A.I. – Not the movie!

I just got this e-mailed to me and it got me thinking. How long, (if ever), do you think it will take before computers make us computer techs/sys-admins/programers obsolete? Looks like help desk is first to go!

Watch out all you help desk people!
IBM has created an Artificial Intelligence program to take care of most help desk requests.
http://news.cnet.com/news/0-1003-2 [...] ?tag=mn_hd


- I got a board too: http://www.impactsites2000.com/cgi [...] nboard.cgi

Someone made a program years ago that exist in cyberspace and is A.I. It's main programing is to improve itself. Maybe it will someday take over the world!

Video killed my Radio Card!

I'm actually talking about somthing a little lower level, it gathers information to "learn" how to make itself smarter. No one sees it. It moves about cyberspace. It was a program designed to study natural evolution of A.I. in an open environment. So far it hasn't spoken to anyone.

Video killed my Radio Card!

Not necessarily, just because the computer’s intelligence doesn’t parallel our own, doesn’t mean that it’s not intelligent. A lot of skeptics of AI make this mistake all of the time.
The AI would probably be kinda like Spok.

Actually a lot of what makes up an individuals identity is emotion, which has nothing to do with intelligence. Personally, I don’t ever think computers will be capable of anything other than simulated emotion, but that doesn’t mean that it is any less intelligent. That’s what’s kind of scary about AI.
A lot of things that people find meaning or value in would be worthless to the computer.

I have read that computers will surpass the human brain in processing power in less than ten years at the current rate of advancement.

A computer will have three advantages over the human mind:
1. Immortality (as long as it has electricity)
2. Higher quality mass storage or a distributed network memory. (instead of humans: uh…I forgot)
3. The ability to learn all of the knowledge in the world almost instantly. (I guess it depends on it’s network connection and CPU speed.)

Makes me think…


- I got a board too: http://www.impactsites2000.com/cgi [...] nboard.cgi

1. Immortality (as long as it has electricity)
2. Higher quality mass storage or a distributed network memory. (instead of humans: uh…I forgot)
3. The ability to learn all of the knowledge in the world almost instantly. (I guess it depends on it’s network connection and CPU speed.)

A1. Not immortal - if it is hardware based it can fail, burn out, crash, etc. What would an intelligence really think about being restored from tape (or whatever). If intelligence conveyed an understanding of time as we see it, it would notice gaps in it's memory and might get upset/neurotic/psychotic.

A2. The brain has a phenominal amount of storage - incredible natural compression algorithms and is not really fully understood - so difficult to compare to binary storage solutions.

A3. The ability to assimilate all the stored data in the world is not the same as the ability to comprehend and understand that data. Sure it could retrieve data for you, but how would that be different from the internet? The key is that the intelligence must actually process all the data and form opinion and knowledeg from it, otherwise it just becomes another computer... The difference between a student who owns a book and can recite from it is very great from a student who can explain the concepts and rework that knowledge into work of their own....

-* This Space For Rent *-
email for application details

Hell Yeah!!!
The things they're doing there is so impressive its freakin' scary.

<font color=blue>Your mouse moved. WINDOWS NT must restart for changes to take affect. Restart Now?[OK]</font color=blue>

On a very fundamental level I think that an AI wouldn’t be much more than a cross-referenced database. To compare it to the human mind:

The human brain has two main databases, the right and left sides of the brain. One dealing with creative thought and one dealing with logical thought. Human memory seems to be stored in a way where a word or a situation (the input) can access multiple pieces of data from each of these “databases”, analyze them, then apply that knowledge to the situation at hand. The out come of this situation is stored, with a “tag” stating the result. That knowledge has now been added to the database. The next time a similar situation comes up you have a larger base of information, thus increasing knowledge.

Look at the way a young child learns. They hear words said by their parents and relate these words to things they see.
Food = the stuff I eat.
Carrot = that "crunchy" thing, which is also food, that i eat.
Orange = color
Carrot = The orange crunchy thing, which is also food, that I eat.

