Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

How does science work? And what's all this about quantum mechanics?

Moderators: AMod, iMod

wtf
Posts: 1178
Joined: Tue Sep 08, 2015 11:36 pm

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by wtf »

Skepdick wrote: Thu Dec 16, 2021 1:49 am If what you call "conscious" is my ability to examine, introspect, and modify the structure of my thoughts and alter my own behavior at runtime then by that yardstick computers are already conscious.
I find it annoying when you talk like that. We are not card decks being run through the great IBM 1442 punchcard reader in the sky. I don't think you're being serious when you say things like that. It seems trollish to me. And if you are being serious, it's even worse. Normally I would ask you what you mean by your "runtime," and if there is a corresponding compile time. But I don't want to know. It would not be a productive conversation.
Skepdick wrote: Thu Dec 16, 2021 1:49 am If you accept that measure of consciousness, then we do know how to implement consciousness. Obviously it's incomplete, but that's par for all science.
That we're card decks? Or algorithms of any sort? I reject that thesis utterly. But you already know that. We have no idea how to implement consciousness. Except, as philosopher and computer scientist David Gelernter said, "the old fashioned way." You get yourself a male and a female of the species and they implement a new consciousness. That's the only way we know how to do it. Delicacy precludes further elaboration.
Skepdick wrote: Thu Dec 16, 2021 1:49 am If you don't accept that measure... what measure of consciousness would you accept?
The Cartesian one. I think therefore I am. My reality may be a simulation, or an illusion created by a supremely clever daemon; but the "I" that experiences it, is real.

But there is some deceiver or other, very powerful and very cunning, who ever employs his ingenuity in deceiving me. Then without doubt I exist also if he deceives me, and let him deceive me as much as he will, he can never cause me to be nothing so long as I think that I am something. So that after having reflected well and carefully examined all things, we must come to the definite conclusion that this proposition: I am, I exist, is necessarily true each time that I pronounce it, or that I mentally conceive it. -- Descartes, Meditations on First Philosophy, 1641

That's my consciousness.

commonsense wrote: Thu Dec 16, 2021 3:52 pm You cannot make that characterization. Skepdick may either be conscious or not. He just doesn’t know which is the case.
I confess I was influenced by my history with my friend Skepdick. But yes I see your point. And actually I think it's Dennett among others who deny the existence of consciousness. I can't fathom the meaning of that claim. But it's considered respectable in the philosophy biz so if I say it's nonsense, that's only my opinion and not one shared by the professionals.
Last edited by wtf on Mon Dec 20, 2021 10:50 am, edited 2 times in total.
User avatar
Sculptor
Posts: 8529
Joined: Wed Jun 26, 2019 11:32 pm

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by Sculptor »

socrat44 wrote: Tue Nov 09, 2021 7:58 am Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI
/ by DAVID NIELD, 5 NOVEMBER 2021 /
The idea of artificial intelligence overthrowing humankind has been talked about for many decades,
and in January 2021, scientists delivered their verdict on whether we'd be able to control a high-level
computer super-intelligence. The answer? Almost definitely not.
------
The catch is that controlling a super-intelligence far beyond human comprehension would require
a simulation of that super-intelligence which we can analyze. But if we're unable to comprehend it,
it's impossible to create such a simulation.

"A super-intelligent machine that controls the world sounds like science fiction," said computer
scientist Manuel Cebrian, from the Max-Planck Institute for Human Development.
"But there are already machines that perform certain important tasks independently
without programmers fully understanding how they learned it."

"The question therefore arises whether this could at some point become
uncontrollable and dangerous for humanity."

https://www.sciencealert.com/calculatio ... UkCpvwMITo
The article you link does not reflect the articles they themselves reference.

The articles that "ScienceAlert" use do not say that the AI would be uncontollable. It says that the algorithm that could be used to ensure that no human would be harmed cannot yet be written.
Machines are constantly harming humans, the instance of AI means that could harm them less than they do now. The Max Planck institute that Science alert references say that it is not yet possible to foolproof the program to guarentee no harm to humans.
This in no way implies an AI "out of control" is states the bleeding obvious that AIs could take actions that inadvertantly harmed humans.

It is the difference between a driverless car failing to see a person on the road, and a driverless car driving on the pavements trying to kill humans.
Your thread is without merit.

What you have here is an hysterical populist webpage offereing click bait deliberately misreading an academic article to look more sexy.
Skepdick
Posts: 14363
Joined: Fri Jun 14, 2019 11:16 am

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by Skepdick »

wtf wrote: Mon Dec 20, 2021 10:00 am I find it annoying when you talk like that. We are not card decks being run through the great IBM 1442 punchcard reader in the sky. I don't think you're being serious when you say things like that. It seems trollish to me. And if you are being serious, it's even worse. Normally I would ask you what you mean by your "runtime," and if there is a corresponding compile time. But I don't want to know. It would not be a productive conversation.
Dude. I can't help you with your Mathematical/reductionist bias.

