On the Topic of Artificial Intelligence

[Shaggai]

An AI has a weapon inherently built into it: its intelligence. It may not have physical weapons to start out with, but it's still dangerous as hell.

How would an intelligent AI be more dangerous than an intelligent human?

A sufficiently intelligent AI would recursively self-improve, until it reached the limits of its hardware. It would most likely be vastly more intelligent than a human.

[Valica]

(1a) Sentient AI is a long, long way away.
(1b) It's not a threat to us simply because it won't exist in our lifetimes.

(2) If it did exist, however, it wouldn't be a threat.
(3) An AI has as much potential to be a threat as a human does, it depends on what you give them.
(4) An AI with access to nuclear weapons is a potential threat, but so is a human.

(5) Anyway, by the time we have true AI, trans-humanism will have made us so similar to them it would be hard to tell the difference.

1a. Not really.
1b. We will almost certainly have full AI within 100 years. An optimist would say by 2040.
2. That's iffy.
3. If they were a full AI, the goal would probably be to make them as humanoid as possible at some point. So everything a human has.
4. An AI with a knife is a threat. Same as a human. Why are you trying to make them out to be separate entities?
5. Doubtful. Trans-humanism doesn't directly include removing the blood, brain, etc. AI humanoids wouldn't exactly need these.

The only threat is if we crate an AI meant to function through (a) computer(s).
If we create a standalone AI that functions on a single "brain", then we are almost certainly safe.

[Shaggai]

(1a) Sentient AI is a long, long way away.
(1b) It's not a threat to us simply because it won't exist in our lifetimes.

(2) If it did exist, however, it wouldn't be a threat.
(3) An AI has as much potential to be a threat as a human does, it depends on what you give them.
(4) An AI with access to nuclear weapons is a potential threat, but so is a human.

(5) Anyway, by the time we have true AI, trans-humanism will have made us so similar to them it would be hard to tell the difference.

1a. Not really.
1b. We will almost certainly have full AI within 100 years. An optimist would say by 2040.
2. That's iffy.
3. If they were a full AI, the goal would probably be to make them as humanoid as possible at some point. So everything a human has.
4. An AI with a knife is a threat. Same as a human. Why are you trying to make them out to be separate entities?
5. Doubtful. Trans-humanism doesn't directly include removing the blood, brain, etc. AI humanoids wouldn't exactly need these.

The only threat is if we crate an AI meant to function through (a) computer(s).
If we create a standalone AI that functions on a single "brain", then we are almost certainly safe.

I think that the 100-year estimate is rather quick. I would be willing to bet a nonzero amount of money that it would happen in the next two or three hundred, certainly, but we're far enough away from AI that 100 years is rather unlikely. I mean, I would only be a little bit surprised, but it's nowhere near "almost certain".

[Valica]

I think that the 100-year estimate is rather quick. I would be willing to bet a nonzero amount of money that it would happen in the next two or three hundred, certainly, but we're far enough away from AI that 100 years is rather unlikely. I mean, I would only be a little bit surprised, but it's nowhere near "almost certain".

Technology evolves at exponential rates.
What we do now was thought impossible 20 years ago.
What we do in 20 years will probably be called impossible now.

Some people say the singularity of technology will occur in 2040.
I think that estimate is possible but not likely.
I'm betting on 2050-2070.

[Lordieth]

How would an intelligent AI be more dangerous than an intelligent human?

A sufficiently intelligent AI would recursively self-improve, until it reached the limits of its hardware. It would most likely be vastly more intelligent than a human.

Also the ability to reproduce beyond human means. If released into the wild, with modern connectivity it could control a vast array of electronic devices. A single A.I could bring about a technological dark age.

[Shaggai]

I think that the 100-year estimate is rather quick. I would be willing to bet a nonzero amount of money that it would happen in the next two or three hundred, certainly, but we're far enough away from AI that 100 years is rather unlikely. I mean, I would only be a little bit surprised, but it's nowhere near "almost certain".