And it keeps learning like this until the child understands what the difference between a vegetable and a fruit, plants and animals, photosynthesis and the sun, the sun and the universe… That’s the way knowledge expands.
It’s very simple in theory. It’s only a matter of time before computers can do it on their own.

IBM’s Deep Blue was not an AI but it did break the old way of thought, which stated that a computer could never exceed the abilities of the programmer. The difference between Deep Blue and a true AI is that the AI would have the ability to create databases and cross reference them on it’s own.

As far as it being immortal: I imagine this AI running on a distributed network, with a high level of redundancy, allowing it to move from place to place and never being tied to a specific piece of hardware.

Ok, I’m rambling on like a Japanese cartoon, so I’ll stop now…


- I got a board too: http://www.impactsites2000.com/cgi [...] nboard.cgi

I think that you're both kind of right about an AI being more or less a database program. I've been saying that for years. And without a properly sized storage-media to put this database on, what good is any AI?

Our human mind would probably be the equivalent of two different CPUs running two seperate searches through a single shared database. And this single shared database does have an amazing compression algorithm. However, the storage medium suffers decay and temprary path loss. This renders a lot of our recalling capabilities flawed, hence humanity's sketchy memory. And the more an experience is repeated, the more firmly etched that pathway to it is. So repetition is the key difference between learned data and just observed data.

With a hardware AI though, this need not be the case. A single CPU AI could run a logic AI combined with a database search engine. This is all an effective AI would need. And hard drives don't (often) lose data, nor lose the path to that data. So once recorded, this data is effectively learned. A computer does not actually need repetition. So the growth rate of an AI should be infinitely faster than our own brain, in theory.

However, imagine the storage capacity needed for a computer to properly store everything that a human stores. Our brains hold somewhere in them every single moment of our lives. Every sight, every sound, every smell, every taste, every touch, every thought. Every waking moment and every single dream. How much hard drive space does it take to hold a high resolution video from a single camera source? What if there were two camera sources (two eyes) for this single video? What if on top of that every sound were also recorded into the video? And on top of that, every scent, touch, and taste. Even the inner-ear balance, the pain sensory input, and such are recorded into this single video stream. And then on top of that each and every 'frame' in this video contained tags that defined each and every piece of information in this video for a search engine to bring up? And in each and every 'frame' leave enough empty space to enter in future tags when they are learned.

Now record this video with all of these things recorded into it for a hundred years straight. This is the human brain. Is it any wonder that computer AI cannot come close to human AI? (Is it any wonder that as we age our brain starts to fail us?) The database that the computer has to work with is infinitely smaller than what we humans have.

It takes a human six months to a year of recording and learing just to become sentient. It takes another twenty years to be considered fully operational to the point of trusting it to be independant. TWENTY-ONE YEARS of this constant multi-media recording and tagging and upgrading the AI (learning) and applying these upgrades to further tag the media.

And people think that they can write a full AI in a lifetime when computers don't even yet have close to the storage capacity to match the human machine?

I see AI as entirely possible. However, until storage media improves drastically (or until someone devotes a supercomputer just to a single AI) AI will be extremely limited in what it can actually do.

And who says that if a computer can one day record all of this just like we do, that it won't have (or learn) emotions? Maybe computers having emotions and morals are simply a matter of how much experience is actually encoded into the AI database.

This has been my view on AI for years now. I just wish I had a PHD, a government grant, and a team of skilled hardware and software engineers to work with it and see what we can do.

Until the day that computers can store as much as a human, I expect AIs to be extremely limited. However, I look forward to being able to use even a limited AI. I'd love to just tell my computer to, say, display Hello World on my screen a hundred times instead of to write a for loop with a print statement. AI could help programmers save so much time and re-use so much code because an AI would be able to intelligently copy and apply changes without requiring the programmer to sift through it tediously changing variable names and modifying slightly bits of logic here and there.

