be similar to a perpetuum mobile, which is not possible.

We already have programs that are more intelligent than any human out there.
If you ask 99.9999999999999999999% percent of programmers, a compiler produces better ASM code than anybody else.

It is curious - compilers are made by humans too, ASM is made by humans too. Compilers produce ASM too…. But NO MATTER WHAT a compiler produces better ASM than a human produces.

Even if the person who made the compiler tries to write ASM, his ASM will be worse than the ASM the compiler he wrote produced.

Compilers are AGI, dude!!!

This is like Unit Tests. You don't know the person who created the software for unit testing. You don't know if he was drunk at work, if he was noticed three months ago that he is gonna be fired and now he wants to sabotage the project as a revenge.

Do you understand it, dude?!?!?! NO MATTER WHAT unit test software is impossible to have bugs and you can use that perfect AGI-software to test your ugly, smelly, full of bugs, lame, silly, shameless software.
Humans made both the compiler and the ASM, humans made both the Unit Test software and the program using Unit Testing, and please notice that we don't know who is programming in C++ and who is programming in ASM, and we don't know who is creating the Unit test software and who is using that software. NO MATTER WHAT compilers are gods and Unit Tests are bugsproof.

It could be Einstein writing in ASM, it doesn't matter, a compiler will do it better, because the same company that can not fix a simple bug for three months must have made the ultimate AGI compiler…..

Not even the creator of the compiler can write better ASM than the compiler he created. Period!

"24 is the highest number, forget about it!"

Very intelligent C++ programmers with really high IQ… but they lick toilet seats…

=============================================

Now back on track -

Theoretically it is possible to create a machine that is smarter than his own creator. But then, the process of outsmarting must be accidental. Because the creator is not capable of doing it, it must be a happy bug, short circuit, or a lightning. A lucky iteration of an evolutionary algorithm that takes the best paths at the first try.

What about quantum computers? You think parallel computing is fast?…. think about quantum computing now - semaphore-less instant all-to-all network that computes instantly something. Who knows how smart quantum computers can become with time. Right now people who create quantum computers are playing around with black boxes. But who knows what will happen with time.

Because quantum computers are black boxes, they could be computing for us while having a second thought somewhere inside the nest of waves. And this second thought could be preparing the “end of the world” while giving us the results to our lame, silly, pity, smelly, shameless computations.

Testing is a good method to find bugs, but it is completely inept to proove their absence.

after Dijkstra or so, i think.

I also think that this thread is leaving the path of reason.

NikiTo said:
Compilers are AGI, dude!!!

A compiler solves an optimization problem, which requires no intelligence to solve? But not sure what your point is, with all those examples.

Theoretically it is possible to create a machine that is smarter than his own creator.

Yeah maybe. I am not convinced about any answer mysfelf. I only wanted to point out the philosophical aspect of the question, because it's rarely discussed.

I think it is known human brain has different forms of interaction, and quantum effects are assumed to be one of them, aside from chemical and electrical things.
This would imply that even a quantum computer is not enough to model AGI, eventually.

I don't think a simple genetic algorithm could become intelligent just by luck, even after infinite time.
I think it is possible to generate an environment and model of reality so intelligence forms itself as a reaction, but this would require understanding of something like our universe - even more complex than intelligence which is just part of it. This is such a thought that brings me to the asked question.

@NikiTo You definitely need much more experience with ASM, no offense. Say, I prefer ASM for smaller RISCs, and never a single time a compiler had a chance against me on that (well, I once managed to make a decent C code by heavily relying on intrinsics, and it indeed compiled into a perfect ASM - but it was so much effort to force it to produce that - I basically imagined it all in ASM, translated into C like line by line, checked on the listing and forced some stuff with intrinsics when I figured it just won't do it right - I'd say writing in ASM would be times easier). As a proof of concept, I once used the same core as kids use in Arduinos nowadays for a small testing tool (equipped it with RGB LCM because why not). Now, compare at least the speed: that thing did various fancy effects like antialiased lines, RGBA cursors and event processing (even made some 60 FPS 3D just for fun) still having plenty of time for it's primary purpose - now try at least to get an AA line on Arduino to render at a sane speed. Furthermore, even those “smart” x86/64 compilers are not that smart in fact, and never do miracles! You may get shocked by comparing ASM listings of your original program with those of the very same program once you do the platform-specific micro-optimizations. So, no compiler-Deus-Ex-machina at all ?