The life cycle of software objects tackles this question in depth, and i think it was a good read.

Especially addressed is the necessity of experience and our expectations of a new Ai to perform as though it were an adult human.

Good book.



http://www.amazon.com/Lifecycle-Softwar ... 756&sr=1-1

johnson1010 wrote:Suppose an AI is created that seemlessly re-creates human behavior, including creativity, imagination, and self knowledge.

How do we treat such an entity?

Does it have a right to life, liberty and the pursuit of happiness?

Is it a crime to turn it off, or alter its programming? If not a crime, is it an injustice?

AI's limit would always be to mimic emotion. Nothing more.

AI's limit would always be to mimic emotion. Nothing more.

"Always"? Certainly now that is the case. No pre-scripted character of skyrim actually has emotions. They are performing the tricks that the programmers intended them to perform.

But how do you impose that limitation on a future AI which perfectly duplicates human experience?

Take the example of a computer back-up of a person's mind. If it is perfectly running a copy of, say, your mother's mind, when it tells you it's lonely what do you have to say about that?

It's just pretending to be lonely?

What separates your set of interconnected chemical receptors from an equally powerful and complex machine’s set of interconnected chemical receptors?

In that case, both your neural net, and the computer’s processors are working with equally complex systems. Both are powered through a form of electricity. Both are comprised of chemical compounds.

Why the special privilege of your emotions being real, and the computer’s being fake?

You are both chemical machines. Your brain arose from evolution, the computer’s brain arose from human construction. But does the simple fact that we created it mean that we can ignore the fact that it weeps?

Does the fact that you created your children mean you can abuse them without consequence?

Mind you, I am not talking about a scripted set of responses, designed to fool people into thinking they are talking to a human. I’m not talking about video game avatars with a total set of 20 possible responses pre-determined by user actions.

I am talking about a fully sentient, thinking artificial intelligence. One that has to suss out answers to changing input. One that learns how to behave, learns how to create, learns how to interact with other. One that can be wrong through trial and error, but is capable from learning from it’s mistakes.

This machine cannot have emotions, simply because it’s been fabricated?

Think of it this way.

Humans are a very specific group. Nothing not born of a human could ever be a human. A person, though, could be anything.

Could we deny the person-hood of extra-terrestrial aliens who traversed the galaxy to come say hi to us?

If you stepped into a time machine, went forward 200,000 years to find the world dominated by sentient dogs, could you deny their personhood? Could they deny YOU yours, simply because you aren't a dog?

Why the barrier on artificial intelligence?

johnson1010 wrote: Why the barrier on artificial intelligence?

I agree, a lot of people have an implicit assumption of dualism, that there is something other than normal, physical matter that differentiates human minds. But just because we don't know everything about consciousness doesn't make it so.

AI's limit would always be to mimic emotion. Nothing more.

Initially they will mimic emotional responses. I don't think AI will ever need emotion, and I don't think there's anything wrong with that. Humans have emotions to influence our actions, in the direction of greater survivability. Guilt, shame, love, fear. We need this to survive, as evolutionarily stable strategies are only undertaken via the influence of our emotions(in general, of course). The same behavior can arises from systems without emotion. Emotion works, we wouldn't be here without it, but there are other mechanisms for altruistic behavior control.

I'd venture to say that the behavior control system for AI would be an upgrade from emotion.

Lets look at emotions for a second.

Winning the lottery is not in itself enough to make a person elated.

First they have to understand that they won.

Emotions are biochemical reactions to things which we understand consciously.

Why do we feel shame? We check our situation against the way we believe things ought to be. Whether that be our percieved inferiority in looks or sports or financial achievement, or we can judge that we have exceeded expectations and be proud, or that we have nothing to be ashamed of. in all these cases, it is comparing the status of some thing against some standard and measuring the result. Shame does not encompass all failures, though. For an honest best effort that still falls short can still be a proud moment.

Some emotions serve social roles, such as crying and laughing, and then they also serve physical needs such as fear, which lets us tap into reserves of physical energy and sensory acuity.

Sentient machines will probably not have the biochemical responses engineered into their systems, but they certainly will have the intellectual cross checking processes in them that lead to our own outward expression of emotion.

A sentient machine, in essence, will also know whether it should feel pride or shame. It will have no adrenaline to express pride, nor blood to pump to it's cheeks, but the realisation will be real and determinate.

In other words, though lacking the biochemical response, it really will have the equivalent of emotion.

johnson1010 wrote:

AI's limit would always be to mimic emotion. Nothing more.

"Always"? Certainly now that is the case. No pre-scripted character of skyrim actually has emotions. They are performing the tricks that the programmers intended them to perform.

But how do you impose that limitation on a future AI which perfectly duplicates human experience?

