How will AI change the world?

[Comp]

1. In how long are we likely to develop an AI at par or surpassing human intelligence?

2. If such an AI was made, would this invalidate science, that is, would would AIs be mass produced to solve all the problems, faster than any human could ever hope to solve em?

3. What math subfield is mostly used in theoretical CS in general, and what math is especially useful for the AI subfield?

4. What are the current roadblocks for AI? Are there any particular problems which if solved would completely turn the current research landscape upside down? Is strong AI impossible without solving these problems?

5. If AI gives a way to a type of immortality, what do you think will be the political implications of this? Full legalization? Banned? End of the world due to energy overconsumption?

[ProgrammingFun]

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

1. In how long are we likely to develop an AI at par or surpassing human intelligence?

I believe that this is impossible in the real world, because AI is only based off of human intelligence. AI only learns what humans already know, or are close to knowing.

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

2. If such an AI was made, would this invalidate science, that is, would would AIs be mass produced to solve all the problems, faster than any human could ever hope to solve em?

Such AI would still be based off of science because that is what developed it to begin with. AI cannot solve all problems associated with science because of one major flaw: it cannot adapt to and learn from its mistakes.

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

3. What math subfield is mostly used in theoretical CS in general, and what math is especially useful for the AI subfield?

No idea...

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

4. What are the current roadblocks for AI? Are there any particular problems which if solved would completely turn the current research landscape upside down? Is strong AI impossible without solving these problems?

Depends entirely on your portrayal of "Strong AI"

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

5. If AI gives a way to a type of immortality, what do you think will be the political implications of this? Full legalization? Banned? End of the world due to energy overconsumption?

If AI was to accomplish this, it would lead to an all out nuclear war to control that AI...which would lead to the destruction of such AI and the humans that created it.

This is what happens when AI gets too smart (note that it is still using human intelligence ): http://digitalwizardgames.com/strategy-games/i-am-an-insane-rogue-ai

[RandomLetters]

1. 2029

2. No. Understanding the universe IS science.

3.

4.

5. If I'm immortal, it won't matter what the government says, now will it?

[ultimatebuster]

Read Ray Kurzweil's book. EVERYTHING WILL BE ANSWERED.

[mirhagk]

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

1. In how long are we likely to develop an AI at par or surpassing human intelligence?

ProgrammingFun @ Sun Apr 24, 2011 4:51 pm wrote:

I believe that this is impossible in the real world, because AI is only based off of human intelligence. AI only learns what humans already know, or are close to knowing.

At the very least, it is theoritcally possible to map the human brain exactly, meaning AI is of course possible, it's all a matter of time. I'd suggest looking at Jeff Hawkins' book "On Intelligence". He proposes that an AI could be built to run on simple current generation laptops, that would surpass our intelligence.

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

2. If such an AI was made, would this invalidate science, that is, would would AIs be mass produced to solve all the problems, faster than any human could ever hope to solve em?

ProgrammingFun @ Sun Apr 24, 2011 4:51 pm wrote:

Such AI would still be based off of science because that is what developed it to begin with. AI cannot solve all problems associated with science because of one major flaw: it cannot adapt to and learn from its mistakes.

Not true many AI have learnt from mistakes, and have learnt to adapt to new environments, some even discovering things their creators did not know. (Their ability to adapt is of course not as good as ours.... yet)

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

3. What math subfield is mostly used in theoretical CS in general, and what math is especially useful for the AI subfield?

All fields of math are useful, some of the most are probability, and computation. Psychology and Biology are both just as useful, as we can then model our brain.

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

4. What are the current roadblocks for AI? Are there any particular problems which if solved would completely turn the current research landscape upside down? Is strong AI impossible without solving these problems?

The current roadblocks according to some are slow computers, to Jeff Hawkins it's the lack of research in modelling our brain accurately, rather than coming up with odd ideas that may work. If the heirarchal memory system is completed, a computer can learn just as well as humans, and will completely overturn research. Strong AI is possible without solving this, as we already have very sophisticated AI, which can learn, fight, adapt, and recognize objects in pictures that they've never seen.

