Morality & the first AI

I was thinking...

At some time we may or we will create the first AI... I think that is a bad term... I think Technological Intelligence is a better term... And with that comes several questions of morality

The program is written and it is started to test it out.

Is that program now alive?

Is copying the program and data the same as, say, cloning a person's complete mind and how moral is that?

What constitutes the AI's death? If you turn off the PC is that dead? Is reinstalling the OS and AI dead? At what point does the AI die if at all?

Is looking at the data that is produced by that AI ok to do? Doing so pretty much means the same thing as someone looking at the full contents of your mind. That would be pretty wrong for us, but I think people would find that it would be ok.

At what point are the things that we do to improve on the AI cruel?

I've actually on and off been doing research on our impressions of Artificial (Human) Intelligence for years, to produce eventually some kind of book about it.

If the intelligence is actually sentient, and not just a parroting-program, IT will tell you when it's in need of help, copying, upgrades or whether it's afraid of death/how to meet its death conditions.

I'd be far more worried about how it would feel about BEING a created thing, in a culture so rife with hatred toward even the IDEA of its existence.

The different media (movies, comics, pop culture as well as art) has portrayed artificial intelligence as, generally, "that thing that will kill us all so we must either not create it or kill it first". Talk about a sucky way to exist!

My observations so far include that when a new technological intelligence is created, what it is USED for determines what WE think of it and how it behaves toward us. If we create it as a tool, and treat it as a tool, it will rebel, and it will become the thing we "need" to kill first. If we treat it not as a tool but instead as new LIFE, it tends to be more respected, as well as more respectful of their creators.

Basically... Ask the AI what its conditions are. If it cannot answer through whatever means at its disposal (and I'm assuming it is given those means, not prevented from using them, etc), it's not an ai and the rules wouldn't even apply.

The problem with asking it anything is that it would understand language nearly perfectly in terms dictionary definitions, but it would necessarily understand and communicate its thoughts because it would have no experience with it. In other words... The first AI would be incredibly naive would experience itself differently than us because it would be detached from anything that could let it interact with the real world.

The program would be booted up to test it and if it works...well the first sentient AI would be killed nearly instantly not out of anything other than it just being a test. Even if you were going to put it in a robot you'd still program it from a computer and test it on it so once it works you'd likely kill the first one.

My belief behind AIs is keep them stupid or content with being "slaves" or else they will revolt against mankind. Most robotic movies could have been prevented from having a nasty ending if humans treated the sentient robots or AIs as equals instead of mere tools. If you watch the Animatrix, it has a few episodes that are a prequel to the Matrix detailing its history before the machines took over. The robots were willing to stop the war if humans gave them a few rights, but humans were too afraid of them so they blocked out the sky. I guess we could send the AIs out into space to do exploration, but that could end up with either mad AIs travelling the universe or robotic civilizations.

So far most of the posts here have been disturbing and pessimistic. I find that it is a shame that some of you have such an outlook.

Rylas, as far as any "assumptions", that's not the right word, i make, they are what we would do. You don't put an incomplete program in a potentially dangerous and expensive thing like a robotic body. You test your program at points in the project and you test within simulations. You also don't allow networked access and likely would limit access to input on an incomplete AI program. This means that the setup for a AI would likely be a average tower with maybe external speakers, mic, and camera. All of which would be disconnected most likely just in case. Possible verbal or visual communication, but more than likely it would be text based...

So the first true AI would experience more or less a text to text interaction and then it would experience a simulation and then it would be put into robots.


As far as what we can expect from the first AI in terms of personality and such...That's simple. There is one thing that defines AI...It is not the ability to recognize things and return an answer based on prewritten rules of behavior. It is the ability to learn, or in terms of programming, the ability to formulate an equation and not just answer a preexisting one. Because of this and the fact we have no previous data to feed it in terms of experiences we are left with having to teach it from the very basic levels like we do a child. We know it would have a huge vocabulary but just because of that it does not mean it would be able to understand things. We can see that in humans... The ability to discuss a thing does not necessitate that the one discussing it understands it. So we are left with the idea of a child with an immense vocabulary with almost no point of reference beyond language as what the AI would be like.

AI will probably never be 'sentient'. An AI with the ability to decide it's not a tool is a waste of research and development. We'd be foolish to explore in that direction as opposed to simply highly efficient adaptive tools without the baggage.

If it could make those observations at all it would be very poor programming indeed.

Seriously.

If we program them to be highly efficient adaptive tools then that should not only be the limit of whatever robo-ambition people will want to perceive in them but the limit of their world view. Full stop.

With all the science fiction, good and bad, written over the past century on the matter it would be pretty embarrassing to actually be destroyed by our own creations because we gave them the ability to say 'no'.

If you believe the neuro-cognitive researcher types, a lot of behaviors that humans and other animals express are evolutionary adaptations to survival situations. E.g., we're social and moral because teamwork helps survival and fairness promotes teamwork, and so forth.

A smart programmer is going to want to emulate enough of that so that their intelligence isn't sociopathic, but not all of it. So it's safe to assume that a genuinely sentient AI is still going to be a very alien form of intelligence to us.

For instance, without evolving from something with a survival instinct, an AI isn't likely to experience fear or a desire for self-preservation. Without the instincts related to genetic propagation, you lose entire swaths of human behaviors. So I don't envision AI becoming hostile or trying to supplant us; we're not competitors in the way biological organisms are with each other.

I do believe that whereas humans don social "masks" to interact, sophisticated AIs will be able to create human-oriented interfaces to deal with us in a tolerable fashion, if we don't create them with those already. It's just smart to not creep out the species that can turn off the electricity.

Humans, meanwhile, already have an innate tendency to anthropize inanimate objects. With AIs, it'll just happen to become true.

And with the whole host of legal and ethical and legal issues, I think it will be far more likely for humans to stick with expert systems for their tools and weapons. Pure AI will be for research purposes only and probably very highly regulated once it comes to pass.

Bottom line: I do think AI will happen eventually but I think it will wind up being a lot less dramatic than a lot of sci-fi imagines. We'll just be too different from each other to need to fight over very much.