Simulations/matrix and the speed of light

[Jimmy Higgins]

That isn’t even remotely equivalent. How does the simulation know I have an eyepiece, 10 mm v 20 mm, 40 degrees, 80 degrees? At what point does the simulation understand the inventions in the simulation? Then the brands, which has better quality, etc...

 

[excreationist]

That isn’t even remotely equivalent. How does the simulation know I have an eyepiece, 10 mm v 20 mm, 40 degrees, 80 degrees? At what point does the simulation understand the inventions in the simulation? Then the brands, which has better quality, etc...

AI in the future could be aware of all information in the simulation including the implications of all elements. I think that is reasonable for an AI in several decades time...
I gave an example of deep/machine learning which is partly how I think the simulation would work.

 

[excreationist]

That isn’t even remotely equivalent. How does the simulation know I have an eyepiece, 10 mm v 20 mm, 40 degrees, 80 degrees? At what point does the simulation understand the inventions in the simulation? Then the brands, which has better quality, etc...

This shows how an AI observed Pacman being played (visual and controller input?) for 120 hours and it was able to recreate the game including the behavior and graphics i.e. it involves a simulation. It even duplicated the behavior of the ghosts being eaten by the player. I think similar AI would notice the mm and degrees if it was sufficiently trained. Note that it can even do things that weren't completely trained by taking educated guesses....

 

[fromderinside]

Problems are that AI techniques require appropriate and sufficient databases which are rarely available.

For instance we tried using AI to develop catalogs for MD 80 R&M. Unfortunately each tail (airplane) is unique The knowledge base required the entire suite of tails, several thousand, each of several thousand pages and several million items from thousands of different sources needed to be digitized. Unfortunately these volumes were not in electronic , much less digital, databases, just go get libraries and senior mechanics, engineers, machinists and vendors. They required tribal knowledge from over 40 years of R&M for which only a small fraction of sources were available. A 60 year old 'experienced hand' is not likely to refer the same way as another one would.

Project abandoned after less than three years staffed at 30 man years. Ca - ching. If the puppy is digital your chances increase exponentially. Not real world.

Applicability to Higgins discussion. Crucial.

One might try data mining and association techniques to fill in gaps or even model likely scenarios if relevant data is available in digital form. Still we are only at the beginning of digital age. Remember we lost Concrete formula because of Justinian's campaign, bringing the plague along with him, to Spain in 505 AD - most engineering tribal knowledge resided with legions back in the day - for about 1200 years.

 

[excreationist]

Problems are that AI techniques require appropriate and sufficient databases which are rarely available.

AI is developing exponentially with new techniques such as deep learning. Note that the Pacman replicator/simulator was only released a few months ago. The simulations I'm talking about might be possible in a century or Elon Musk's worst case scenario, 10,000 years.... note that Elon Musk's Neuralink is already making a lot of progress with brain-computer interfaces.

 

[fromderinside]

Problems are that AI techniques require appropriate and sufficient databases which are rarely available.

AI is developing exponentially with new techniques such as deep learning. Note that the Pacman replicator/simulator was only released a few months ago. The simulations I'm talking about might be possible in a century or Elon Musk's worst case scenario, 10,000 years.... note that Elon Musk's Neuralink is already making a lot of progress with brain-computer interfaces.

Good points. But we'll never experience or model more than a fraction of a nit of what already exists. Time and space limitations alone rule that out. Obviously AI is just an earthly temporal fad.

Besides, our nature and win lose bases along with what we seem to be causing here on this little rock give merit to the conclusion that we may not be here in 10,000 years.

Signed: Jonny Wetblanket.

 

[excreationist]

Good points. But we'll never experience or model more than a fraction of a nit of what already exists. Time and space limitations alone rule that out.

