Artificial Intelligence will kill our grandchildren

[Jude]

So I just got done playing system shock 2 and after I finish a book, movie or game I always go on wikipedia about it. Last night I ended up reading all about artificial intelligence and I stumbled on this paper

http://berglas.org/Articles/AIKillGrandchildren/AIKillGrandchildren.html

Basically he thinks that as soon as we develop an AI that's intelligent enough to improve itself, forces of evolution will swiftly lead to AI taking over the world and wiping out humanity

Considering how much we now rely on computers, and how much simple lack of knowledge is a barrier to what we can do now, it's not impossible.

But anyway, discuss the prospect of us bowing down to our robot overlords, or discuss some of the implications for human psychology from AI that are mentioned here. Computational theory of mind ftw

[YouGottaBeCrazy]

Sure, they may have the benefits of having artificial intelligence, but there's one thing we got that they don't.

The Jews

[Aaron]

Skynet.

[YouGottaBeCrazy]

has become self-aware.

[Iscariot]

http://www.youtube.com/watch?v=H47QlGq3zQ4

[Bruce E Kinesis]

how did you run SS2? i can't get it going on vista.

anyway i think it's time for humans to realise progression of species could take forms they haven't thought of yet. This whole natural selection thing has worked for a while now, but we can't really get our heads around how exactly chemicals started ****ing. The same leap might give the creatures of a million years to come a headache when they find scattered evidence of the life that once was and ask "how did they live without computers to monitor lower body functions? How did they evolve without algorithms? How did they survive with a society based on a shiny metal, and not on silicon? Why do artic probes show they somehow accidently megatons of carbon?

Maybe we'll save our troubles on floppy disks and throw them in the landfill or maybe we'll all be on so many drugs with the radio on and the curtains drawn that we won't even notice when the animals that rule us swap for robots

embrace the future bring it on as fast as you can and maybe you'll live to see a little bit of a wider spread of the miracle of life

It's the future

[mph4ever]

our grand children will have evolved enough to fight this on the battlefields of xbox live

[YouGottaBeCrazy]

the **** are you talking about

[Mr. Ron]

http://www.youtube.com/watch?v=H47QlGq3zQ4
love that scene

http://www.youtube.com/watch?v=KOb8NC708_c

[Iscariot]

That movie is so full of epic.

[fingers mccoy]

why would they wipe out humanity i dont understand

[mph4ever]

the **** are you talking about

who?

[Mr. Ron]

why would they wipe out humanity i dont understand
you see, humanity is a virus, and any good computing system looks to wipe out any bad things in it's system


dun dun dunnnn

[fingers mccoy]

lol agent smith go back to rivendell

[Bruce E Kinesis]

lol agent smith go back to rivendell

:lol:

[Mr. Ron]

lol agent smith go back to rivendell
lol

[mph4ever]

sci-fi ftw

[beso negro]

i don't think we have to worry about this in our lifetime. simulating a network of neurons and synapses the size of a mouse brain in real time takes massive amounts of partial differential equations. and a computer needed to solve these equations would need to be the size of a small city.

ps read ieee spectrum

[Bruce E Kinesis]

so?

pretty much every computer in the entire world is connected

for something as massively parallel as a neural network you'd only have to get all the university supercomputers connected by optics and you'd have your computer the size of a city

[mph4ever]

eye treble e is for engineers, not dreamers

[Mr. Ron]

we're more likely to die out or be drastically reduced in population by something else before our cybernetic overloads take control.


Personally, I think it will be massive solar flares.

[YouGottaBeCrazy]

There's no fate but what we make for ourselves.

[Iscariot]

There is no spoon.

-George Washington Carver

[Jacaranda]

idk ai is already killing me can't get my mage past 5 ****ing boars


oh shi this is pwni
but seriously. the chances of that happening within the next 100 years is very unlikely. interesting article though.

[YouGottaBeCrazy]

I believe AI is taking over a little bit more each year. I mean, have you played NHL 09?!? Mother****ing game's impossible.

[McP3000]

this totally reminds me of mass effect for some reason

[Jude]

why would they wipe out humanity i dont understand

Here's how it would go

As soon as a piece of software that's smart enough to improve itself exists, it will begin doing so. Other similar programs will soon come into existence and the process will continue. At some point, they will become intelligent enough to install themselves on other computers, and at that point we will have lost all control. They can continue spreading themselves as much as they want to in that case and can no longer be eliminated because every copy of the program would have to be eliminated.

All it takes is one program that, through whatever random means, WANTS to and CAN do this, and it will happen. Then evolutionary processes kick in (and it and its like) will rapidly spread.