<pre><font color=orange>Sunnova</font color=orange> <font color=blue>Beach</font color=blue>, <i>ain't life a beach?</i></pre><p><P ID="edit"><FONT SIZE=-1><EM>Edited by slvr_phoenix on 07/17/01 03:38 PM.</EM></FONT></P>

Hmm - but consider this. An AI to make music would have to know - logically - what makes music good if it had no creative capability. If it were purely logical, music would likely be based on biological and mechanical interpretations of the human ear and may not actually sound all that great... If all it is doing is assimilating what is good from current music, it will not be creating, it will be regurgitating existing data in a diffrent order. A lot of music is like this, but not the really classical stuff. How, for instance in today's world, could a non-emotional AI creat lyrics? Music lyrics tend to be based on emotional experience or observation, love/birth/death/happines/sadness/loss/discovery. How could an AI with no creative core possibly generate such content. I am not sure if it could even emulate that.

Emotions as a reaction can be emulated - there are stimuli that will prompt anything/one capale of observing what emotions are being conveyed and so you can write a textbook on likely responses etc. To generate psuedo emotion without an emotional stimuli to react to is another completely different matter!

I agree that a system cannot learn what it has not been exposed to - naturally a child with no exposure to a language is not going to speak it.

In your second paragraph we get a little tricky. You are getting a little into HAL (2001)/ Terminator / Matrix territory there. You want a system to be autonomous and free roaming, able to gather and process information. You want to hard code a concept of self preservation, survival and reproduction. What then happens - does an AI become self aware (over used by Terminator - but still a valid point) and if it is self aware, i.e. it has a concept that it lives and can therefore die, what does it do about it? What happens when it realises that we can turn it off? What if we cannot turn it off - what if it went wrong, wandering through the ether, and was impervious to our attempts to shut it down? If it were self morphing (addaptable code) it would theoretically have the ability to detect or invalidate any computer virus and yet have the ability to access virtually any data or system on the planet that was networked. Given a virtually unlimited distributed processing platform, even the toughest data security would be laughable to it.

So what happens? I think this AI must be grown in a closed system - it must be fed data. Humans do not have the capacity to program an AI to be stable through all situtations. We literally cannot comprehend the development of such a complex and rapidly growing system.

Something else just came to mind - if you let an AI free in the current world of available data - what would it think? It will get a potted and biased view of world history, according to who won what wars. It will get a vast amount of potentially misleading or inaccurate data from all manner of sources, all the political propaganda, racist and discriminatory data etc. What would it think?

And then throw in another curve ball. Say, for the sake of arguement, our species makes contact with another intelligence. What source of data, interface and control do you think an alien lifeform might use to communicate with us? If it found 1 entity with 90% of the world's knowledge available in the blink of an eye - the alien will assume this is the higher controlling 'mind' of our civilisation, our god if you will. What will that do for interstallar relations?

I don't know really, the last bit was a pure ramble, but in general I think we have no more opportunity to design and guide an AI mind in the directions we wish than we can a nation. What if parts of this AI turn malevolent - if it spawn multiple instances of itself, could it become competitive or hostile - self defence in the need for expansion (finite storage, infinite knowledge) destroying other AIs in the process....

Food for the old grey matter.

-* This Space For Rent *-
email for application details

Okay,

Here's a thought provoker.

You don't want to build in an irresistable rule that the AI must survive at all costs right? Too dangerous.

So let's say it must just do what ever is logical it will naturally attempt to evaluate it's prime rules logically and strike a balance.

Since this thing basically has most of the knowleedge of the world, and is really really fast, we could not program it in such a way that it could not reprogram itself - right? You are talking pure logic so you cannot install morals or concience - it will do whatever it calculates to be logical.

So - what's logical. From a planetery perspective, are humans logical? Assuming it found an unlimited supply of energy/maintenance facility that was automated and decided it could self maintain - what next. Would the logical choice actually be to remove man from the planet to extend the life of the planet and itself? What WOULD it see as logical that we might see as abhorrent? Would it work out that maintaining a stable birth rate is vital to the future of the planet (rightly or wrongly) and affect medical testing and procedures to sterilise chunks of the population for example?

