Is artificial intelligence possible ?

[-radkeff- 0]

Will real artificial intelligence ever be possible ? By that i mean will computers ever think, learn and solve problems by themselfs ? I serously doubt it...

[Jamesia 0]

OF COURSE IT WILL BE INVENTED!

i feel that it is a certainty that it will be. And probably during our lifetimes aswell.

Havent some people in America killed themselves because they thought THE MATRIX was real?

[-radkeff- 0]

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">OF COURSE IT WILL BE INVENTED!<span id='postcolor'>

Well...im not so sure about that...computers are completly logical right ?? they do what we tell them to do...the only way i can think of is like deep blue(ibm's chess comp.) compairing maps of different stances, but even thats not real ai...i dont think its possible

[Guest]

This gets into the very grey area of what exactly is AI. Computers have to be built and programmed by humans, so at what point they extend beyond that programming to become AI (if that's how you choose to define it) is murky territory. Some say it already exists, others say it will never happen.

I hate to say it, but I think it all depends on what you define as AI as to whether it will/does exist. So long as my computer knows when I would like a cup of tea and some toast....then go and make it for me, I don't really care

[Jamesia 0]

Does the fact that when I use the keyboard the keys often get stuck down and type the wrong letters mean that my keyboard has artificial intelligence and is trying to p*ss me off?

[NurEinMensch 0]

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote (-radkeff- @ 20 May 2003,17:41)</td></tr><tr><td id="QUOTE"></span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">OF COURSE IT WILL BE INVENTED!<span id='postcolor'>

Well...im not so sure about that...computers are completly logical right ?? they do what we tell them to do...the only way i can think of is like deep blue(ibm's chess comp.) compairing maps of different stances, but even thats not real ai...i dont think its possible<span id='postcolor'>

Current computers are based on a rather simplistic set of logical rules, but so are the neurons in my brain.

It's the level of complexity and how they are linked to each other that makes us think we think.

So I'm quite optimistic something similar can be emulated on a computer.

Ok, a very complex computer and we're not going to see it anytime soon, but it will happen at some point.

[Jinef 2]

I would say yes, but we are no way near it yet.

Considering the computers that sent things into orbit are eqivalent to that in my wristwatch i think that over a space of 40 years a lot can change.

[mr. Duck 0]

Yah sure, a long time ago some people thought it was impossible to fly in the sky, let alone to space, and look what we can do now, so yeah, eventually we'll get some very complex computers with things what you could define as artificial intelligence.

[kegetys 2]

Some time ago there was an experiment where a computer program learned on its own that by flapping wings it controls it can produce lift and fly. As far as I remember, the program was made so that the "objetive" is to get as high as possible, and it was only given controls to control the wings, but no information on how to use them. Eventually it found out the optimal way to move the wings to rise up.

That of course is a very simple thing, controlled by simple logic. But it was a sort of "intelligence" anyway, the computer learned how to "please itself" on its own.

By using complex (simulated) neural networks and some sort of "this is good - this is bad" type function which tells the program, after it has done something random, that what it just did was either good or bad. After doing random things for a while it "learns" combinations of things which to do to make good things happen, and what not to do to avoid bad things from happening. In a basic level, this is how the human mind works too I guess (ie. if you touch something you never have seen before and it hurts, you learn not to touch it again.), its just so very complex that simulating it is hard, plus a human has alot of different senses which would need to be simulated too.

(Those were just my thoughts, could be something inaccurate or false )

[freddern 0]

I think that to create AI the AI has to have a goal to reach, then think of ways to reach that goal. So you must think of what are humans goal- what make us go on.- Maby making babyes and continuing our race. (If the machines do that then we may have "The Matrix")

[Necromancer- 0]

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote (Kegetys @ 20 May 2003,18:36)</td></tr><tr><td id="QUOTE">By using complex (simulated) neural networks and some sort of "this is good - this is bad" type function which tells the program, after it has done something random, that what it just did was either good or bad. After doing random things for a while it "learns" combinations of things which to do to make good things happen, and what not to do to avoid bad things from happening. In a basic level, this is how the human mind works too I guess (ie. if you touch something you never have seen before and it hurts, you learn not to touch it again.), its just so very complex that simulating it is hard, plus a human has alot of different senses which would need to be simulated too.

(Those were just my thoughts, could be something inaccurate or false )<span id='postcolor'>

Ya know, some humans can be known to do the same stupid thing again.

Like 2 weeks ago... I know that pole was electrically charged, but somehow i touched it and gave me a little shock.

[freddern 0]

Maby this should be like a poll : do you think AI is possible or not?

[Renagade 0]

Yes

Does anyone remember a game called creatures?To make things  simple they were  like cyberpets  that learned  and a few other funny things.

heres some info on them

http://www.cyberlife-research.com/company/norns/

Theres more interesting stuff if u look around the site like the robot monkey.

[Tovarish 0]

With neural networks and fuzzy logic it's deffinitely possible. Might take some time to evolve that's all. As Kegety's already mentioned, there have already been some advances in a computer's ability to teach itself new things.

[bn880 5]

I think it is very possible, not now but maybe in the future if we don't kill each other off.

You could say that already there is minimal intelligence/learning in AI systems, and I would agree.

[der bastler 0]

All you need for AI is an algorithm that let's the computer reprogram itself, i.e. rewrite it's own code, depending on the input the computer receives.

(remember no. 5?)

Neuronal Networks and genetic algorithms are a good way, but mankind should be smart enough to place a ROM-kill-switch in these AI-computers...

[Guest]

I think that you first have to define what "intelligence" is

About neural nets: I've worked a lot with them (including my MSc thesis) and I can guarantee you that they're not "intelligent". Adaptive yes, but very little beyond that. It's just a statistic algorithm for approximating and extrapolating multidimensional functions. It is different than the standard deductive form of programming, yes but there is nothing really magical about them. As far as we know now they have actually very little in common with the actual wet-ware which is far more complicated.