Take the example of a computer back-up of a person's mind. If it is perfectly running a copy of, say, your mother's mind, when it tells you it's lonely what do you have to say about that?

It's just pretending to be lonely?

What separates your set of interconnected chemical receptors from an equally powerful and complex machine’s set of interconnected chemical receptors?

In that case, both your neural net, and the computer’s processors are working with equally complex systems. Both are powered through a form of electricity. Both are comprised of chemical compounds.

Why the special privilege of your emotions being real, and the computer’s being fake?

You are both chemical machines. Your brain arose from evolution, the computer’s brain arose from human construction. But does the simple fact that we created it mean that we can ignore the fact that it weeps?

Does the fact that you created your children mean you can abuse them without consequence?

Mind you, I am not talking about a scripted set of responses, designed to fool people into thinking they are talking to a human. I’m not talking about video game avatars with a total set of 20 possible responses pre-determined by user actions.

I am talking about a fully sentient, thinking artificial intelligence. One that has to suss out answers to changing input. One that learns how to behave, learns how to create, learns how to interact with other. One that can be wrong through trial and error, but is capable from learning from it’s mistakes.

This machine cannot have emotions, simply because it’s been fabricated?

Although it's fun to pretend we are a step away from having a Blade Runner problem, we are far, far away from actually having to worry about that.
The smartest computers we have now are not even close to the level of a rat's brain, let alone, say, a chimpanzee.

The gulf that will never be bridged by AI is life context and semantic meaning. For instance, how would AI create on i'ts own, semantic meaning from nothing unless it was given a ready made storehouse (and god only knows how large of a storehouse that would need to be) of historical experiences unique to that AI?

Emotional responses are a mind/body connection. Are you also ready to proclaim that the future of AI will also include this connection? An AI's "body" is not a chemical creation - it's purely a mechanical creation. Unless of course you are willing to proclaim that we will be adding flesh and blood to our future robots.

The "abuse your child" question is just way too out in left field for me to address. I actually don't understand how you fitted it into this conversation. I'll just let it go as a non sequitur.

Are you familiar with the Chinese Room Thought experiment? I'd like your thoughts on that.

No, i agree we are a few jumps away from really having to deal with this issue of human-level AI.

What this question allows us to explore, however, is our evaluation of the "other" whatever that may be. How do we evaluate what it means to be a person, regardless of whether that person might be human?

AI seems the most likely avenue for us to discover other non-human people. Aliens are far away, and the time issues probably don't permit us to meet them face to face. Evolution is hamstrung by our existence, as just about anything not good for wearing or eating isn't really a part of the sere we impose. That leaves AI.

I don't know how we get from really good number crunchers and simulations to actual sentience. If i did know that i would be very rich.

The point i tried to make about emotions is that the intellectual status produces the chemical responses in our bodies.

Perfect execution of some feat is acknowledged, we realize what an achievement that is, our bodies kick off a chemical reaction, and we shout with pride and exhilaration.

Sentient AI probably won't have a body like ours, and that includes the same chemical responses that we identify with this exhilaration, but does that really detract from the intellectual cross-checking it does to realize that it has similarly performed some task excellently?

Isn't that the essence of emotion? The realization of some status? I performed well, leads to pride. I am underperforming leads to shame. I just missed being killed leads to relief. These are intellectual cross-references we perform that proceed our chemical responses. Does a lack of chemical response negate the realization, and the root of that emotion?

I don't think it does.

The abuse your child thing was in reference to a common argument that since robots are our creation and thus owe their existence to us, then they are nothing more than property and can be treated as non-persons, regardless of their intellectual, self-knowing attributes. (future sentient AI's).

My point being, that your children are likewise reliant on their parents and would also not exist without them, but they are sentient, and that demands it's own respect. It's the crux of this thread. what do we owe sentient machines?

We created them, but do we still have the right to un-plug them?

Yeah, i'm familiar. The pre-cursor of the turing test.

I don't think the turing test, or whether or not you can fool a native speaker into thinking you are one of them, is an adequate identification of true sentient AI.

It's a very involved thing, trying to identify sentience.

I don't have a good definition on hand. "I'd know it when i saw it" doesn't really say what needs to be said, but that is kind of where it stand these days.

That's why i like having these conversations.

ant wrote: Are you familiar with the Chinese Room Thought experiment? I'd like your thoughts on that.

Some interesting stuff on Wikipedia about it:

"The overwhelming majority," notes BBS editor Stevan Harnad, "still think that the Chinese Room Argument is dead wrong." The sheer volume of the literature that has grown up around it inspired Pat Hayes to quip that the field of cognitive science ought to be redefined as "the ongoing research program of showing Searle's Chinese Room Argument to be false."

http://en.wikipedia.org/wiki/Chinese_ro ... experiment

I'm in the camp that thinks it must be wrong. I defer to the smarter critics.