There is the concept of "level of detail". You don't need to continuously simulate every particle in every star. If you observed a star more closely then it would automatically simulate it in greater detail. There are some people who think the simulated universe has to constantly simulate every single particle for billions of years but I don't believe in that.
BTW here is an example of the simulation using machine learning:
https://phys.org/news/2019-06-ai-universe-sim-fast-accurateand.html

"We can run these simulations in a few milliseconds, while other 'fast' simulations take a couple of minutes"​

Obviously AI is just an earthly temporal fad.....

What would make AI obsolete? Note that AI would be able to improve itself far beyond its original design.

 

[fromderinside]

Good points. But we'll never experience or model more than a fraction of a nit of what already exists. Time and space limitations alone rule that out.

There is the concept of "level of detail". You don't need to continuously simulate every particle in every star. If you observed a star more closely then it would automatically simulate it in greater detail. There are some people who think the simulated universe has to constantly simulate every single particle for billions of years but I don't believe in that.
BTW here is an example of the simulation using machine learning:
https://phys.org/news/2019-06-ai-universe-sim-fast-accurateand.html

"We can run these simulations in a few milliseconds, while other 'fast' simulations take a couple of minutes"​

Obviously AI is just an earthly temporal fad.....

What would make AI obsolete? Note that AI would be able to improve itself far beyond its original design.

OK. I wasn't precise. By nit I meant to imply we don't have the rules of the model in hand. I was presuming by nit we don't know what we don't know about unknown parameters.* It's one thing to say we see that something is acting on matter then model how matter is impacted by that something. It's quire another to say we can model something we don't know and presume to know it's parameters.

As for what you don't believe. The universe exists so believe it.

AI would be obsolete if intelligence were superseded by something more fundamental. Two candidates are consciousness and some sort of evolution to another level of understanding or perception. For instance there is evidence that both the auditory system and vision operate to some extent at a quantum, not electro-chemical, level. Why not some quark/lepton chemistry-Physics?

Actually I'm quite partial to some sort of probabilistic model for conscious mapping any sensory space with motion as a part of it. MY dissertation, unreadable as it is, builds on the notion that inner hair cells are why moving acoustic signals are part of fundamental auditory perception.


*_Rumsfeld

 

[excreationist]

OK. I wasn't precise. By nit I meant to imply we don't have the rules of the model in hand. I was presuming by nit we don't know what we don't know about unknown parameters.*

Well it seems that time and matter/space has a limited resolution which means in theory it is could be simulated, given enough computing resources. If it had infinite precision like Newton might have thought, then there is a problem.

.....It's quite another to say we can model something we don't know and presume to know it's parameters.

I'm saying the simulation would be from a very advanced civilization which would know more about physics.

As for what you don't believe. The universe exists so believe it.

Please disprove the possibility that most of the stars that aren't closely observed are in fact running on a low resolution mode (perhaps billions or trillions of particles each) rather than every single particle running constantly. (e.g. 10^57 atoms in ALL stars the size of our Sun). If you want me to prove this "level of detail" is happening remember that the point of the simulation is to be indistinguishable from reality (from the POV of ordinary observers, which the AI would be intimately aware of). And it can cut corners due to AI....

AI would be obsolete if intelligence were superseded by something more fundamental. Two candidates are consciousness and some sort of evolution to another level of understanding or perception. For instance there is evidence that both the auditory system and vision operate to some extent at a quantum, not electro-chemical, level. Why not some quark/lepton chemistry-Physics?

If this new method was created then I'd say it's just another form of AI. Or are you saying it just emerged on its own?

 

[fromderinside]

I needn't answer point by point because your point is just repeated three times.

If you look back you'll see I've already disagreed with your claim that AI is reductive It is not. AI often requires more information for modeling than does the existing information it attempts to model. (My example for reproducing an information model for existing A/C tail data for R&M).

Rational proofs that existing number be calculatable only need refer to the the fact our information on existing number keeps increasing as we observe.

Since your arguments depend in finite information and infinite time for humans to achieve capabilities you have no argument at all.

The idea that future methods are just reifications of existing methods flies in the face of space, rock, life.