Technology evolves at exponential rates.
What we do now was thought impossible 20 years ago.
What we do in 20 years will probably be called impossible now.

Some people say the singularity of technology will occur in 2040.
I think that estimate is possible but not likely.
I'm betting on 2050-2070.

We're far enough away from AI that strong AI is unlikely to happen in the next hundred years, even given exponential growth.

A sufficiently intelligent AI would recursively self-improve, until it reached the limits of its hardware. It would most likely be vastly more intelligent than a human.

Also the ability to reproduce beyond human means. If released into the wild, with modern connectivity it could control a vast array of electronic devices. A single A.I could bring about a technological dark age.

Yup. Anything which potentially could contain a copy of the AI becomes an existential threat.

[Sociobiology]

Technology evolves at exponential rates.
What we do now was thought impossible 20 years ago.
What we do in 20 years will probably be called impossible now.

Some people say the singularity of technology will occur in 2040.
I think that estimate is possible but not likely.
I'm betting on 2050-2070.

We're far enough away from AI that strong AI is unlikely to happen in the next hundred years, even given exponential growth.

with the advent of memristors AI became much more likely, because they use the same type of information as neurons.
Our real limitation is how well we can image the brain, since it shows how to do most of the things we want an AI to do.

Also the ability to reproduce beyond human means. If released into the wild, with modern connectivity it could control a vast array of electronic devices. A single A.I could bring about a technological dark age.

Yup. Anything which potentially could contain a copy of the AI becomes an existential threat.

not as much as you think, a mind capable of creative thought must by definition be equally capable of error, so don't expect too much from an AI. the more intelligent we make them the more error prone they become.

[Salandriagado]

How would an intelligent AI be more dangerous than an intelligent human?

A sufficiently intelligent AI would recursively self-improve, until it reached the limits of its hardware. It would most likely be vastly more intelligent than a human.

That depends entirely on how much hardware you give it. Additionally, "more intelligent" does not imply "more dangerous". Quite the opposite, in my experience.

[Wisconsin9]

Solution to an AI threat: pull the plug. Literally. How is this not ludicrously obvious? Keep 'em on a leash until we make sure that they're as safe as we can make them.

[Shaggai]

A sufficiently intelligent AI would recursively self-improve, until it reached the limits of its hardware. It would most likely be vastly more intelligent than a human.

That depends entirely on how much hardware you give it. Additionally, "more intelligent" does not imply "more dangerous". Quite the opposite, in my experience.

If it's got enough hardware to be smart enough to improve itself, it probably has at least as much hardware as a human. Since it can recursively self-improve, it can make much better use of said hardware than a human.

That's because you were dealing with humans, presumably humans with relatively similar goals. An AI would not necessarily share the same goals as a human, and if you think intelligent people aren't dangerous then you have only been interacting with a specific subset of intelligent people.

[Shaggai]

Solution to an AI threat: pull the plug. Literally. How is this not ludicrously obvious? Keep 'em on a leash until we make sure that they're as safe as we can make them.

That works if and only if you can be sure they're safe.

[Wisconsin9]

Solution to an AI threat: pull the plug. Literally. How is this not ludicrously obvious? Keep 'em on a leash until we make sure that they're as safe as we can make them.

That works if and only if you can be sure they're safe.

You can never be 100% certain that something's safe.

[Vladislavija]

Technology evolves at exponential rates.
What we do now was thought impossible 20 years ago.
What we do in 20 years will probably be called impossible now.

I tend to disagree on all three parts.

[Aleckandor REDUX]

AI would change the face of global human society, culture, economics, and perhaps biology forever. If the applied sciences can deliver it to the world correctly, there would be some major historical shift (whether or not this'll be bloody or not is uncertain) that I guess would see Earth connected to some post-scarcity grid of sustainable efficiency, automating virtually all labor and industry. Human endeavors, since all immediate physical necessities and luxuries can now be met, will probably focus on areas of cultural creativity or now wonderfully AI-assisted scientific research. What I'm hoping for is some sort of Iain M. Banks-esque Culture kind of pan-human anarcho-socialist stellar civilization to arise from this, that is, if we can program these predominant AIs to develop humanocentric morals and the ability to empathize on a meaningful level.

