Quantum Mechanical consequences for AI

[Sithdarth]

I will preface this by saying that this is not about how quantum mechanics might someday allow us to make an AI. This is about the consequences that might arise from an AI making observations.

To start with we must rehash Schroedinger's Cat. We probably all know about the cat in the box with the poison in a supposition of released and not released making the cat alive and dead. Now Schroedinger himself sidestepped the paradox by saying the cat as a conscious being was perfectly capable of collapsing the wave function for the entire universe. The cat in effect determined its own reality and was never in a supposition. That is to say there is something about a "conscious" mind that directly impacts reality in a way that anyone else can clearly see.

The Quantum Eraser Experiment (http://en.wikipedia.org/wiki/Quantum_eraser_experiment) and The Delayed Choice Quantum Eraser Experiment (http://en.wikipedia.org/wiki/Delayed_choice_quantum_eraser) show us that even an observation by a non-conscious observation will also collapse the wave function but if that information is removed the wave function goes back to normal. In fact it can even retroactively return to normal. That is the effect propagates backward in time.

The combination of these two things is the real problem. If a conscious observer looks at the path information before it gets erased then that information now exists in their head. That information can then not be erased unless it is also erased from the observer's memories and the wave function remains collapsed.

Now we have two real possibilities from here. The first is that it is just hard to erase the information from a mind but once done the wave function would return to normal. One potential way this could work has to do with entropy changes included in the production and storage of information. This of course means there is nothing special about conscious observation beyond the way in which the information is stored. There is a certain attractiveness to this in that it removes all life from any place of importance. It does still allow for a definition of consciousness as any system in which there is a certain level of difficulty in erasing information to the point of reverting a wave function. Anything with a difficulty level above a certain threshold is conscious and anything below is not. Finding that threshold might be hard and might exclude or include somethings that are problematic. Like plants might be able to collapse a wave function to the point were reversing it is hard enough to be called conscious. If it is related to entropy changes we could actually have a mathematical formula to predict the exact point where the entropic changes do to information flow in a system produce consciousness.

The second is that there is something inherently different about "consciousness". Which I personally dislike in that it sounds a little hooky and it seems to me to violate Occam's Razor. It's something that cannot be measured and makes no predictions. Like the first one it allows for a both a definition of consciousness and a surefire test to determine consciousness. Of course you could still end up with things that you are uncomfortable with being defined as conscious. Also, if there is something special about consciousness the question becomes is it intrinsic to biology or could computers actually become conscious. If they can't then its intrinsic to biology which makes it an even stranger metaphysical concept. If they can it poses the opposite question. That is if consciousness is special but both machines and living things of sufficient complexity can have it then what the heck is it and how was it imbued in the machine if we accept that consciousness would arise gradually with increased complexity. (That is as it did with life.)

In either case instead of subjecting an AI to a Turing test you could instead incorporate it into a delayed choice quantum eraser experiment and then delete all its memory before anyone looks at it or the experimental data. If the interference reappears your AI is not conscious if it doesn't then it is conscious. You have used quantum mechanics to prove the existence of another mind and effectively countered solipsism. Of course if machines can't become conscious there are interesting implications. Like say machines can't be conscious but you can still transfer an entire mind into one. Then what becomes of the mind even if it can still pass a Turing test. Worse yet if the universe evolves to the point where only this artificial life remains what happens to causality when any collapsed wave function can revert if everyone that saw it or the consequences of it forgets about it? Alternatively, if machines can't be conscious you might not be able to transfer a mind into them. It also raises questions about maybe the universe evolved in specifically the right ways to create conscious observers or maybe the first life to evolve retroactively fixed the history of the universe the instant it attained consciousness.

In short something really weird seems to be going on here and I'm sure there is a great Scifi story hidden in their somewhere.

[Magus]

Does this have something to do with why a watched pot never boils?

[Sithdarth]

Does this have something to do with why a watched pot never boils?

That's more to do with subjective time measurement via the workings of your internal clock.

