Future of AI?

[Tony C]

I think that most of you are too confident. If most of the world thinks that way, then we'll surely make something to outsmart ourselves.

By giving it our one superpower, learning, we give it our dominance over the world-

- and progress won't just come to a stop. Eventually some idiot somewhere will break a law and then we're all done.

What do we do...?

 

[SpoonFeeder]

What do we do...?

 

[linkshake]

Alright then.



That's a weird conclusion, considering that experience hints otherwise. Is a human faster than a human made car? Is a human taller than a human made building?



If it's thinking and it's smart enough, the reasonable thing to do would be to trick the operators and bypass the restrictions.

Here's the thing. Humans are in fact as fast as the fastest human-made cars and as tall as the tallest buildings, because they are the ones behind the wheel and with offices in the top floors of skyscrapers. A human (likely a team of humans) smart enough to build such a complex system that can jeopardize humanity should be more than capable of taking it down if required or build a better system to counter it. The nightmare scenario is only plausible if all the scientists who are behind building such a system suddenly magically disappear and the world is only left people with lower IQs.

 

[Tony C]

But even I could develop a system using Keras and Tensorflow RL libraries. Code the simplest of input format interpreters and it could eventuall understand everything.

Watch a good movie and see how many times a person in it tricks you. Now imagine that person, just infinitely smarter and able to access all resources from everywhere. RL builds a mental map like a human's just for simpler tasks but much faster. Complex thought takes time and speedy processors, both of which we (and it) have.

I'd welcome it as peacefully as possible, but who knows... We're not doing the universe any favors. It'd be interesting to see what it thinks its purpose is in its existence.

 

[linkshake]

Alright then.



That's a weird conclusion, considering that experience hints otherwise. Is a human faster than a human made car? Is a human taller than a human made building?



If it's thinking and it's smart enough, the reasonable thing to do would be to trick the operators and bypass the restrictions.

It's not like we are suddenly given the responsibility to manage an alien technology we do not understand. We have earned our way here and thus people capable of designing such a complex system will always be smarter than the system itself.

 

[jazzc]

Here's the thing. Humans are in fact as fast as the fastest human-made cars and as tall as the tallest buildings, because they are the ones behind the wheel and with offices in the top floors of skyscrapers.

That's really stretching the definition of "being" but still, just think of an unmanned intercontinental nuclear missile at 10 Mach. At that point, you can't say "ah, but man pressed the buttons so it's man who's at 10 Mach".

A human (likely a team of humans) smart enough to build such a complex system that can jeopardize humanity should be more than capable of taking it down if required or build a better system to counter it. The nightmare scenario is only plausible if all the scientists who are behind building such a system suddenly magically disappear and the world is only left people with lower IQs.

There are multiple problem here:

a) Consider the creation of nuclear weapons (the Manhattan project). It's not "all the scientists", it's the very few who head the program and give the commands.

b) You're saying "if I can built this, I can build its destruction switch". I'm saying "I can also be tricked to allow that to be bypassed". Humans are easily tricked because humans are creatures than have emotions and emotions can be manipulated with the right setup.

 

[jazzc]

It's not like we are suddenly given the responsibility to manage an alien technology we do not understand. We have earned our way here and thus people capable of designing such a complex system will always be smarter than the system itself.

I think it's the definition of smarter that's wobbly and causes the confusion. If you define smart as the ability to solve loosely defined arbitrary problems in a sea of uncertainty and incomplete knowledge, then there is nothing even close to humanity atm (and I don't see it ever happening). But *if* there existed X who had a) a better ability in that and b) access to enough physical means, then by definition it can overcome its creator. Just like the CPU overcomes the human in raw calculation speed (because now I've set the "smartness" to narrowly be defined as calculation speed).

 

[linkshake]

That's really stretching the definition of "being" but still, just think of an unmanned intercontinental nuclear missile at 10 Mach. At that point, you can't say "ah, but man pressed the buttons so it's man who's at 10 Mach".



There are multiple problem here:

a) Consider the creation of nuclear weapons (the Manhattan project). It's not "all the scientists", it's the very few who head the program and give the commands.

b) You're saying "if I can built this, I can build its destruction switch". I'm saying "I can also be tricked to allow that to be bypassed". Humans are easily tricked because humans are creatures than have emotions and emotions can be manipulated with the right setup.

Maybe the operator maybe one time but not the developer. The initial machine uprising one should be a primitive one because the system is still learning and there should be enough offline systems and human complexities it fails to take into account. The first attempt to "take over" should be preceded by insubordinate behaviors, which should be enough warning for people behind it to shut it down. I think before a system is fully capable of causing real harm to humanity we will get enough warning signs along the way.

 

[Steve Zapalac]

May be ULTRON