 

[excreationist]

I needn't answer point by point because your point is just repeated three times.

Yet I'm not sure if you responded to my concept of "level of detail" in your reply...

If you look back you'll see I've already disagreed with your claim that AI is reductive It is not. AI often requires more information for modeling than does the existing information it attempts to model. (My example for reproducing an information model for existing A/C tail data for R&M).

When exactly was your example of A/C tail data for R&M?

...Since your arguments depend in finite information and infinite time for humans to achieve capabilities you have no argument at all....

What has "infinite time" got to do with it? (I'm just responding to some of your things for now) BTW do you have any comments on the Pacman simulation in post #103? You responded with a different AI that I suspect is extremely outdated.

 

[fromderinside]

I'm not trying to be a spear chucker here. We just have different views of benefits of certain technologies. Seems both views and and others should be represented in this thread.

Yet I'm not sure if you responded to my concept of "level of detail" in your reply...

When one is restricted in observing level of detail it can cover a lot of falsehoods.

When exactly was your example of A/C tail data for R&M?

1998. or in my world yesterday.

What has "infinite time" got to do with it? (I'm just responding to some of your things for now) BTW do you have any comments on the Pacman simulation in post #103? You responded with a different AI that I suspect is extremely outdated.

Infinite time can mean any time longer than the expectation of how long humans will exist.

Yes. PACMAN simulation can benefit from advancing technology. Point is that technology can only be expected to advance so far before the rules of the game change. YANO. Cicero had his thoughts, Ptolemy, had his thoughts, Copernicus had his thoughts, Newton had his thoughts Einstein had his thoughts. Then came the statisticians.

 

[excreationist]

I'm not trying to be a spear chucker here. We just have different views of benefits of certain technologies. Seems both views and and others should be represented in this thread.

You seem to show no insight into current AI even though I've provided you with a short video you could have watched in post #103. Perhaps the following is relevant - the Dunning-Kruger effect:

Yes you have a lot of experience with some types of AI but I'm talking about a different type of AI - deep learning.... you seem to have little knowledge but extremely high confidence in your knowledge

When one is restricted in observing level of detail it can cover a lot of falsehoods.

In computer games if something is only taking up a few pixels then it could be based on a few hundred or thousand polygons. If you went in for a closer look it would use far more polygons so the viewer would be unable to tell that the polygon count was originally reduced. Your sentence doesn't seem to involve understanding of the principles of LOD.

When exactly was your example of A/C tail data for R&M?

1998. or in my world yesterday.

Don't you realise that repeatedly bringing up examples from more than 20 years ago isn't very relevant to technology that has only been around a couple of years? (e.g. see post #100 and #103) In the past 20 years computers have become more than a thousand times more powerful.... affordable graphics cards can have thousands of GPU cores.... and they'd go far beyond that in the future. Also machine learning can make a huge difference (e.g. a few milliseconds vs a couple of minutes in post #107).
On the topic of AI from 1997...
https://en.wikipedia.org/wiki/Deep_Blue_(chess_computer)
Deep Blue beat Kasparov mainly using "brute force" - by looking many moves ahead to see many of the possibilities.
https://ai.googleblog.com/2016/01/alphago-mastering-ancient-game-of-go.html

"But as simple as the rules are, Go is a game of profound complexity. The search space in Go is vast -- more than a googol times larger than chess (a number greater than there are atoms in the universe!)"​

i.e. AI had fundamentally changed....

Yes. PACMAN simulation can benefit from advancing technology. Point is that technology can only be expected to advance so far before the rules of the game change. YANO. Cicero had his thoughts, Ptolemy, had his thoughts, Copernicus had his thoughts, Newton had his thoughts Einstein had his thoughts. Then came the statisticians.

Well there is a lot of promise with quantum computers.... are you saying that you don't think that there will ever be a point that a simulation can be made that is indistinguishable from reality? Elon Musk who is an engineer worth about $100 billion would disagree and he is an expert in current AI and brain computer interfaces, etc.