[Valica]

Technology evolves at exponential rates.
What we do now was thought impossible 20 years ago.
What we do in 20 years will probably be called impossible now.

I tend to disagree on all three parts.

And? That simply makes you wrong.

Technology does grow at exponential rates. It's a fact.

We went thousands of years with slow innovation and inventions.
Then came the 19th and 20th century.

After the industrial age, machines were booming.

Fast forward to WW2; we were pumping out inventions like it was nobody's business.

Even further to the Cold War; we went to fucking Luna.

Then you get the 90s and the dawn of serious home computing.
Internet speeds were atrocious and all we had was dial-up, but people loved the internet.

Then you come upon the last 10 years and what do you find?
Fiber internet, 3D printing, virtual reality technology, cloning advancements, the ability to render the Earth in 3D in a browser, etc.
I could keep listing shit, but you get the idea.

Technology advancements allow for more technology advancements. It's literally exponential.

Secondly, imagine going back to 1994 and tell people we would have gigabit internet or that we could play games at 4k resolution...
Or show them a picture of the newest iPhone.
They would have called you insane.

Don't even get me started on quantum computing.
The technology wasn't even starting to emerge 20 years ago.

The third point might be the most flimsy because we understand now that tech evolves rapidly, so we can better gauge what will exist.
But even then, there's no telling what kinds of advancements might be made.

[Vladislavija]

I tend to disagree on all three parts.

And? That simply makes you wrong.

Technology does grow at exponential rates. It's a fact.

We went thousands of years with slow innovation and inventions.
Then came the 19th and 20th century.

After the industrial age, machines were booming.

Fast forward to WW2; we were pumping out inventions like it was nobody's business.

Even further to the Cold War; we went to fucking Luna.

Then you get the 90s and the dawn of serious home computing.
Internet speeds were atrocious and all we had was dial-up, but people loved the internet.

Then you come upon the last 10 years and what do you find?
Fiber internet, 3D printing, virtual reality technology, cloning advancements, the ability to render the Earth in 3D in a browser, etc.
I could keep listing shit, but you get the idea.

Technology advancements allow for more technology advancements. It's literally exponential.

Secondly, imagine going back to 1994 and tell people we would have gigabit internet or that we could play games at 4k resolution...
Or show them a picture of the newest iPhone.
They would have called you insane.

Don't even get me started on quantum computing.
The technology wasn't even starting to emerge 20 years ago.

The third point might be the most flimsy because we understand now that tech evolves rapidly, so we can better gauge what will exist.
But even then, there's no telling what kinds of advancements might be made.

I don't see how any milestones you mention mean anything. Here let me show you some additional "milestones".

Now I wonder what do those flat lines mean. Could it be that we're reaching a plateau in our exponential technological growth? If you want I can match for every milestone you make one limit we have or are approaching.

Extrapolation is very dangerous.

[Lordieth]

We're far enough away from AI that strong AI is unlikely to happen in the next hundred years, even given exponential growth.

with the advent of memristors AI became much more likely, because they use the same type of information as neurons.
Our real limitation is how well we can image the brain, since it shows how to do most of the things we want an AI to do.

Yup. Anything which potentially could contain a copy of the AI becomes an existential threat.

not as much as you think, a mind capable of creative thought must by definition be equally capable of error, so don't expect too much from an AI. the more intelligent we make them the more error prone they become.

I'm not sure I agree. Computers are capable of a great level of accuracy, and creativity doesn't necessarily mean error. A lot of what causes human error wouldn't apply to an artificial intelligence. At the very least, its capacity to learn will be beyond any level of a human, so while it may make mistakes, it will almost certainly learn from them.

It could make errors in judgement, yes. Morality is subject to such things.