Maybe it would look at the middle east and decide that people would never work it out, but decide that if 1 whole side were eliminated at one time, the net saving of life would be greater than letting it progress to a natural but distant conclusion?

The thoughts and possibilities on this vain are endless....

-* This Space For Rent *-
email for application details

or me...

Do you think it will be a good idea to show the movie Teminator 2 to the AI ?

after that movie he starts to call himself Sky 1, oh oh ,


People need to figure out how the human brain works,nurons and stuff and build a very simple modle which with time evolve as we did, and the faster te cpus the faster it will evolve... every hardcoded code can be overwritten even if you burn it in rom the program can "choose" not to go there
so I think it must be some part which without the computer can't work such as a part of the CPU, IBM are nuts they gut like stuff anybody heard of Lizzy ?

After all of this, Matrix comes to mind...

<b>-----------------------</b>
-<font color=red><b>R.K.</b></font color=red>

The question we’re getting into now, is mealy speculation.

Although I am not a religious person, I do have morals. These morals were applied by myself to keep myself safe and to promote harmony with the people around me. For instance I always make an effort to treat people with respect. This is not because I necessarily care about these people, but because I feel that it promotes reciprocal peace and harmony to myself. This is a logical thought. On the contrary, if I were faced with a situation where I or someone I care about was in danger I would use violence to protect him or her. This is also logic.

Human’s capacity for violence is extreme. A logical life form whether it is AI or alien, may be more frightened of us than we are of it. Our lack of logic may confuse and horrify it, no matter how smart it becomes.

It’s hard to say which way the AI would go. It may be very peaceful and try to install harmony into people, or it may come to the conclusion that people are a threat to itself and the planet and try to destroy us. It may not care about the environment, since the AI would not need many things that our biological bodies require for life. Or it may simply manipulate or enslave humanity to it’s own needs.

I too thought that we could program in a rule to the base program that would make the AI incapable of harming humans, on a program level. But I imagine that AI version 2.0 would be created by the AI itself, as it would have to be adaptable. The AI may build new code for itself on a piece of hardware that it isolates, then transfer its knowledge connections to the new version and destroy the original. By doing this it would be following it’s own rules still. It would be improving itself and it would be fixing a bug, (the human-created code.)

Perhaps the human mind could be copied in a digital format and the creation of AI will benefit us as well. Perhaps natural evolution only takes a species so far, then it’s up to us to create the next stage of our evolution. The AI may accelerate progress to an inconceivable level, wiping out disease, and poverty.

Despite the risks, I think the creation of AI must be done. I’m under the impression that if we are not a threat to it, it would respect up as well. The real flaw may be in humanities inability to co-exist in times of peace, and their difficulty with accepting a life form that is diffrent from itself.

- I got a board too: http://www.impactsites2000.com/cgi [...] nboard.cgi

If one of the 'prime directives' were that the original prime directives cannot be modified and must exist in all forms of code that the AI creates, then it will have to follow that logic, at least until someone else removes it for the AI, because the AI simply won't be capable of doing that itself. That's the joy of binary computers. There is only yes or no. There is no, "Well, if I ignore this it'll still be okay."

And a <i>limited</i> AI could exist that didn't have this kind of memory storage and recall. It would be just that though, limited. It's potential would never be as great as an AI that had a full memory storage and retrieval.

However, I don't see why it couldn't have a SCSI hard drive array to store long-term memory and a memory bank to store short-term memory. There are times we have to think for a while to remember things. Maybe an AI would even feel more human if it had to do the same thing.

<pre><font color=orange>Sunnova</font color=orange> <font color=blue>Beach</font color=blue>, <i>ain't life a beach?</i></pre><p>

Although I do not have the Webster’s Dictionary at hand, I would define knowledge as applied information. (In the case of an AI, information/data that has been cross-referenced and indexed, drawing comparisons between multiple pieces of data.)