[Magus]

I'll be slightly serious and expose my complete lack of knowledge of quantum mechanics: 1. how do we know that time isn't a completely subjective illusion of the human mind, simply a measuring device rather than a "thing", and 2. why does observing photons change how they would have reacted had they not been observed, unless it's in an actual thing that actually manipulates the photons and I'm misreading "observe" as "me looking at something" when it's actually "people looking at things in a special device that lets them look at them and which actually acts on them in some fashion."

EDIT: I think I might have answered my own question since the quantum eraser experiment involves shooting light through a crystal at least, right? Or photons or whatever.

[BitVyper]

It's something that cannot be measured and makes no predictions.

What if you can figure out and define what it is about consciousness that is inherently different?

Out of curiousity, would you consider a computer that developed consciousness gradually in response to external stimuli (like those little robot communities they made on the basis of finding food and avoiding poison, but with actual consciousness) less artificial than one that could be programmed so from the start?

[Geminex]

1. how do we know that time isn't a completely subjective illusion of the human mind, simply a measuring device rather than a "thing"

I guess we could assume that. But it wouldn't be fun. Also, you couldn't use that assumption to blow up cities or build AI.
Killjoy.


2. why does observing photons change how they would have reacted had they not been observed, unless it's in an actual thing that actually manipulates the photons and I'm misreading "observe" as "me looking at something" when it's actually "people looking at things in a special device that lets them look at them and which actually acts on them in some fashion."
The entire principle of quantum mechanics is that we change things by knowing. Even if the method by which we observe it doesn't affect whatever we're observing, the fact that we're observing it does.

Edit: Also, I hate Sith. Before this I was trying to think of more global warming puns. Now I'm busy comprehending that (simply put) information can time-travel. I'll get back to you on the AI issue.

[Magus]

The entire principle of quantum mechanics is that we change things by knowing. Even if the method by which we observe it doesn't affect whatever we're observing, the fact that we're observing it does.

But why? What "energy" or "force" is acting on it to change it? People aren't psychic, after all. Observation acts on the observer, not the observed. When we look at a star our eyes are catching light from billions of light years away; how is that thing which is affecting our biology instead affecting it? Why does knowing change it? What "force" is at work?

Unless it's like a multiverse thing or something (pretty sure quantum mechanics has multiverse in it, right?)

[Geminex]

It's not so much energy or force as it is... probability.

When an "Either Or" situation occurs on a quantum (that is to say, extremely small, tiniest-things-in-existence-small) level, such as a photon of light nearing a sheet of paper with two closely-placed slits in it, then the result will usually be a combination of both "Either" and "Or".
For the photon going through the slits, if it isn't observed, this'll mean that, contrary to all expectations, it'll go through both slits. Simultaneously. And then it'll interact with itself.

If, on the other hand, we observe it, and find a way to determine which slit it went through, it only goes through one slit, and, thusly, doesn't interact with itself.
The reason for this is that the photon's pretty much a particle and a wave (a probability function) at the same time.

And that's just the basic principle. One of the experiments that Sith's posted above proves that not only does observing the experiment change the result, but it will change the result retroactively. Pretty much, if you're going to observe it, then the result you get will be the result you would have gotten through observing it, even if you haven't actually observed it yet.

Apparently, information can fucking time travel.
Because fuck you, Einstein.

[Sithdarth]

But why? What "energy" or "force" is acting on it to change it? People aren't psychic, after all. Observation acts on the observer, not the observed. When we look at a star our eyes are catching light from billions of light years away; how is that thing which is affecting our biology instead affecting it? Why does knowing change it? What "force" is at work?

To give you a simple but probably unsatisfying explanation it all comes down to the uncertainty principle. You see there is a set of classical uncertainty equations that relate the various measures of a wave to other measures of a wave. Like wave number and extent in space and frequency and duration of the pulse. These uncertainty equations put a limit on how well you can know one of these quantities depending on how well you know the other. Perhaps a better way to say it is that if a wave packet is packed into a very short distance it will have a very large number of different wave numbers (which are related to wavelength) and a wave packet with a very short duration in time will have a large number of different frequencies. This is just a consequence of how waves work. The mechanics behind waves just mean this happens.