Comp @ Sun Apr 24, 2011 4:35 pm wrote:

5. If AI gives a way to a type of immortality, what do you think will be the political implications of this? Full legalization? Banned? End of the world due to energy overconsumption?

If AI ever get created, we're basically screwed. If a true AI gets access to the internet, it can replicate itself on anyone's machine, and then hack into and control many automated factories. I'd give us about 3 days to live once an AI decided we were useless.

[ProgrammingFun]

mirhagk @ Sun Apr 24, 2011 9:03 pm wrote:

At the very least, it is theoritcally possible to map the human brain exactly, meaning AI is of course possible, it's all a matter of time. I'd suggest looking at Jeff Hawkins' book "On Intelligence". He proposes that an AI could be built to run on simple current generation laptops, that would surpass our intelligence.

Will definitely look into that book...but how can AI possibly surpass the collective human intelligence? It might surpass an individual AI but I don't think that it can collectively surpass all of it.

mirhagk @ Sun Apr 24, 2011 9:03 pm wrote:

If a true AI gets access to the internet, it can replicate itself on anyone's machine, and then hack into and control many automated factories.

But the easiest way to destroy it would be to remove its electrical supply or its internet supply...manually...something that such an AI cannot stop us from doing.

mirhagk @ Sun Apr 24, 2011 9:03 pm wrote:

I'd give us about 3 days to live once an AI decided we were useless.

That immediately reminded me of GLAdos

[mirhagk]

Yes we could manually remove it's electrical suuply. Assuming it has not already taken advantage of all electronically locked doors, and factories, ie creating it's own bots.

Oh and there's a good argument for why computers are actually faster than our brain (contrary to popular belief).

The brain has 100 billion neurons, 15% of which are the neocortex (the intelligence part, no emotions or w/e).

It's in these 15 billion neurons that we do things like catch a ball, or analyze a picture. Now a neuron takes approx. 5 ms to fire (really slow), so in 1 second, only a chain of 200 neurons can fire (yes you can multithread, but each spawn of a new thread is another firing used, ie to spawn and synchronize 5 threads, takes 10 neuron firings). this means it has a very small limit to the number of threads it can mantain, and still get the results into a single neuron

This means that to catch a ball, which can be done in a second easily, the brain can up to 15 billion 200 Hz processors, and we all know that multi-core does not scale proportionally. (so a 600 core 3 Ghz processor is definetly superior).

Many of these cores will need to synch on many different things, so a very generous estimate is that it can handle 1000 threads running at 200 HZ, which means that a single 200 Mhz processor beats it.

The specifics of all of this varies but the general idea is that the brain is a very slow computer, and so is more likely something like smart memory (where each sector can also do some logic) instead of the classic graphics card idea of multi-core processor.

[ProgrammingFun]

mirhagk @ Sun Apr 24, 2011 11:42 pm wrote:

Oh and there's a good argument for why computers are actually faster than our brain (contrary to popular belief).

They may be...but I still believe that an AI cannot surpass humans' collective intelligence (I'm making us sound alien aren't I )

mirhagk @ Sun Apr 24, 2011 11:42 pm wrote:

The specifics of all of this varies but the general idea is that the brain is a very slow computer, and so is more likely something like smart memory (where each sector can also do some logic) instead of the classic graphics card idea of multi-core processor.

The brain may be a slow computer, but it is also the most stable one, and one that carries the power to regenerate (to some extent), something that AI cannot do.

[ultimatebuster]

ProgrammingFun @ Mon Apr 25, 2011 7:13 am wrote:

mirhagk @ Sun Apr 24, 2011 11:42 pm wrote:

Oh and there's a good argument for why computers are actually faster than our brain (contrary to popular belief).

They may be...but I still believe that an AI cannot surpass humans' collective intelligence (I'm making us sound alien aren't I )

mirhagk @ Sun Apr 24, 2011 11:42 pm wrote:

The specifics of all of this varies but the general idea is that the brain is a very slow computer, and so is more likely something like smart memory (where each sector can also do some logic) instead of the classic graphics card idea of multi-core processor.