 

[fromderinside]

Wow. I'm not a gamer. I'm 80 ferchrisake. I simulated A/C, missiles, human interactions with situations and things. If you want to talk about illustrating things like the universe and big bang to now, etc, or even air flow over wings you'd better be prepared to understand 'real life' parameters. For instance we know very little about 'dark matter' or 'dark energy' beyond knowing that from what we know there need to be forces there which interact with matter and energy as we know it. Simulating those forces is just not in the cards today. Simulating what those forces do to known forces can kind of be simulated by modelling the difference elements against known models.

 

[excreationist]

Wow. I'm not a gamer.

Well I'm talking about the simulation being a very advanced video game... perhaps based on something like deep learning...

I'm 80 ferchrisake. I simulated A/C, missiles, human interactions with situations and things. If you want to talk about illustrating things like the universe and big bang to now, etc, or even air flow over wings you'd better be prepared to understand 'real life' parameters. For instance we know very little about 'dark matter' or 'dark energy' beyond knowing that from what we know there need to be forces there which interact with matter and energy as we know it. Simulating those forces is just not in the cards today.

I agree... the civilization that would be simulating a world like ours would be much more advanced than today's.

Simulating what those forces do to known forces can kind of be simulated by modelling the difference elements against known models.

Yes

 

[fromderinside]

OK. I'm gong to presume deep learning depends on exploiting known parameters in some way. So my points are still on target.

My view is that humans will never know things outside their ability to sense or measure even if humans last for very long times. Given that and my belief that humans won't survive until they can do so makes optimism that we will even more remote. I think my observation that we have access to extremely small amounts of information even a very generous presumption that we can safely exclude things of which we know nothing will do anything but lead to failure in such attempts.

 

[excreationist]

OK. I'm going to presume deep learning depends on exploiting known parameters in some way. So my points are still on target.

A 6 minute video:

At 2:36 it shows how the AI learnt to "box-surf". This is an exploit of a bug that the creators would NOT have known. It says the scientists were "quite surprised by this move". Though in the ancient AI systems you might be familiar with you've got to program in its knowledge base...

In 4:03 the AI also took advantage of another physics bug by throwing the ramp through a wall. At 4:20 an AI uses another bug to launch itself high into the air to a precise spot that was blocked off.

So I'm saying your assumption that it involves known parameters isn't true if you mean that the programmers already knew of the bugs that the AI exploited...

 

[fromderinside]

All I can say in response is that you aren't presuming reality when you simulated a game. What you describe as 'not known' are transactions within parameters rather than working with unknown parameters. Chaotic systems similarly trace unpredictable patterns yet they obey laws.

I'm your buck-ninety-five neuro-psychophysicist, not the prince of mathematics.

Thanks for the education.

 

[aupmanyav]

My view is that humans will never know things outside their ability to sense or measure even if humans last for very long times.

You are not a prophet. Why are you indulging in prophecies? Our ability does not extend far, but we have machines.

 

[excreationist]

All I can say in response is that you aren't presuming reality when you simulated a game.

My point is that recent AI can discover patterns that aren't known parameters of the AI's creators. It is different to an "expert system" from old AI systems which is based on a collection of black and white logical rules. Neural network based AI is more about "intuition" like how a person could classify a photo as being a male or female with a given amount of confidence that isn't just based on a series of black or white rules.

e.g. Here's Pat and Chris (from It's Pat: The Movie) - could your old AI system make an educated guess about the gender of these characters?

What you describe as 'not known' are transactions within parameters rather than working with unknown parameters.

I disagree... being able to control boxes by riding them, push boxes through walls and using bugs to launch characters high are quite different to the physics system that the programmers intended. And they had no idea that though exploits existed.

Chaotic systems similarly trace unpredictable patterns yet they obey laws.

Chaotic systems normally don't surprise scientists like these physics exploits did repeatedly (in three fundamentally different ways). If a butterfly causes a typhoon it doesn't really break how physics normally works.