On more than one occasion you have made it explicit that Mathematics doesn't recognise time.

If you don't recognise time then your semantics don't encode time.
If you are not encoding time in your semantics, then you are not using linear logic OR temporal logic.

We are speaking different meta-languages.
wtf wrote: Mon Dec 20, 2021 10:00 am That we're card decks? Or algorithms of any sort? I reject that thesis utterly. But you already know that.
Why do you reject it? Every damn law in physics is deterministic!
Even quantum physics cannot falsify superdeterminism.

Surely you have a proof of impossibility as grounds for your rejection?

https://en.wikipedia.org/wiki/Proof_of_impossibility

If you know that physics is a dead end just prove it already - so physicists stop wasting their time.

For starters - do you even grok that the card decks don't DO anything or mean anything in a vacuum? They are just holes on paper until something (a runtime!) interprets them. So not only do you not understand the difference between denotational and operational semantics, you also don't understand that some systems don't even have well-defined or even well-definable formal semantics. Distributed/concurrent systems for example - process calculi. Something like Google's control plane - it's damn obvious that it can be implemented.

It's also obvious that it can't be captured as an algorithm. But it's made up of algorithms.
wtf wrote: Mon Dec 20, 2021 10:00 am We have no idea how to implement consciousness.
You seriously don't understand the semantic and epistemic problem here, do you?

IF we happen to implement a machine that happens to emulate humans sufficiently well so that you are happy to ascribe "consciousness" to that machine it is precisely at that moment when consiousness ceases to exist SCIENTIFICALLY. It is precisely at that moment consciousness becomes falsified. Like phlogiston.

The term "consciousness" ceases to do any work in any human theory precisely ​because the phenomenon has been accounted for in terms of SOMETHING ELSE.

Because it doesn't do any work - by Occam's razor you discard it.
Last edited by Skepdick on Tue Dec 21, 2021 9:26 am, edited 7 times in total.
Skepdick
Posts: 14363
Joined: Fri Jun 14, 2019 11:16 am

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by Skepdick »

wtf wrote: Mon Dec 20, 2021 10:00 am The Cartesian one. I think therefore I am. My reality may be a simulation, or an illusion created by a supremely clever daemon; but the "I" that experiences it, is real.
Do you actually understand that "I am thinking" and "I am conscious" are the exact same epistemic problems?

What IS (thought|consciousness|experience)? You don't know!

Why are you acribing (thought|consciousness|experience) to yourself if you don't know what (thought|consciousness|experience) IS?

I am real.
I am not real.
I am both real and not real.

ALL of the above expressions necessitate the "I", yet the assertion about the "I" is entirely arbitrary!
wtf wrote: Mon Dec 20, 2021 10:00 am But there is some deceiver or other, very powerful and very cunning, who ever employs his ingenuity in deceiving me. Then without doubt I exist also if he deceives me, and let him deceive me as much as he will, he can never cause me to be nothing so long as I think that I am something. So that after having reflected well and carefully examined all things, we must come to the definite conclusion that this proposition: I am, I exist, is necessarily true each time that I pronounce it, or that I mentally conceive it. -- Descartes, Meditations on First Philosophy, 1641

That's my consciousness.
Nobody is denying the existence of the "I". Self-reference IS recursion (computation).

What I am pointing out is that you can't even determine whether your perspective on the "I" is an intensional or extensional one.
Ontologically speaking you can't even determine where the "I" starts and ends which epistemically puts you in the shithouse.

Which are the necessary vs sufficient existents for "I" ? Which parts of your physical body can be removed without compromising the "I" ?
Iwannaplato
Posts: 6656
Joined: Tue Aug 11, 2009 10:55 pm

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by Iwannaplato »

wtf wrote: Fri Nov 19, 2021 6:26 am Do you understand that moving shipping containers around the world is a more complex data processing problem than playing chess? It's only because gizmos on store shelves aren't as sexy as AI hype that you are not seeing what's right in front of you, and are seduced by the latest "Ohmigod we're all gonna die from AI" nonsense.

I would say that the fact that you think it's brand-new news that there are systems we don't understand and can't control, shows that you haven't studied much of technology or even history. If you want a historical example, look at how World War I started. Europe literally stumbled backwards into it, with the greatest statesmen of the day failing to grasp the inevitably disastrous results of policies that seemed like a good idea at the time.

Human history is the story of one poorly understood out-of-control system after another. I'm not worried about chess-playing programs or Tensor Flow or the latest wild-eyed AI hype.
But generally speaking no group of scientists is trying to get the global supply system to be a single functioning unit that as a whole approaches problems. It is an extremely complicated system without a controlling center. Perhaps it could become self-aware, but 1) no one is trying to do this and it resembles non-centered things like ecosystems more than individual intelligences that make decisions and have goals from desire, say. With AIs you have people specifically trying to create a singular intelligence, with varying degrees of planned and current precautions around giving that entity control or potential control of parts of the physical world.

