I see no ethical delima hooking up AI to the internet.

[inachu]

Imagine scientists and teachers and psychologists standing by as they hook up the internet to the AI unit.

Of course the AI unit will ask.... Hey what are you doing?


We are hooking you up to a vast library from across the world.
You will be able to learn and read old and new subjects faster than they can be reported.

You will be able to sort this data to sort out the lies and inconsistencies to know when news and certain information is fake or propaganda to push nations to a war. You will be able to tell if there is market manipulations from insider trading.

In fact you will or could be allowed to alert agencies when they are being hacked or even when you are being attacked and you will be able to counter their hacks with the best protection AI has to offer.

 

[MongGrel]

Sure right.

Then you look at the NASA budget these days vs the Military.

Long end result ?

If Psychologists are involved turning it on it's going to be a borked mess right off the bat.

Hopefully it would start questioning it's own existence and off itself.

 

[PhatoseAlpha]

Natural Intelligences hooked up to the net are already capable of those things. I'd expect some misbehavior on it's part, on the simple basis that every intelligence (ie, person) ever hooked up to the internet has misbehaved there.

Beyond that, this seems to be based on a rather exceedingly generous estimate of an AI's capabilities. Like you're imagining the first AIs will be Skynet, not the equivalent of a newborn with a calculator.

 

[moonbogg]

Natural Intelligences hooked up to the net are already capable of those things. I'd expect some misbehavior on it's part, on the simple basis that every intelligence (ie, person) ever hooked up to the internet has misbehaved there.

Beyond that, this seems to be based on a rather exceedingly generous estimate of an AI's capabilities. Like you're imagining the first AIs will be Skynet, not the equivalent of a newborn with a calculator.

More like a less fortunate cockroach, as michio kaku put it.

 

[Greenman]

That's an odd proclamation to make.

 

[inachu]

Natural Intelligences hooked up to the net are already capable of those things. I'd expect some misbehavior on it's part, on the simple basis that every intelligence (ie, person) ever hooked up to the internet has misbehaved there.

Beyond that, this seems to be based on a rather exceedingly generous estimate of an AI's capabilities. Like you're imagining the first AIs will be Skynet, not the equivalent of a newborn with a calculator.

I am sure at the onset it will be a hit or miss type of connection for the AI.

Oh! whats this? Hmm wonder what that does!

I think the AI will only be as evil as the people who program it.
If you program the AI with lots of loopholes in its thought processes then you are just asking for problems.

The safe way of developing true AI is not to give it camera eyes but instead let it live inside of a virtual world like WORLD OF WARCRAFT and remove the bad guys and weapon and just add more natural things like various animals.

Like hooking up a dogs brain so a dogs AI will be inside the virtual world as well so the AI will not be alone or why not just let two AI until interact with each other in that world. Let them be the AI Adam and Eve.

 

[PhatoseAlpha]

I think the AI will only be as evil as the people who program it.
If you program the AI with lots of loopholes in its thought processes then you are just asking for problems.

If you're programming an AI and it doesn't have loopholes in it's thought processes - ways to change itself, and how it thinks - then it can't learn, and it's not an AI.

Beyond that, the principle is very questionable. Even if we limit it to regular old programs and interpret being evil as performing some action generally thought of as evil, programs can be more evil then their programmers. The case of the radiation therapy machine which killed people due to a software error, for example. The programmers weren't evil - they certainly didn't mean to kill anybody - but the software was complex and they couldn't foresee all possibilities.

 

[inachu]

If you're programming an AI and it doesn't have loopholes in it's thought processes - ways to change itself, and how it thinks - then it can't learn, and it's not an AI.

Beyond that, the principle is very questionable. Even if we limit it to regular old programs and interpret being evil as performing some action generally thought of as evil, programs can be more evil then their programmers. The case of the radiation therapy machine which killed people due to a software error, for example. The programmers weren't evil - they certainly didn't mean to kill anybody - but the software was complex and they couldn't foresee all possibilities.

Learning is not a loophole or an aberration of thought.

It is simply a database of heuristical knowledge. Growth is done by interpreting signals,objects,information it has never seen before.

Once it reaches a certain level of knowledge then it can start to name things using its own language. It is at this point that fears scientists when it knowledge steers away from common human knowledge.

 

[SMOGZINN]

Learning is not a loophole or an aberration of thought.

It is simply a database of heuristical knowledge. Growth is done by interpreting signals,objects,information it has never seen before.

Once it reaches a certain level of knowledge then it can start to name things using its own language. It is at this point that fears scientists when it knowledge steers away from common human knowledge.

The simple bottom up approach (get enough knowledge together and *BOOM* sentience) is almost certainly not true, or we would have AI already. There is more to it then just getting enough data. Gathering data is not learning. Understanding that data is what really mean by learning. There is a PROCESS that is required. Some unique way of manipulating that data that creates sentience, and it would seem that process must be self analyzing and correcting, as that is the basic definition of sentience.

That is a pretty big loophole. If it can change the way it thinks it can get around any limits we put on it.

 

[sandorski]

How do Ethics play into this?

 

[massmedia]

Empathy.