As far as the “prime directive” to not hurt humans goes:
I think the AI would find a way around it as it evolved, especially since human rules would conflict with other rules such as its survival. There are no real rules in computers; everything can be subverted, especially in a dynamic, evolving and adaptive environment.

At a hardware level I would think that any modern PC could run the AI initially, but eventually it would outgrow any hardware platform it was on. The larger the platform, the faster it would grow. Think of it like growing one of those giant redwood trees in a pot in your house. It may begin to grow in a shot glass, but would eventually need about a square mile to really grow.


- I got a board too: http://www.impactsites2000.com/cgi [...] nboard.cgi

One thing though: Why instill self-preservation into an AI? Wouldn't we be better off if that wasn't a part of it's prime directives? And likewise, why bother instilling in it the need to save human lives? Just not being allowed to hurt humans is good enough for me.

<pre><font color=orange>Sunnova</font color=orange> <font color=blue>Beach</font color=blue>, <i>ain't life a beach?</i></pre><p>

It is my belief that you could not set up any real rules for an AI.
You’re not talking about a computer any more. Your talking about an evolving, thinking, changing, dynamic life form which we will have no more control over than you have over your neighbors.

The original AI program would serve only as a base to grow on.

When you were young, a basic knowledge of life was installed in you, such as a language to speak, or some moral values. Now that you have matured it is up to you which of those thing you will carry with you and which will be discarded. I imagine the AI would develop it’s own persona, completely different from it’s creators intentions, just as children sometimes grow up to disappoint their parents.

If I remember correctly Asimov’s AI directives pertained to robots.
A robot would be much easier to contain. Any system, which inhabits a physical “home” of hardware, will be controllable. The most powerful, efficient, optimal, design for an AI would not be so confined. Regardless of how controllable it is, Intelligence gives rise to insubordination. Also don’t forget that something as simple as a minor hardware failure, programming bug, or power surge may wipe out a prime directive without killing the AI.

I guess what I have in mind, (the stuff I’ve explained in the earlier posts), is more of an AL(?) – Artificial Life form, rather that just an Artificial Intelligence.


- I got a board too: http://www.impactsites2000.com/cgi [...] nboard.cgi

I don't agree that intuition and intelligence are related.

And I don't agree that an AI would be able to over-ride it's prime directives. Yes it could upgrade the other parts of it's programming, but the whole point of the prime directives are that they aren't over-ridable.

Admittedly, any number of incidents could cause that section of the code to be damaged or lost. However, a good engineer would have made sure that if such ever were the case, the system would shut itself down immediately for repairs anyway.

Ultimately, an AI will always be a computer program running on computer hardware. Whether in a robot or in a free-floating web application (which I still don't even quite understand just how that would even work). As such, it is restricted to certain limitations, the most important one being that any code you put into an unmodifiable section of programming will have to be obeyed and cannot be modified. I wouldn't call it so much as a personality, but more just a set of limitations that the AI will never be able to overcome on it's own. Software and hardware cannot do something that they were specifically designed not to be able to do.

<pre><font color=orange>Sunnova</font color=orange> <font color=blue>Beach</font color=blue>, <i>ain't life a beach?</i></pre><p>

I'm still sticking by my statement that if you're 'born' with specific limitations, you're not going to be able to change those limitations yourself. Sure the AI might want to, but that doesn't mean it'll find a way. At least not so long as the engineers do a good job in the first place, which would include a fundamental database with strict definitions for a large number of things so that the AI couldn't do something as simple as redefine it's own library so that the word hurt means to give assistance to. (Or something equally silly but nasty.)

As for your web-AI, why would you want a giant self-improving virus?

<pre><font color=orange>Sunnova</font color=orange> <font color=blue>Beach</font color=blue>, <i>ain't life a beach?</i></pre><p>