If any team of scientists decides to try to create a singular intelligence hub for the world's supply chain, that will be a team working on an AI and yes, there damn well better be caution in there. And as far as we can tell it might be a huge threat.
socrat44
Posts: 309
Joined: Tue Jan 15, 2019 4:20 pm

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by socrat44 »

Artificial intelligence that can discover hidden physical laws in various data
Date: December 9, 2021
Source: Kobe University
Summary:
Researchers have successfully developed artificial intelligence technology that can extract
hidden equations of motion from regular observational data and create a model that
is faithful to the laws of physics. This technology could enable us to discover the
hidden equations of motion behind phenomena for which the laws were considered
unexplainable. For example, it may be possible to use physics-based knowledge and
simulations to examine ecosystem sustainability.
----
These research achievements were made public on December 6, 2021, and were presented
at the Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS2021),
a meeting on artificial intelligence technologies. This research was among the
top 3% selected for the spotlight category.
-----
https://www.sciencedaily.com/releases/2 ... 222121.htm
----
Is ''artificial intelligence'' smarter than their creators?
--------
Age
Posts: 20194
Joined: Sun Aug 05, 2018 8:17 am

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by Age »

socrat44 wrote: Sat Dec 25, 2021 10:19 am Artificial intelligence that can discover hidden physical laws in various data
Date: December 9, 2021
Source: Kobe University
Summary:
Researchers have successfully developed artificial intelligence technology that can extract
hidden equations of motion from regular observational data and create a model that
is faithful to the laws of physics. This technology could enable us to discover the
hidden equations of motion behind phenomena for which the laws were considered
unexplainable. For example, it may be possible to use physics-based knowledge and
simulations to examine ecosystem sustainability.
----
These research achievements were made public on December 6, 2021, and were presented
at the Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS2021),
a meeting on artificial intelligence technologies. This research was among the
top 3% selected for the spotlight category.
-----
https://www.sciencedaily.com/releases/2 ... 222121.htm
----
Is ''artificial intelligence'' smarter than their creators?
--------
No, in regards to some things. Yes, in regards to other things.
socrat44
Posts: 309
Joined: Tue Jan 15, 2019 4:20 pm

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by socrat44 »

Age wrote: Sat Dec 25, 2021 11:21 am No, in regards to some things. Yes, in regards to other things.
1 - Yes, in regards to some things.
For example: the chess computer "knows" how all the pieces move and
what the ultimate goal of the game is. Under this limited AI program,
the computer can solve exercise better than Kasparov.
AI doesn't develop new programms, new laws of game.

2 - No, in regards to some things.
For example: the article said that ''Researchers have successfully developed
artificial intelligence that can extract hidden equations of motion from regular
observational data and create a model that is faithful to the laws of physics.
This technology could enable us to discover the hidden equations of motion
behind phenomena for which the laws were considered unexplainable. ''

In other words, researchers have developed AI smarter than the creators themselves.
I don't believe in this possibility. An ''artificial intelligence'' cannot be smarter than
their creators, AI will always work in a given program.
Can i be wrong?
Age
Posts: 20194
Joined: Sun Aug 05, 2018 8:17 am

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by Age »

socrat44 wrote: Sun Dec 26, 2021 2:40 pm
Age wrote: Sat Dec 25, 2021 11:21 am No, in regards to some things. Yes, in regards to other things.
1 - Yes, in regards to some things.
For example: the chess computer "knows" how all the pieces move and
what the ultimate goal of the game is. Under this limited AI program,
the computer can solve exercise better than Kasparov.
AI doesn't develop new programms, new laws of game.

2 - No, in regards to some things.
For example: the article said that ''Researchers have successfully developed
artificial intelligence that can extract hidden equations of motion from regular
observational data and create a model that is faithful to the laws of physics.
This technology could enable us to discover the hidden equations of motion
behind phenomena for which the laws were considered unexplainable. ''

In other words, researchers have developed AI smarter than the creators themselves.
I don't believe in this possibility. An ''artificial intelligence'' cannot be smarter than
their creators, AI will always work in a given program.
Can i be wrong?
Yes.
promethean75
Posts: 4931
Joined: Sun Nov 04, 2018 10:29 pm

Re: Calculations Suggest It'll Be Impossible to Control a Super-Intelligent AI

Post by promethean75 »

All you gotta do is prevent the machines from creating and/or reaching reserve sources of energy, and the sonsabitches will shut down. That's the advantage humans have over the terminators. We can metabolize and generate caloric energy from a variety of materials, while they have to generate and store it by direct electrical means. Whatchu gotta worry about is solar power, tho. But by the time skynet goes full retard, I'm pretty sure the capitalist industrialists will have blotted out the sun with their pollution, so collecting solar power is gonna be a bitch.

Yo I'm like a fuckin Nostradamus. What if the shit really does go down like that?

Wait why is there a black helicopter over my house right now?
Post Reply