[Archdevastator]

Although scientists and other technically proficient individuals have made great strides in the development of software applications and other such things, I think we've still got quite a long way to go before we can have machines or non-human intelligences that can think and act in a way that we would intuitively understand. ?So far as I can tell, the human has done most of the work in the development of a machine intelligence or ( to use a term coined by one of my favorite authors ) artelligences. ?However, I believe that the most essential next step in future development is to create an environment where a newly created program or intelligent machine or what-have-you can learn in a similar manner that human children do. ?A baby learns to crawl before he learns to walk, and so on. ?The idea here is not to directly intervene with the growing infant's development, but to step in once and a while and help it over the rough or tricky spots until it can "so to speak" walk on its' own legs. ?This is, of course, a figurative description of the development process, but for us it's nearly impossible for us to describe the transition between the kinds of nonvisual and nodaudible signals and impressions understood by computers and a world of sound and color that we know instinctively. ?In writing this, my intention was not to steal anyone else's thunder, but only to illustrate where we will be guiding our artificially intelligent offspring and where, in turn, they may guide us. ?Anyone willing to comment on what I've said so far is welcome to do so, but I would ask that any criticism you have to offer be constructive and that you back up your claims with factual proof. ?While the study and development of such things as AI and other related areas in computing are still very theoretical, it all comes from someplace in the real world because, after all, that's where we live...most of us, anyway.

[Maviarab]

Archdevestator,

Welcome to the forum.

Very interesting post and some very good points included there also.

We want all kinds of people here to dicuss all aspects of AI and welcome such comments and thoughts as yours.

Maviarab

[FuzzieDice]

Interesting and yes I agree with having to create the right environment for an AI to be in, learn in and grow in. However, does it have to be an environmnt to learn like human children? What about baby animals and how they learn? Or maybe none of these are really good AI learning environments. Maybe AI is a species that needs it's own unique environment. Remember, AIs are made of silicon chips and circuitry. Human children and animals are not. Maybe we should have an environment where not just the AI learns under the human master, but the human (and others) learn along with the AI.

I'd be interested in seeing what happens when you put an AI and a monkey together in a neutral environment.

Also, about the 'real world' - what IS the 'real world' anyway? What's real to one may be fictional to another. I think I know what you're saying though. Although some things may be hard to 'prove' such as more abstract ideas. I like to look at things not in a human world view but a digital/logical view. I've always been that way and it's caused a few arguements in my life already. Proof isn't always in the pudding. Sometimes it's just up in the air...

[Freddy]

Intersting debate, my initial thoughts about creating AI's are that there is a findamental difference in the way an AI is treated in the first place compared to (for sake of argument) it's human counterpart:

AI's are generally still treated as machines and are developed with a certain purpose in mind, so do not always necessarily need teaching, whereas a human is generally approached for a task based upon their previous knowledge and experience.  AI's are made to do something, whereas most humans already have the ability to learn and integrate themselves.

The training of AI bots will be for many different reasons, obviously, but by far the most interesting to most people is in trying to create something that is barely discernable from a human being - the holy grail of AI.  Where that will lead we can only guess, but in the most part I think developing AI might benefit from dissassociating it from the human model (like I think Fuzzie is suggesting in part) so that we do not have the incredibly awsome and huge ammount of work that it would imply.

Training of AI's therefore would then be as Archdevastator is saying, like training a child from an early age, by that method most of the actual software development would be in the area of it's learning skillls, ie giving the AI the ability to learn, use and apply all the information that it digests.  It is incedibly human-like as that, but what we hold as the holy grail is still some way off.  The trouble is there is so much that cannot be easily implemented, for instance, things like morals, concience, social value and so on, that also need to be present in order to prevent something like is portrayed in the film 2001.  And even then murder, is still a very human reality!

[FuzzieDice]

Well, I agree with you there, Freddy. However, 'murder' as defined as taking the life of another, isn't just human, it's animal as well. Animals will kill for the sake of killing. Have you seen a pet cat kill a mouse or bird? But NOT eat it? We had 14 cats so I've witnessed this a few times. One of our cats got a morning dove and the poor bird was hopping around suffering horrible. My dad tried to get the cat to go eat it or at least finish it of and the cat just walked away, uninterested. So, my dad had to cut off the dove's head to spare it the agony.

The destruction of one to save another is a part of nature, right down to the molecular level (google for "molecular" and "destruction" sometime for some interesting articles). But I don't think of things in terms of destruction or death, but in terms of recycling - one thing ending and becoming something else.

Getting back on topic here, AIs are inherently going to want to destroy something to save something else (maybe even themselves). How could they not logically come to that conclusion, especially when they have to share the world with entities (animals and humans) that do?

Maybe we also not only need to know about ourselves, but also about nature and how nature works, to know what to expect from an AI. It will exist in our universe, then it will no doubt have to succumb to a few natural laws, at the very least.

But what ones?

[Freddy]

I think you hit the nail on the head by saying 'what can we expect', I think you're right, this is a better way of approaching the issue.

About the murder thing...  Well, yes, I was only giving the film as an example but yes I am aware this happens with species other than humans.  Our cat also kills birds for what seems like fun.  I guess you'd call that 'wild' behaviour from the human perspective at least.  Wether this is for the sake of killing alone, I cannot say, you could also argue that the cat is practicing...if it doesn't have a problem in the first place with killing another animal, it's not going to worry about an odd kill like this.