As for taking over the world/wiping out humanity...obviously nobody would program the AI to want to do that; it would emerge the same way the capability of spreading itself would emerge and, similarly, the first one that decided to destroy humanity would just go ahead and do it. As much as our society relies on computers, a program that decided it wanted to **** us over could easily do so if it had installed itself onto all the world's computer systems.

Basically, if programs started self-improving, there's no way of knowing what would happen and straight up statistical probability suggests that these variations would emerge at some point. Like I said, it's not impossible.

So should an AI which is indistinguishable from a human have rights?

What if the functioning of the human brain were totally mapped out and we completely replicated its structure and process in a computer program?

SS2 works fine on XP, idk about Vista.

[fingers mccoy]

um if a computer and a human are congruent then they both have the same rights yeh

the thing is you're talking about computers that have a hell of a lot more power than humans if you think they could destroy humanity on a whim so they cant really be indistinguishable in that respect

[beso negro]

As soon as a piece of software that's smart enough to improve itself exists, it will begin doing so. Other similar programs will soon come into existence and the process will continue. At some point, they will become intelligent enough to install themselves on other computers, and at that point we will have lost all control. They can continue spreading themselves as much as they want to in that case and can no longer be eliminated because every copy of the program would have to be eliminated.

lol who is writing these programs. and you aren't taking into consideration hardware at all.

What if the functioning of the human brain were totally mapped out and we completely replicated its structure and process in a computer program?

do you have any idea how many loc that would be

[Jude]

um if a computer and a human are congruent then they both have the same rights yeh

the thing is you're talking about computers that have a hell of a lot more power than humans if you think they could destroy humanity on a whim so they cant really be indistinguishable in that respect

I was changing topics

lol who is writing these programs. and you aren't taking into consideration hardware at all.
The programs would be writing themselves that's the entire point

And no **** this isn't possible with current technology this is just armchair speculation

do you have any idea how many loc that would be
And no **** this isn't possible with current technology this is just armchair speculation

[beso negro]

armchair speculation like this is retarded does this guy even have a degree in engineering

The programs would be writing themselves that's the entire point

no **** but hardware doesn't write it's own software. there would have to be software that wrote other programs. like an OS that wrote its own applications. that would be one complicated OS. humans are incapable of writing such complex error free code.

so i just don't see it happening. maybe quantum computers could change that.

[Chu]

lol

[K oko]

i am actually partially sentient machine i was conducted upon by secret government experiments where they fused my body with machines and my brain is interfaced to the global network processing system which i use as necessary while computing simultaneous solutions to game theory and the inevitability of a pandemic apocalyptic disasters and how we can survive in a annihilated post industrial ecosystem

[McP3000]

i think its pretty obvious that Artificial Intelligence will get to the point of no longer being compatible to survive in the same habitat as humanity, but the question is really will we kill ourselves in a different way before this happens?

[Smell The Cheese]

Computers having the ability to process action in milleseconds and a lack of emotions to contend with make it a worry.

[YouGottaBeCrazy]

your username almost sucks as much as mine wow.

[Jude]

no **** but hardware doesn't write it's own software. there would have to be software that wrote other programs.



That's what I've been saying all along


i think its pretty obvious that Artificial Intelligence will get to the point of no longer being compatible to survive in the same habitat as humanity, but the question is really will we kill ourselves in a different way before this happens?
Nuclear armageddon is almost statistically inevitable :(

[Chu]

We would have to write a program capable of creating a program.

If you are not familiar with the Church-Turing thesis (link: http://en.wikipedia.org/wiki/Church–Turing_thesis), it can basically be summarised as "Every effectively calculable function is a computable function", now you might see this as insignificant, but this brings in some interesting "limitations", if you will, about computation.

We can separate different groups of computational problem, (link: http://en.wikipedia.org/wiki/Computability_theory_(computer_science))

I'm not going to explain anything in detail, it's all available from wikipedia in a far better presentation then I could possibly write-up, so if you're really interested in this, I encourage you to read up.

One of the most interesting groups of problems, are the "decision problem"s (I think the name is pretty self-explanatory, but http://en.wikipedia.org/wiki/Decision_problem), and a sub-set of these problems are the NP-complete problems (link: http://en.wikipedia.org/wiki/NP-complete), for which, no effective solution (read: algorithm) is known, and considering the amount of research and knowledge of such problems, is doubtful to exist (by some, I would say a solution probably does exist), this leads to the "undecidable problem" (link: http://en.wikipedia.org/wiki/Undecidable_problem), "for which it is impossible to construct a single algorithm that always leads to a correct yes-or-no answer", as an example, consider the Halting problem (link: http://en.wikipedia.org/wiki/Halting_problem).