[Vladislavija]

I'm not sure I agree. Computers are capable of a great level of accuracy, and creativity doesn't necessarily mean error. A lot of what causes human error wouldn't apply to an artificial intelligence. At the very least, its capacity to learn will be beyond any level of a human, so while it may make mistakes, it will almost certainly learn from them.

It could make errors in judgement, yes. Morality is subject to such things.

I say computers would be as prone to errors as humans. Software bugs can be a bitch to weed out. Especially when using data mining. Just remember that AI which was supposed to diff tanks from humans and failed miserably.

[Utceforp]

I'm not sure I agree. Computers are capable of a great level of accuracy, and creativity doesn't necessarily mean error. A lot of what causes human error wouldn't apply to an artificial intelligence. At the very least, its capacity to learn will be beyond any level of a human, so while it may make mistakes, it will almost certainly learn from them.

It could make errors in judgement, yes. Morality is subject to such things.

I say computers would be as prone to errors as humans. Software bugs can be a bitch to weed out. Especially when using data mining. Just remember that AI which was supposed to diff tanks from humans and failed miserably.

The difference being that you can fix bugs in AI, but bugs in humans are hard/impossible to identify or fix.

[Vladislavija]

I say computers would be as prone to errors as humans. Software bugs can be a bitch to weed out. Especially when using data mining. Just remember that AI which was supposed to diff tanks from humans and failed miserably.

The difference being that you can fix bugs in AI, but bugs in humans are hard/impossible to identify or fix.

Uhhhhh, I'm really not convinced about that. Have you worked with AI ever?

[The Black Forrest]

An AI has a weapon inherently built into it: its intelligence. It may not have physical weapons to start out with, but it's still dangerous as hell.

How would an intelligent AI be more dangerous than an intelligent human?

How would it understand compassion?

[Norstal]

The difference being that you can fix bugs in AI, but bugs in humans are hard/impossible to identify or fix.

Uhhhhh, I'm really not convinced about that. Have you worked with AI ever?

I'm not sure a lot of people worked with computers here.

AI has many things but they are mainly these:

• Getting to the correct state
• Optimization of state traversal

The danger in AI is not that they become sentient. The danger of AI is that it is too optimized. An AI that staples paper can be become dangerous as it may be able to staple everything and anything.

[Mefpan]

How would an intelligent AI be more dangerous than an intelligent human?

How would it understand compassion?

By programming the AI in a way that includes giving actual values to formless things like trust, loyalty and reliability which could very well be seen as valuable resources given that they're non-negligible factors in causing a preferred reaction in a person?

I mean, we write numbers on fancy slips of paper, call it money and suddenly it's extremely important. Shouldn't be too difficult to make it so that repeated positive interaction with us fleshbags is given some value as well.

[Shaggai]

How would it understand compassion?

By programming the AI in a way that includes giving actual values to formless things like trust, loyalty and reliability which could very well be seen as valuable resources given that they're non-negligible factors in causing a preferred reaction in a person?

I mean, we write numbers on fancy slips of paper, call it money and suddenly it's extremely important. Shouldn't be too difficult to make it so that repeated positive interaction with us fleshbags is given some value as well.

Define trust, loyalty, reliability, compassion, and positive interaction.

[Mefpan]

By programming the AI in a way that includes giving actual values to formless things like trust, loyalty and reliability which could very well be seen as valuable resources given that they're non-negligible factors in causing a preferred reaction in a person?

I mean, we write numbers on fancy slips of paper, call it money and suddenly it's extremely important. Shouldn't be too difficult to make it so that repeated positive interaction with us fleshbags is given some value as well.

Define trust, loyalty, reliability, compassion, and positive interaction.

Well, that teaches me to keep my mouth shut because I'd have to avoid answering that request due to a lack of ability to do so in any satisfactory way.

Though, I'm still holding my fingers crossed that people smarter than me can figure out a system of recognizing those things based on an evaluation of past man-machine interaction records. I mean, facial recognition's a thing to some extent now and I think companies are already looking into ways to decipher the exact mood a person is in based on their facial expressions. Can't find a reputable source on that though, so take that with a mountain of salt.