The brain may be a slow computer, but it is also the most stable one, and one that carries the power to regenerate (to some extent), something that AI cannot do.

Add the word "yet" to all your responses.

[ProgrammingFun]

ultimatebuster @ Mon Apr 25, 2011 9:31 am wrote:

Add the word "yet" to all your responses.

I don't agree...

[RandomLetters]

mirhagk @ Sun Apr 24, 2011 11:42 pm wrote:

Yes we could manually remove it's electrical suuply. Assuming it has not already taken advantage of all electronically locked doors, and factories, ie creating it's own bots.

Nukes > doors

[mirhagk]

ProgrammingFun @ Mon Apr 25, 2011 7:13 am wrote:

They may be...but I still believe that an AI cannot surpass humans' collective intelligence (I'm making us sound alien aren't I )

Why not? If a human intelligence AI were to be created, it could interface with the internet in the same way we interface with our arms. That's the thing about our brains, any knew thing is interfaced the same way, everything is processed the same, it's all just sensory data to our brain. So interfacing with new things is as easy as connecting it, and then our brain goes ahead and learns how to use it, and read the inputs. So with something where making connections is very easy, it could interface very easily, and add new components to itself very easily. (Meaning that it could grab the robotic arms and such and actually use them)

Anyways my point is that it could easily interface with the internet, being able to search that for knowledge, and quickly be able to learn new facts we don't know. Plus the fact that there is much more available computing power than brain power, and all of it is connected smartly, rather than communicating in super inefficient speech.

ProgrammingFun @ Mon Apr 25, 2011 7:13 am wrote:

The brain may be a slow computer, but it is also the most stable one, and one that carries the power to regenerate (to some extent), something that AI cannot do.

Well this simply isn't true, the brain is not the most stable computer, just ask someone with a brain disease. A computer can quickly be fixed, but brains are generally irrepairable.

Yes to a VERY basic extent our brain can regenerate, but a computer can easily be given tools with which to regenerate itself, and new components can easily be added and used, ie you can increase a computer's CPU speed, and memory etc, but you cannot speed up a brain.


PS: @RandomLetters, how are nukes controlled? Def not by computers right ;p

[Aange10]

Very interesting post. And I have to say, I honestly think it would be possible for an AI to be much more intelligent than a human. For instance, if you have ever seen the movie "Limitless", it talks about a guy who takes a pill that makes his brain remember everything he has ever done/read. Even though that is science fiction, what is stopping it from becoming a reality on a computer? If a computer never forgot anything, and was smart enough to learn, what is stopping it? Certainly not us.

If I remembered everything I have ever read/ will read I'd become the smartest man in the world over night. There is no question about it, I'd simply remember everything.

[mirhagk]

There's not enough bits in the world to remember everything. Even everything that a human remembers, a computer would have a hard time remembering (right now).

The thing is that our brain is SUPER effecient at storing data. If you picture a tree in your head, your calling on your memory. But does this tree look exactly like any tree you've seen? Likely not. Your brain stores rules for how things look, patterns,

Picture it this way, instead of storing how the tree looks, it stores that it has a general shape, and leaves and bark. Then it remembers how the leaves and bark look, and recreates the tree from that.

You'll quickly realize that this is considered very lossy compression since most of the original data is lost, and the picture is just reconstructed to a similar picture.

The way our brains work, we could not remember everything, it's designed not to, in fact it forgets things on purpose (remember when you were born?)

EDIT: Also intelligence has very little to do with how much information you remember. Wikipedia knows ALOT of facts, but it still can't answer a trivial question, or solve a problem, because it doesn't know how to respond to things, or understand things. Wolfram alpa is alot smarter than wikipedia because it can *somewhat* understand what you ask it, even though it has only a tiny fraction of the number of facts.

[Nick]

you guys are doing his homework btw