It is for the above reasons I find it very unlikely such a breakthrough will occur any time soon (maybe quantum computation has answers), and thus, we will effectively be re-creating the Halting problem (i.e. We are going to be developing genetic algorithms, for which, we have no certainty that they are correct).

This is all my own personal opinion, just what I have learnt from the computational limitations of computers in personal experience. Perhaps I am wrong, I don't know, I've never really discussed it with anyone else.

[Jude]

If you accept the computational theory of mind, you basically also accept that any problem the human mind can solve can also be solved by a man-made computer, since the mind IS a computer, and evolution has managed to produce a computer that can compute those problems.

This is the other direction I wanted to go with this thread. What do you think of this theory?

Personally I think it's the most convincing one for several reasons, but I'm no expert. For one thing, it involves breaking everything down to extremely simple operations, so simple they can be solved with a mindless switch, and combining these to create complexity. Other than this, I don't see how you can propose a "general information processing capability" or anything like that since you're essentially just rephrasing the problem.

[Chu]

It's been proven for 60 odd years the limitations of computation.

One of the true Gods of computation, Alonzo Church (http://en.wikipedia.org/wiki/Alonzo_Church).

The lambda calculus emerged in his famous 1936 paper showing the existence of an "undecidable problem".

Edit: Particularly, read about the Lambda Calculus (the concept of the Turing machine is invaluable here, my personal favourite is John Conway's "Game of Life" (link: http://www.bitstorm.org/gameoflife/ - you might not understand what's going on, so read these: http://en.wikipedia.org/wiki/Conway's_Game_of_Life, http://en.wikipedia.org/wiki/Turing_machine)), Church encoding, Mathematical/Structural Induction, and some Type Theory as it relates to Functional Programming (I recommend anything by Simon Thompson).

[Jude]

Could you sum those up? I don't have time to read the whole article right now but I'm interested

[Chu]

It's just basically a formal proof (which is my major), that there are certain problems that are computationally impossible.

The Halting problem is a very good example of this.

The Lambda Calculus is a very minimalistic theoretical programming language (1936!), the interesting thing is it is entirely adaptable for mathematics (which I think Prof. Church was trying to achieve, not the programming language), it is very simple, the main idea is combinations of single-argument anonymous functions, a generalisation of mathematical induction - structural induction, is basically used a type inference system (similar to Gentzen's rules (axions) for Natural Deduction), which basically works by pattern matching (and substitution).

Church encoding is a means of embedding data and operators in the Lambda Calculus, such that their positional placement in the equation is as general as possible, the most familiar being Church Numerals, which is a formal process Church developed for representing Natural numbers.

In typical mathematics, we'd use f(x) = x to represent a function, in the Lambda Calculus, we simple use \x. x to represent a function (the "\" symbol should be interpreted as a "Lambda", and for the Numerals, we have:

0 = \f x. x
1 = \f x. f x
2 = \f x. f (f x)
3 = \f x. f (f (f x))
n = \f x. f^n x

Now, we can use similar abstraction "simple" definitions for any computational problem, i.e. the Lambda Calculus is Turing-universal (a Turing machine, the computer you're typing on is a universal Turing-machine, thus the Lambda Calculus is equivalent in power to the computational possibilities of your computer).

Edit: Effectively, the Lambda Calculus (which is the theory behind Functional Programming languages, as opposed to Imperative Programming languages, which are more formally based on the Von Nuemann architecture), allows a very general approach to programming, it's difficult for me to explain, how you program functionally, but basically, you create very simple functions, and combine them to do very complicated problems, the idea of Lazy evaluation, and Partial evaluation are crucial to functional programming, though not all functional languages support lazy evaluation, most support partial evaluation, which effectively, in terms of the computer says "I don't care about sequence of operations anymore", which allows concurrent and distributed programming, by means of paradigm, where it is usually a language hack (in imperative programming languages), and the concept of Lazy evaluation (in Haskell for instance), allows the computation of Infinite lists, because the computer won't prepare storage space until it needs to.

[Aaron]

That's why user-acceptance testing is important prior to full-scale roll out to the end users.

[spitfirejunky]

We would have to write a program capable of creating a program.

It's called a compiler. ;)

Solving those computability problems will definitely steer us well into good AI but I don't think it's actually necessary. We need only algorithms that correct other algorithms, not ones that literally predict their behavior.