In quantum mechanics everything is both wave and particle so as one might expect these uncertainty relations come into play but in slightly different forms. But again the minimum uncertainty exists simply because that is how Quantum Mechanics works and has nothing at all to do with how you observe the waves. It is these uncertainties that "collapse" when you observe a system. You can find the position but then the momentum goes crazy and vice versa. There is no special thing happening here its just how the mathematics that describes Quantum Mechanics works out and damned if the universe doesn't obey the math just to foil our understanding of what the hell is going on.

Also, I'm really damn tired so please forgive any really strange typos. Like I found some really crazy ones on my last read through and I don't trust myself to have found them all.

[Ryanderman]

Re the OP: You said that:
It does still allow for a definition of consciousness as any system in which there is a certain level of difficulty in erasing information to the point of reverting a wave function. Anything with a difficulty level above a certain threshold is conscious and anything below is not. Finding that threshold might be hard and might exclude or include somethings that are problematic.

And also that:
In either case instead of subjecting an AI to a Turing test you could instead incorporate it into a delayed choice quantum eraser experiment and then delete all its memory before anyone looks at it or the experimental data. If the interference reappears your AI is not conscious if it doesn't then it is conscious.
Maybe I'm missing it, but what sets the threshold of difficulty to determine conciousness above the difficulty of erasing an AI's memory, and below the difficulty of erasing a human's memory?

[Sithdarth]

Maybe I'm missing it, but what sets the threshold of difficulty to determine conciousness above the difficulty of erasing an AI's memory, and below the difficulty of erasing a human's memory?

Nothing. The test doesn't tell you if the machine is not conscious because its not complex enough or if its not conscious because machines can't be conscious. (Unless you can prove somehow its a matter of complexity and prove your machine has enough complexity but still is not conscious.) For all I know the threshold could be at plants or it could be that only biological things can be conscious. Basically the test in and of itself only tells you if its conscious not why or why not.

[Azisien]

I love Sith threads. People post not understanding some of the physics terminology and his explanatory post involves more physics terminology.

I need to start reading up on QM again though, I feel rusty after reading this thread.

[Professor Smarmiarty]

Don't worry- in theory my job is fiddling with quantum mechanics but I don't understand lots of it.
As for the topic- it is certainly interesting but I don't have much to add unfortunately. It seems plausiable enough.

[Ryanderman]

Nothing. The test doesn't tell you if the machine is not conscious because its not complex enough or if its not conscious because machines can't be conscious. (Unless you can prove somehow its a matter of complexity and prove your machine has enough complexity but still is not conscious.) For all I know the threshold could be at plants or it could be that only biological things can be conscious. Basically the test in and of itself only tells you if its conscious not why or why not.
That's not quite where my confusion lies. How does the test tell that the AI is not conscious? You suggested in your scenario #1 that if the information could be erased from a human's mind, it could uncollapse the wave function. And your scenario #2 suggests that the is some siginifcance to conciousness that would cause the wave function to remain collapsed, even if the memory were erased. But then you said that in either case if deleting the AI's memory caused the wave function to uncollapse, it would prove it was not conscious. I guess I'm failing to see how that result is any different from the conscious human in scenario #1.

Sorry to pick at this detail, but I felt like I understood most of what you said, so I want to clear up this sticking point to my comprehension.

[Sithdarth]

Ah in case one it'd be all about how difficult it is to erase the memory. Like if the machine is complex enough you could erase the area where the information is stored in the AI's brain without actually removing the information from the universe. Its kind of a strange thing to think about but information is essentially negative entropy. If in the first case deleting the computers memory doesn't revert the wave function than it is obviously conscious and erasing things from the human mind would probably have the same result. In the case where even erasing human memory reverts the wave function you're essentially screwed in that you know have to arbitrarily choose a point where the difficulty in erasing memory indicated consciousness. Then you perform the test on your AI and try everything up to that level of difficultly. For example you might be able to revert a wave function by killing everyone who observed it or the consequences of it and incinerating their corpses. That would be a very difficult erasing procedure. So you might choose the point of having to destroy the system to erase the data the point of consciousness. If that is the only way to revert the wave function after the AI looked then its conscious if hitting the delete key is enough then its not. Its a bit of a subtly but it still works just a bit more arbitrarily.

The alternative in case one is to accept there is nothing special about consciousness and making the distinction is pointless. But that's just no fun really.

