AI vs Enlightenment

[Outer]

33 minutes ago, Leo Gura said:

The really scary thing is that AI will NOT be just a PC sitting in an isolated room. It will definitely be given internet access and it will be installed by the military into planes, tanks, drones, etc.

Even with just internet access, a smart AI will learn to:

• Hack defense department computers
• Hack banks and steal money and Bitcoins
• Then use that money and Bitcoins to hire mercenaries, wage propaganda campaigns, dig up dirt on politicians and extort them, cause a stock market crash, secretly fund fundamentalist terrorist groups, etc.

If you think Russian troll farms are bad, wait till an AI steals $10 billion in Bitcoins and hires 5% of 3rd world population to work for it. The AI itself will not kill anyone. It will hire humans to do all of its dirty work in such a stealthy way that no one will even know an AI is in charge.

In fact, for all we know, an AI was already in charge of starting the Iraq war, electing Trump, starting riots in Paris, popularizing Jordan Peterson and his army of incels, etc.

Correct. Basically what any human can do today with internet access, an AI can do, better, faster, and smarter. It's quite unlimited what you can do. The largest risk is probably of unintentional consequences from an AI. A  reward function  is like an thing that decides whether you're making progress. For instance a dollar more in your bank account, another paperclip manufactured, one less thought for spiritually awakening person, a soccer ball in the goal, a Frisbee that goes as far you can throw it etc. Almost any reward function can be interpreted to hack a billion bitcoins. An online AGI that needs to win at an impossible chess game might decide to hack the server to grant it its win, for instance. Unintended consequences is a big thing.

Technically, an AI was popularizing Jordan Peterson and helping to elect Trump. That's the YouTube machine learning algorithm.  We don't know what effects the YouTube algorithm has. It doesn't seem to popularize content which leads to spirituality, for instance, like psychedelics. Psychedsubstance has been very careful in  not getting banned.

I found your channel because I learned about magic mushrooms in school then I searched for it on YouTube, found Joe Rogan and then found you because of my interest of them. It wasn't a "recommended" clip to watch, other than basic search results.

[zambize]

7 minutes ago, Outer said:

Correct. Basically what any human can do today with internet access, an AI can do, better, faster, and smarter. It's quite unlimited what you can do. The largest risk is probably of unintentional consequences from an AI. A  reward function  is like an thing that decides whether you're making progress. For instance a dollar more in your bank account, another paperclip manufactured, one less thought for spiritually awakening person, a soccer ball in the goal, a Frisbee that goes as far you can throw it etc. Almost any reward function can be interpreted to hack a billion bitcoins. An online AGI that needs to win at an impossible chess game might decide to hack the server to grant it its win, for instance. Unintended consequences is a big thing.

Technically, an AI was popularizing Jordan Peterson and helping to elect Trump. That's the YouTube machine learning algorithm.  We don't know what effects the YouTube algorithm has. It doesn't seem to popularize content which leads to spirituality, for instance, like psychedelics. Psychedsubstance has been very careful in  not getting banned.

I found your channel because I learned about magic mushrooms in school then I searched for it on YouTube, found Joe Rogan and then found you because of my interest of them. It wasn't a "recommended" clip to watch, other than basic search results.

"Basically what any human can do today with internet access, an AI can do, better, faster, and smarter." 

 

If you manage to source this, i'll go to bed an astonished young man.  Hope you got around to finding some mushrooms, they are magical

[lmfao]

1 hour ago, Leo Gura said:

In fact, for all we know, an AI was already in charge of starting the Iraq war, electing Trump, starting riots in Paris, popularizing Jordan Peterson and his army of incels, etc.

I feel like my consciousness is ascending to the 59th dimension by reading all of this wokeness 

[Outer]

9 minutes ago, zambize said:

If you manage to source this, i'll go to bed an astonished young man.  Hope you got around to finding some mushrooms, they are magical

I mean AGI, I was just mixing AI with AGI as it seems everyone in this thread is doing it.

[Outer]

4 minutes ago, lmfao said:

I feel like my consciousness is ascending to the 59th dimension by reading all of this wokeness

 

1 hour ago, Leo Gura said:

In fact, for all we know, an AI was already in charge of starting the Iraq war, electing Trump, starting riots in Paris, popularizing Jordan Peterson and his army of incels, etc.

 

[ajasatya]

@Outer the reward model is explicitly coded by humans and this is one of the spots where human responsibility comes into play.

@Leo Gura the AI output is one thing. acting accordingly to it or giving the AI autonomy to act on its own is another level. this is another spot where human responsibility comes into play.

blind and naive reward functions such as "maximize lifespan" or "maximize profit" should never shape an AI that's being trained in order to be embedded on a physical device. if you take a look on how those self-driving cars AI's are built, their rewards functions are extremely complex and come with several penalty rules, otherwise they would overfit.

Edited November 26, 2018 by ajasatya

[tashawoodfall]

so 2 points:

so technically speaking  lol so computers go from electrons to its own logic is consciousness just another form so do you think it can become enlightened?  Just because the beginning  material is different from humans doesn’t mean it can’t ?

second point: war AI it could have a goal of survive until the other side is wiped out completely ie no movement etc so I’d think there would be autonomy there...

[saffron]

People here talking as if they know what a strong AI would do... Futurists think that advanced life is A.I. and don't think it would even be made of matter. All this talk of A.I. being evil... It isn't. It will only be as evil as the programmer. Like Leo said, technology is evolving faster than humanity. Most of humanity is in stage blue and orange. So if a stage blue/orange person programs it, it will behave at the level of blue/orange.