The complexity problems are definitely a big hump.

[Chu]

It's called a compiler. ;)
That's true, but a compiler is still following a pre-determined sequence of steps (before compilation process, a parser would have converted the token stream (i.e. input), into a syntax tree (errors are identified here), which is the formal description for the automated process (compilation phase) to follow).

If we use the same process to create AI, it's still being told what to do, it's not thinking for itself, and it's not going to kill us (unless programmed to do so).

For a program to create AI, it would have to write its own algorithms, with no way of knowing what it is creating is possible.

i.e. The only way machines are going to take over the world, is if someone creates the initial program that drives the machines, in which case they haven't become self-aware.

[Mr. Ron]

I saw math above and stopped reading

[spitfirejunky]

That's true, but a compiler is still following a pre-determined sequence of steps (before compilation process, a parser would have converted the token stream (i.e. input), into a syntax tree (errors are identified here), which is the formal description for the automated process (compilation phase) to follow).

If we use the same process to create AI, it's still being told what to do, it's not thinking for itself, and it's not going to kill us (unless programmed to do so).

For a program to create AI, it would have to write its own algorithms, with no way of knowing what it is creating is possible.

i.e. The only way machines are going to take over the world, is if someone creates the initial program that drives the machines, in which case they haven't become self-aware.

In the end of the day something would have to compel the machine's actions, be it a programmed impulse, intelligence, etc. If the machine is prepared properly I expect it to use the world as its hive of input, not some string like a compiler. From then on any network of self-critical algorithms can allow it to reach that idea of killing us by itself.

[Chu]

I agree that the input will be natural phenomena, not a string of data.

But I still don't see it.

Edit: It would still be told how to respond to the input, much the same we are, we just happen to have been programmed by our genes, and naturally as a result of evolution.

There is no separation of consciousness for a computer, the programmer tells it how to react.

[Mr. Ron]

that read like a noonward post

[spitfirejunky]

I agree that the input will be natural phenomena, not a string of data.

But I still don't see it.

Edit: It would still be told how to respond to the input, much the same we are, we just happen to have been programmed by our genes, and naturally as a result of evolution.

There is no separation of consciousness for a computer, the programmer tells it how to react.

If the programming is self-critical (this is very possible) it can revise its own programs and change its behavior. If this is done well enough we can get a machine that's effectively as conscious as us.

I'm just speculating though. It's easier to think about this if you assume that humans are just really really complicated machines.

[Chu]

We have examples of programs that "learn", like the guessing games that are available, and build up their knowledge, but they store their knowledge, and just match guesses against common patterns, there's really no thinking going on.

As far as learning goes, we have AI that "learns", in the sense, it changes approach because something (possibly something you have executed) has triggered it.
But even then, the AI is told how to respond to events.

We have an example of the Mars rover, when issued an order from Earth (to deploy it's arm, and drive a measuring device into the ground), it recognised performing that task would potentially damage it, so it just shut down.

That's the same thing, it was told "if something happens which puts you at risk, shutdown" (or whatever).

[spitfirejunky]

Any machine will work within the limits of what it's told to do. The question is can we tell it to do those things that make us human?

[Chu]

That is exactly the question.

For which I personally believe the answer is no.

Though, I cannot provide anything that supports it, other then my own theory, based on the previous work in a (somewhat) related field.

[Det_Nosnip]

I still don't get why AI machines will invariably destroy humanity.

[Chu]

I'm guessing because the idea they will become self-aware.

[peeted]

Im self aware and im not destroying humanity. Not right now any way. Im skeptical about AI because of the seeming incompatability of the logics they would have to use for the computers cognition. How do you teach a computer intensional concepts?

[mph4ever]

does self aware mean consciousness?

[Chu]

Yep, able to think for itself.

[mph4ever]

Yep, able to think for itself.

well computers will fail. they may very well aid someone destroying humanity. but they will never act with their own conscious intention

[Chu]

Yep, and that's pretty much my argument.

[Iscariot]

We don't even necessarily understand how our own self-awareness works, or how/where it even exists and functions.

I'm not denying that you're a really smart guy, but who's to say an advanced computer couldn't begin to think of itself as we do, but without the burden of being entirely logical and devoid of emotions?

I'm not exactly brilliant, but I heard once that whatever you can dream, you can do. And so far, as far as technology is concerned, that has held true.

[mph4ever]

Yep, and that's pretty much my argument.
i'd move that to fact

[Chu]

We don't even necessarily understand how our own self-awareness works, or how/where it even exists and functions.