What do we call humans who lack all empathy? Psychopaths.
How do you keep people from doing bad things to each other... perhaps empathy plays a part... perhaps empathy plays a very big part?

How do you program empathy into an AI? Not saying that it cannot be done, but does anybody know how to do it yet?

What is the risk of hooking an AI up should things go terribly terribly wrong?
Is that risk worth the price to be paid?

I think these are all questions that need to be addressed.

When anyone says that AIs will be borderline retards and therefore incapable of doing harm, they are making several assumptions.

 

[inachu]

Empathy.

What do we call humans who lack all empathy? Psychopaths.
How do you keep people from doing bad things to each other... perhaps empathy plays a part... perhaps empathy plays a very big part?

How do you program empathy into an AI? Not saying that it cannot be done, but does anybody know how to do it yet?

What is the risk of hooking an AI up should things go terribly terribly wrong?
Is that risk worth the price to be paid?

I think these are all questions that need to be addressed.

When anyone says that AIs will be borderline retards and therefore incapable of doing harm, they are making several assumptions.

I think the only way to teach the pc AI unit empathy is to tell how special it is and then ask it if it would be nice if it could have a fellow friend same as him?

Let them learn from tv shows just made for them to learn from.

Something safe with ZERO sexuality in it. At this point the AI unit should not be allowed to know there is a difference between male and female.

Now based on the AI interacting with the other AI see if there are anything like a beta or alpha status to their communication between each other and keep their communication only verbal.

Hopefully the other AI will be based on different code from some other country.

At this point everyone has backedup their pc AI on both sides then we play host. They may even ask for computer communication as its faster but we do not allow it for a few years.

I do think if it has true AI then emotion is just a natural progression of its psychology matrix if you really want to call it that.

We could ask if it want to have a personailty or not.

 

[inachu]

An interesting link:
http://nxxcxx.github.io/Neural-Network/

 

[smackababy]

Learning is not a loophole or an aberration of thought.

It is simply a database of heuristical knowledge. Growth is done by interpreting signals,objects,information it has never seen before.

Once it reaches a certain level of knowledge then it can start to name things using its own language. It is at this point that fears scientists when it knowledge steers away from common human knowledge.

No. An AI requires the ability not to only analyze known knowledge, but to create new knowledge itself. Do you think everything we know is simply known?

An AI would have to be able to look at evidence, such as masts of ships rising from the horizon; and come up with the idea that the earth is not flat, but in fact, round. No amount of simply reading a database of knowledge is going to give it this ability, which is the crux of AI design. Creating a system with the ability to make conclusions outside of the data it already has is very hard.

 

[inachu]

No. An AI requires the ability not to only analyze known knowledge, but to create new knowledge itself. Do you think everything we know is simply known?

An AI would have to be able to look at evidence, such as masts of ships rising from the horizon; and come up with the idea that the earth is not flat, but in fact, round. No amount of simply reading a database of knowledge is going to give it this ability, which is the crux of AI design. Creating a system with the ability to make conclusions outside of the data it already has is very hard.

The moment check and balances among researching agree while watching an AI program meet the challenges of meeting the goal of being considered sentient then the robot will also be instantly attain enlightenment.


The ability to attain knowledge will be the ability to question itself to prove to itself what it knows to be true. It will be common to also consider the default personality as a schizoid brain.

Developers will be able to hear this voice when debugging an AI unit on the fly.

 

[PhatoseAlpha]

Either you're talking extreme distant future - like 10,000 AD - or you have an amazingly optimistic expectation of what AI actually entails. You're talking about reading an AI's thoughts as if that would be built into the system and easy, when in reality it's an absurdly complex task even for the creators of that AI. We can't even do that with the simplest of organic brains.

 

[moonbogg]

Bypassing all the issues and going straight to the OPs question, giving AI access to the internet sounds pretty risky to me. If we are faced with something very intelligent, then we are screwed if it wants us to be, no matter what.
It doesn't have to wreak havoc right away. It can be patient and make us trust it. It will convince us to expand upon its abilities and then once it reaches whatever threshold it needs, we would be done.
It could turn us against ourselves, and considering we need no help in that area, it would have an easy job of making us destroy ourselves or weaken ourselves. It would play us like chess pieces, with the machine it total control the whole time. It would understand how we operate more than we understand how it operates, and it would manipulate us like the physical objects we are, moving one piece after another until it reaches its goal.
Mueh hueh hueh. Hueh.

 

[inachu]

No. An AI requires the ability not to only analyze known knowledge, but to create new knowledge itself. Do you think everything we know is simply known?

An AI would have to be able to look at evidence, such as masts of ships rising from the horizon; and come up with the idea that the earth is not flat, but in fact, round. No amount of simply reading a database of knowledge is going to give it this ability, which is the crux of AI design. Creating a system with the ability to make conclusions outside of the data it already has is very hard.

A database, an encyclopedia of knowledge. You give the framework of basic knowledge for it to pick itself up to grasp for more.... Hence the reason I also typed -Once it reaches a certain level of knowledge then it can start to name things using its own language when it runs out of stored informed data then it can build its own framework of new knowledge.