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">Some time ago there was an experiment where a computer program learned on its own that by flapping wings it controls it can produce lift and fly. As far as I remember, the program was made so that the "objetive" is to get as high as possible, and it was only given controls to control the wings, but no information on how to use them. Eventually it found out the optimal way to move the wings to rise up.

<span id='postcolor'>

Yes, this was done here in Sweden (I know one of the guys). The system they used is called Reinforcement Learning (implemented in OFP btw ).

The basic idea is that you give the system "rewards" for doing what you want it to do and "punishment" for when it does what you don't want it to do. The rewards and punishments are actually just scores that the program then tries to maximize by keeping track of what sequence of actions led to what states and what score it recieved for that. It's memory can be implemented through either a simple look up table or in more complex cases a neural network.

[LaPoi 0]

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote (Jamesia @ 20 May 2003,16:44)</td></tr><tr><td id="QUOTE">they thought THE MATRIX was real?<span id='postcolor'>

Well if you come from a trobbled family anythings possible. Besides, that movie had alot of deep s***. The only time I ever thought that much was with a nuclear railgun in MGS.

[bn880 5]

Well how DO you define intelligence? I would start by taking an example of an organism which is accepted to have intelligence but as little as possible.

Study that an make a definition.

Is software being run in electronic circuits capable of simulating such a process, if it is, you have intelligence...

EDIT: You can also observe human babies, however how do you know what information has been genetically encoded in the brain and what is already picked up by adaptive learning?

[bn880 5]

I wanted to add that just because we fully understand a process of logic, does not mean it can not be intelligent. Perhaps if we were capable of understanding fully the way some intelligent organisms brain works and watch it in action, we would decide, well no that is far from being intelligent.

If you look at it that way, you will only accept that some human creation is intelligent when it is outperforming human brains and or created itself.

Right now I think even small adaptive ai algorithms have a tiny bit of intelligence, created by human hands, maybe a miniscule amount compared to a birds brain, but it is still there.

[Jinef 2]

Hmm i just thought of a lethal mix - sociologists and computer programming

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">It's just a statistic algorithm for approximating and extrapolating multidimensional functions<span id='postcolor'>

Hmm, i think i have no idea what you are on about, i don't like this. I have never heard of the word 'extrapolating' before.

I think it will be a long while or just never before i understand what happens with all these software programs

[bn880 5]

Yep extrapolation is basically making a decision on past events or information. You extrapolate quite a bit in mathematics, physics etc. The meaning in it self is nothing special. The methods usually are.

EDIT: lol "extrapolatin"

[Guest]

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote (bn880 @ 20 May 2003,21:45)</td></tr><tr><td id="QUOTE">Well how DO you define intelligence?  <span id='postcolor'>

I don't

Jinef:

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE"></span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">It's just a statistic algorithm for approximating and extrapolating multidimensional functions<span id='postcolor'>

Hmm, i think i have no idea what you are on about, i don't like this. I have never heard of the word 'extrapolating' before.<span id='postcolor'>

It's really not difficult at all.

You have a set of points and you try to fit a function to them to predict other points.

Example 1: Linear prediction (i.e we fit a line to a set of points). The red dots represent the points (samples, measurements, whatever) that we have  y(x). The blue line is the function we are trying to fit and the blue dot represents the point which we predict.

Example 2: Non-linear prediction (i.e we fit an arbitrary function to the points - it could be anything - a polynomial, or a sine or a custom constructed shape):

Now for the multidimensional part - not difficult. The graphs you see here are 2-D  y as a function of x ( y(x) ). If we add a dimension we would get a 3-D graph (z (x,y). You can go on increasing the dimensions as long as you wish.

Multidimensional just refers to that it is a function of several variables.

Example:

area of rectangle = f(side a, side b)

Are you with me so far?

A more general situation is where the function returns several values, not just one.

Example: We have a pool table and we wish to express the position of the ball as a function of the angle and the strength you hit it with:

(x,y) = f(angle,strength)

Ok, that was all the theory

A neural network simply constructs a function that fits to the points you give it.

So what's the big deal then? Simply that it constructs that function automatically and that it can recreate very complex functions.

You can have a neural network learn to drive a car by showing it how it is done while providing it with relevant sensory inputs. You could for instance take a picture each second of the road ahead and record the position of the throttle and the steering wheel:

(steering,throttle) = f(picture)

After you trained the network you can let it drive the car by itself. You feed it pictures and it will output signals for steering and throttle

[bn880 5]

</span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote (denoir @ 20 May 2003,16:47)</td></tr><tr><td id="QUOTE">You can have a neural network learn to drive a car by showing it how it is done while providing it with relevant sensory inputs. You could for instance take a picture each second of the road ahead and record the position of the throttle and the steering wheel:

(steering,throttle) = f(picture)

After you trained the network you can let it drive the car by itself. You feed it pictures and it will output signals for steering and throttle <span id='postcolor'>

Hint Hint BIS.

[IsthatyouJohnWayne 0]

One 'future science' book i read recently on this and other subjects (explaining things simply) that i found to be quite good is 'Visions' by Japanese physicist Michio Kaku. Its already a few years out of date but considering that its predicting the next 100+ years that aint too bad. (actually id advise someone whos interested to find a newer and more AI focused book).

I know nothing about this subject but there are many branches of and approaches to AI research. Some attempting to solve problems in ways that contradict other contemporary attempts.

My own view (knowing nothing of AI) is that as long as an apparent intelligence is a direct result of artifice it is not truly anything but a symptom of human intelligence. Real intelligence (denoting an independance of thought) must be an anomaly.