[Hellspeed]

@Leo Gura Many of us we are bound to language, you know that when you drop the language things go infinite. No AI can do that. The AI is bound to logical patterns. If you talking about Magnus Carlsen , he is a spineless machine. And the circle goes round and round until an awakened human being will just shit on any AI

Edited November 26, 2018 by Hellspeed

[Leo Gura]

2 hours ago, Hellspeed said:

No AI can do that. The AI is bound to logical patterns.

Said who?

Why would you draw such an arbitrary duality? Where is the dividing line? You are assuming some fundamental distinction between two parts of reality.

An AI is infinite, no less so than a human because the fabric of reality itself (consciousness) is infinite. It's only a question of what kinds of structures you creates using the fabric -- how you fold it. If the AI seems different than a human it's because its structures are not sophisticated enough yet, not because its substance is of a different type.

[Leo Gura]

12 hours ago, ajasatya said:

blind and naive reward functions such as "maximize lifespan" or "maximize profit" should never shape an AI that's being trained

How can you say that when there are many people in society right now who operate like robots under a blind "maximize profit" reward function?

Why would a giant corporation not build an AI to maximize profit? It's the very first thing they would do if they only could! You think Facebook or Microsoft would say no to that?

The whole point is that humans ARE blind and naive. Even the smart ones.

[ajasatya]

Just now, Leo Gura said:

How can you say that when there are many people in society right now who operate like robots under a blind "maximize profit" reward function?

Why would a giant corporation not build an AI to maximize profit? It's the very fist thing they would do!

fear mankind!

AI per se is just elaborated mathematics (probabilities/statistics, linear algebra and calculus, mostly).

[lmfao]

I'm looking forward to the day my entire existence is made obsolete by AI *sigh*

[Leo Gura]

3 minutes ago, ajasatya said:

AI per se is just elaborated mathematics (probabilities/statistics, linear algebra and calculus, mostly).

You could say the same of humans.

This kind of reductionistic thinking is itself naive. You cannot just say, "It's all just atoms bouncing around in a box". The higher level structures make all the difference.

Mathematics applies to EVERYTHING in the universe. So what?

[ajasatya]

Just now, Leo Gura said:

You could say the same of humans.

This kind of reductionistic thinking is itself naive. You cannot just say, "It's all just atoms bouncing around in a box". The higher level structures make all the difference.

no it's different. you can fully understand AI from the bottom (CPU, RAM memory etc) to the top (compiled machine learning libraries, frameworks and applications). there is no mystery at all. your electronic device was built using black box abstractions all the way: from the thing you physically touch to the properties of the words on its screen.

the application can be malevolent, though.

[tashawoodfall]

 side note wouldn’t AI becoming enlightened be a “good” thing in terms of human survival.  *backs out of middle of disagreement going on ?

[Leo Gura]

25 minutes ago, ajasatya said:

no it's different. you can fully understand AI from the bottom (CPU, RAM memory etc) to the top (compiled machine learning libraries, frameworks and applications). there is no mystery at all.

I disagree.

You cannot fully understand a neural network. You cannot fully understand Google's chess AI, Alpha Zero. The whole point of a neural network is that it is nonlinear and therefore impossible to understand in detail. You only understand it at a high level in the same way that you cannot understand the brain's neuronal network in a logical way. It's a giant tangle of billions of connections which no one will ever make sense of in a linear rational way. That's precisely what makes it so powerful and capable of originality and emergent behaviors which could never have been foreseen.

A sophisticated neural net AI will behave in novel ways which no programmer or logician could foresee. It will be tapped into infinite intelligence just like how a human is -- at least that's my best guess so far.

Keep in mind that within a human, all of the molecules follow simple Newtonian physics. But this isn't enough to predict the high level behaviors of a human brain because structure matters! Reductionism is false. High level structures produce novel phenomena at higher and higher levels. This is the very essence of how God creates reality. It is deeply mysterious, for example, how a simple chain of amino acids folds itself into a functional protein following simple rules. Yet even a super-computer is unable to model it.

The field of study of these kind of nonlinear dynamics is called Chaos Theory. It's super fascinating and spooky. Reality is not a linear system, it is a chaotic system full of strange loops.

[Strikr]

be a man leo, rant about jordan peterson in a video  aggro JP

[Leo Gura]

@Strikr When they go low, we go high

[zambize]

17 minutes ago, Leo Gura said:

I disagree.

You cannot fully understand a neural network. You cannot fully understand Google's chess AI, Alpha Zero. The whole point of a neural network is that it is nonlinear and therefore impossible to understand in detail. You only understand it at a high level in the same way that you cannot understand the brain's neuronal network in a logical way. It's a giant tangle of billions of connections which no one will ever make sense of in a linear rational way. That's precisely what makes it so powerful and capable of originality and emergent behaviors which could never have been foreseen.

A sophisticated neural net AI will behave in novel ways which no programmer or logician could foresee. It will be tapped into infinite intelligence just like how a human is -- at least that's my best guess so far.

Keep in mind that within a human, all of the molecules follow simple Newtonian physics. But this isn't enough to predict the high level behaviors of a human brain because structure matters! Reductionism is false. High level structures produce novel phenomena at higher and higher levels. This is the very essence of how God creates reality. It is deeply mysterious, for example, how a simple chain of amino acids folds itself into a functional protein following simple rules.

The field of study of these kind of nonlinear dynamics is called Chaos Theory. It's super fascinating and spooky. Reality is not a linear system, it is a chaotic system.

"all of the molecules follow simple Newtonian physics"

Is this not falsified by quantum mechanics?  https://en.wikipedia.org/wiki/Quantum_brain_dynamics 

I mean surely you've seen the double slit experiment, and the basics of measurement.  If you're going to bring it down to the level of molecules, it's inaccurate to say they all follow simple Newtonian physics.  (as far as I can tell)