[Ryanderman]

Ah. That makes more sense. Thanks.

[Hanuman]

In a flat world, West is peril.
http://www.flapjackfan.com/assets/images/wallpapers/west_800x600.jpg

[GrandMasterPlanetEater]

Ah in case one it'd be all about how difficult it is to erase the memory. Like if the machine is complex enough you could erase the area where the information is stored in the AI's brain without actually removing the information from the universe.

Oh, that's easy. Just use "cut/paste" instead of delete.

I learned that from Clippy! I'm sure he'll teach me about quantum chromodynamics next.

[PyrosNine]

Does this long rant have anything to do about that novel where some guy makes an alternate universe by making digital copies of himself and others, and then deleting them en masse, creating a universe that only they observe? That then goes to shit when the people also end up creating life forms in this digital universe, who end up not believing that they (the people) exist, and begin wiping them from that existence as well?

Darned if I can remember the name of the book, but it was weird and seemed to be built on that principle, but followed the assumption that it would be easier to create a virtual duplicate intelligence of a human than create an AI. It had some keen visions of what it'd be like if the Matrix had low bandwidth (a week in digital time would actually be years!)

[Krylo]

In a flat world, West is peril.
http://www.flapjackfan.com/assets/images/wallpapers/west_800x600.jpg

...I've understood something like 80-95% of what Sithdarth has said here.

This, though?


This... is... I don't even...

[MasterOfMagic]

In flapjack, there's an episode about the West. Crazy, crazy things happen there that don't make any sense.

I think he's comparing the current discussion to that.

[PyrosNine]

That book I was talking about? I found it's name. Permutation City. (http://en.wikipedia.org/wiki/Permutation_City)

I think this is something similar to what Sithdarth was talking about. At least more than antiverse flapjack.

[Sithdarth]

So I had a thought the other day. Its been said that our concept of past and future is derived in no small part from thermodynamics. That is the second law of thermodynamics tells us that the future is the direction of increasing entropy. Now when we remember something we do so at the cost of greatly increasing the entropy of the universe. Well greatly compared to the decrease in entropy we introduce through adding the ordered information to our memories. Since entropy must increase to form memories we must always remember the past as the direction in which entropy decreases.

Anyways this got me thinking that there is a certain minimum amount of entropy for a given bit of information. That is to say every object carries all the information there is to know about it in it's structure and the entropy needed to create that object in those conditions is the absolute minimum. Now on the quantum level you can make copies of the state of a particle, i.e. its information via entanglement at this absolute minimum cost. However, actually storing the data in any conventional way would cause a far greater entropy increase. I contend that the only time information is erasable to the point of being able to revert a wave function is when that information was created at the minimum entropic cost. So anything that stores information in an entropically unfavorable way could be considered conscious at least as far as its effect on the universe.

Which begs the question of data storage in quantum computers. If quantum computers where to use entanglement to store and manipulate data and that data might be created at this entropic minimum. If so then as long as its not removed from the computer in an entropically unfavorable way if the computer were to observe something and then delete the results it would fail the consciousness test. Basically while a human mind in a really powerful normal computer might pass the test a human mind in a fully quantum computer might fail the test. That is regardless of their respective cognitive powers. I also wonder what would happen if the result of a quantum observation were say used to generate a random number that was then displayed after all data pertaining to the generation of the number was erased. Such a thing might revert the wave function used to generate the number and that effect could travel into the future changing the number in unexpected ways. Really any sort of calculation could be subject to the past interference if something wasn't done at the end of each processing cycle to ensure that all states used in the calculations stayed the way they were supposed to.

[Hanuman]

In flapjack, there's an episode about the West. Crazy, crazy things happen there that don't make any sense.

I think he's comparing the current discussion to that.
The basic point was that I was comparing current metaphysics to the flat world concept.

[GrandMasterPlanetEater]

Sith, why don't you try posting these ponderances on the xkcd forums? They seem like the type to go over this kind of stuff with a fine tooth comb.

[bluestarultor]

Sith, why don't you try posting these ponderances on the xkcd forums? They seem like the type to go over this kind of stuff with a fine tooth comb.

Because then he wouldn't have the satisfaction of being the guy who knows everything scientific here. :p