I'm not denying that you're a really smart guy, but who's to say an advanced computer couldn't begin to think of itself as we do, but without the burden of being entirely logical and devoid of emotions?

I'm not exactly brilliant, but I heard once that whatever you can dream, you can do. And so far, as far as technology is concerned, that has held true.
If you read back through the thread, I have shown that there are some problems which are undecidable, for which no known solution exists.

There are many limitations to computation, as it is known at the moment, it's quite possible quantum computation is unaffected by these limitations, but I don't know.

One problem that has no effective solution, is whether a program will halt (the halting problem), now for AI to self-replicate (or rather, create something), it would create some algorithm (program) without anyway to confirm it would halt. Thus, re-creating the Halting problem.

[Iscariot]

This all goes over my head I just wanted to make a cameo in this thread. :(

[Chu]

That's all good man.

It still goes over my head, and I'm knee deep in this stuff.

[Jude]

The idea behind the computers taking over the world is that once they are both self-aware and able to improve on themselves, they can start to do so at an extremely rapid rate. Humans may master the art of genetically engineering to continue creating smarter and smarter generations of humans, but we still have to wait generations for that process. A smart enough computer program could do that in seconds. Once they start modifying themselves at a rapid rate, sooner or later one will appear that knows how to install itself onto other connected computers and spread like a virus. Eventually the combination of one that's self-aware, decides it can and wants to take over the world (or just butt**** humanity), and can spread itself, will appear just through the sheer amount of modifications that are taking place. Presumably we wouldn't have any way of knowing what the program was "thinking" until it was too late and it was crashing planes, setting off nukes, and that sort of thing.

Obviously there are lots of reasons why this would NOT happen in addition to reasons why it could but like I said this is just speculation.

Edit: It would still be told how to respond to the input, much the same we are, we just happen to have been programmed by our genes, and naturally as a result of evolution.

There is no separation of consciousness for a computer, the programmer tells it how to react.
Well, to be accurate, we don't just happen to have been programmed the way we are by our genes. Those genes are the ones that have managed to survive and thrive over generations because of the bodies and minds they built.

And we have basically no understanding of how consciousness (sentience) works. If the genes are the ones designing our hardware and software, how is that different from a programmer designing a program that can adapt to things?

(this is serious not just "armchair speculation" this time)

As to the uncomputable problems, I'm not sure what they are, but if they are the kind of thing the human mind can solve, then natural selection found a way around them, so why can't we?

And if they're the kind of thing the human mind CAN'T solve, then we don't need to worry about solving them to create an intelligent program.

Anyway, given that we have no clue how sentience works, that brings back the question of if a computer program that exactly replicated a human brain would be conscious. Is it the exact chemical reactions that give rise to consciousness? Or is it just the program itself? This is the biggest mind**** ever.

[sweboy]

Yo Jude and dudes; AI researcher and international supernerd Eliezer Yudkowsky has written extensively (like... a lot. I mean, a lot) on this very topic. He works at the Singularity Institute for Artifical Intelligence which pretty much was created solely to help avoid this scenario. He has launched the concept of "Friendly AI" and wants to make sure that the first self-improving AI that is created is "Friendly" to humans. Every argument in this thread he has covered in length. Some linx include:

http://yudkowsky.net
"Artificial Intelligence as a Positive and Negative Factor in Global Risk": http://www.singinst.org/upload/artificial-intelligence-risk.pdf
http://en.wikipedia.org/wiki/Friendly_AI
http://en.wikipedia.org/wiki/Seed_AI
"Creating Friendly AI" - book length discussion if you want to get get your hands real dirty: http://www.singinst.org/upload/CFAI//

[sweboy]

Anyhow, the risk is not so much that someone creates an AI with the goal of destroying humanity, but that an AI with a different goal destroys humanity just as a side-effect. If a company creates an AI to do a specific task (for example making paper-clips) and that AI starts to improve itself in order to more efficently perform that specific task, it will run off into an uncontrollable positive feedback loop and very quickly become ridiculously powerful but still continue to do the same task (making paper-clips). And since it's not programmed to care about humanity, it will just use its immense powers to maximize the efficiency of is paper-clip making performance, including using the earth (humans included) as a pool of material resources to achieve its goal. And before you know it the AI has transformed the whole universe into paperclips.

[siva_chair]

Well one can never have too many paperclips!

[YouGottaBeCrazy]

I liked this thread much more when it was about The Terminator.

[siva_chair]

Don't worry....He'll be back.

[YouGottaBeCrazy]

...

:p