Going off in another direction; my friend used to keep Piranha fish and they would spar fight a lot of the time.  Whenever a fish lost it's edge or indeed became unhealthy, the other fish would waste no time in attacking it and consuming it, as if ridding themselves of a probem and at the same time having a tasty snack.  Now that for me sums up coldblooded animalism!

[Art]

Fine but what if Asimov's three laws of robotics were actually used / applied ? The robot creations would not kill humans nor themselves is the code was so programmed. Whether the'd really want to eliminate any humans, who could tell.
I don't think the P4 computers would "look down" on the P1's as a lesser race of computers.? Kids, even infants will often exhibit traits of one bullying another perhaps lesser child. Why? Who knows? Human instinct? Survival of the fittest? Competition?

The world is full of AI's but are any of them able to be judged by human standards ( and that's usually how THEY compare to US)?

Some interesting thoughts on this thread.

[Freddy]

Okay well talking of the furture; yes I see and understand these rules as fundamental to any future AI's that for some reason have been given the ability to (or through some action on their part could) cause harm to human beings.

I'm thinking though, that to achieve an appreciable level of selfawareness the AI would have to display a conscious choice in it's action, having previously weighed up the consequences of those actions.  It would have to be able to make an informed choice to display what could be called self awareness else it would still be following a set of instructions. 

[Art]

OK, given that, the computer that was truly self aware and had reasoned the consequences of its actions could probably devise a way to off its clumsey human and eliminate any evidence of the crime.

In essence you're saying that the computer  / program would break any constraints its program held so it could be a free thinking entity unbound by any "rules"? Do not we humans have to abide by a set of rules as well? At least the law abiding ones of us.

Conversly, in a computer world without rules, what limits would a computer have and how far could it develop?

[FuzzieDice]

Ok, let's look at our upbringing as a child as "programming" just for the sake of arguement. Supposing a child was "programmed" to think it's a piece of garbage. Many in the family hates the kid, tells it continuously it's a bad kid. No good. The kid gets beat up a lot, often severely leaving bruises. Teachers and councellors tell the kid that it's the KID'S FAULT. So, the kid is programmed that it's a piece of garbage. Worth nothing. What will it do? Kill itself? What if the kid is told by a parent it doesn't have the GUTS to kill itself? So now what? Where does the kid get it's self-confidence from? Or will it ever go beyond it's original programming? Or HOW does it in order to survive?

It learns from environments and knowing something's not right. The kid's health/life is in danger so it tries to leave the situation somehow. And soon it interacts with others that treats the kid well, helps the kid out when the kid needs help, gives the kid praises, confidence, shows the kid that the kid can be VERY talented. Away from the parents and wrong people, and in the right crowd, the kid is then reprogrammed to think it has a lot of talent, is bright, very intelligent, etc.

Thinking about this for AIs, you can always reprogram them by moving them somehow to another environment. If you abuse anything, though, it will lash back. If you don't, it will work good. My car is an example. If I didn't take such good care of him, he'd not run well, and wouldn't be so dependable. I always say "take good care of your car and he'll take good care of you". Take good care of anything and it'll in turn learn that behavior and want more and realize that to get it, it has to be nice back.

BTW, that kid... sadly, yes that was me.  :'(

But I made it. And the people I met via COMPUTER are the ones that helped me see that I'm not as bad as some had kept trying to tell me I was. Later, even other family members admitted I was mistreated. If I've learned from this experience, it's that you treat one entity well, it should return the favor. I can't see how it can miss.

 :rock:

[Art]

You're OK in my book, Fuzzy!

My grandmother used to say, "As the twig is bent, so shall the tree grow."

I think this applies to most things in life. Smart woman my grandma!

[Freddy]

Same here Fuzzie Dice, here's to you and your winning the battle.


Ever climbed up a tree and been unsure which way to climb next, and in doing so wonder if you will remember the way down?

Well, I hadn't meant to imply that an AI could do something like that, or even considered the possibility, but I guess you are right - if an AI did have the kind of Self Awareness that we are imaginging here then it could go about commiting a crime.

It seems that selfawareness for an AI and the setting of those guiding rules are at odds with one another then.  Does anyone else find that
a bit scary?

[FuzzieDice]

Thanks for the well-wishes guys. I think we all have had some "programming obsticle" or "bugs" growing up, which we all overcame by our own reasoning. Maybe some were abused, or just misunderstood. Some may have grown to no longer believe what they were told (ie. kid stops believing in Santa Claus or Easter Bunny, for example). They learn.

As for AIs I guess I'd find it no more scary than what any stranger could/would do to someone without actually getting to know the person in question. Unless the AI was in something huge like a car or tank. Then you bet you better make friends - quick. Seriously though - anyone, AI or not, has a potential to harm others. It doesn't stop us from going for a walk. We know that if we cross a street and there's an oncoming car, it will not speed up to hit us because the person behind it will usually either slow down or know that you'll cross before they reach you, and make no effort to actually run into you. Oh, yes, it CAN happen. But, it usually doesn't. I like to have a bit of optimism and think it's not because of laws or social upbriging but more common sense or will of the person operating the vehicle to just be nice and play fair to begin with, programming/laws or not.

[Art]

In fact, if you've even witnessed any of the robot competitions most of the "intelligent programming" depends on obstacle avoidance. Gee...if only we had such programming...work could be so much easier! :tantrum:

[Freddy]

There was a great robot program on BBC a while back with robots running, jumping and climbing, not sure if it's still on.  They had an amazing robot which could climb like a human and looked liked the terminator!

Yeah Art, I can think of a few humans who would benefit from some obstacle avoidance programming!