Robot consciousness

[Piggy]

Given what us "TM-ers" mean when we talk about behaviors or outputs, then a behavior *IS* in fact an output.

A stimulus is an input...

...upon which some type of information processing occurs, to produce...

...a response, which is an output.

If the response happens to be physical movement (intentional or reflexive) then the response can rightly be called a behavior. All behaviors are outputs, even if not all outputs are behaviors.

Then I must have misunderstood drkitten's remark that there are no high level outputs.

And if all high-level behaviors are outputs, then I see no reason to conclude that all such outputs must be sustainable at any operating speed.

Tell me, how do you propose that a "sustained reverberant state of coherent activity" will be achievable by the brain in real time if neuronal activity is slowed to the point where real-time cohesion fails?

 

[Piggy]

I would say the the behavior of a system was characterized by its response to stimulus.

You analyze a behavior by viewing the output but without also knowing or controlling the input you won't get far.

In this case, it is stipulated that the input remains constant -- except for Joe and Jane in the blackout room, where input is extended in real time but otherwise remains constant.

The real question here, it seems to me, is the effect on the generation and maintenance of conscious awareness if data streams that are coherent in real time are drastically slowed so that signaling loses its coherence (from the point of view of the brain functions that are built to handle these data streams).

 

[steenkh]

Tell me, how do you propose that a "sustained reverberant state of coherent activity" will be achievable by the brain in real time if neuronal activity is slowed to the point where real-time cohesion fails?

I do not think that anybody are suggesting that a physical computer like the brain will continue working if slow down activity to a point where the physical components fail. Even electronic computers cannot run at arbitrarily slow clock frequencies.

A TM is a theoretical construct that can run arbitrarily fast or slow.

 

[Philosaur]

This thread just keeps looping back around again and again.

Piggy, no one here--as far as I can tell--is disputing that a human brain has a minimal functional speed (Piggy's Constant?) to produce consciousness. To be clear, this is equivalent to the following scenarios:

If a juggler's hands are slowed down enough (while gravity acts with its usual force), he will be unable to juggle.

If the pistons of an internal combustion engine are slowed down enough (while the rate of combustion remains the same), the engine will cease to turn.

But these speed limits are merely an accident of their particular hardware implementations. They are not inherent in the higher-level processes (i.e. keeping balls in the air, moving a vehicle). There are other ways to produce the higher-level processes where these particular speed limits are not a factor. Granted, some other speed limit will surely come into play, because physical processes happen in space-time and so are subject to the laws of physics. So a robot brain will surely be subject to *some* speed limit, below which it will cease to function--but the limit is an accident of the hardware implementation.

The point that I and others here have been trying to make is that it is provably possible to do all of the information processing a human brain can do on any hardware that can act as a TM. Furthermore, most of us who understand the concept of a Turing Machine probably also believe that consciousness just IS information processing.

You've said repeatedly that you believe that machine consciousness is possible.

But then you say that consciousness can only be produced by hardware sufficiently like a human brain such that it, too, would be subject to Piggy's Constant. If it's subject to Piggy's Constant, then why not stipulate that it would also operate within the same temperature range as the human brain, require oxygenated blood, require organic neurons, etc.?

Why hammer away at the notion of a common speed limit, and not say that all other things must be sufficiently similar, too?

 

[shuttlt]

Just a thought, but there seem to me to be two ways of stating this speed limit problem:

1. Every information processing system that has ever been produced has physical limitations that prevent it's operation once it is slowed down beyond a certain limit.

2. There is no limit (above stationary) to how slowly we can insist our information processing system run for which we could not design an information processing system that would function beyond that limit.

Definition 1 seems to be what Piggy is describing. Does anybody disagree with definition 2?

 

[Pulvinar]

I'm afraid I'm going to have to ask you back up and restate your idea here, because I'm lost. That sentence seems to be saying conscious experience, on the one hand, and a category of retrievable information, on the other, are equivalent.

That's correct, for clear-cut cases. Do you have any evidence that would contradict that?

Your misunderstanding seems to hinge on your idea of consciousness as reverberation, and that reverberation couldn't be sustained by a TM-equivalent machine at low clock rate. I don't subscribe to the reverberation theory, but assuming it were true, there's no reason reverberation can't be done by a TM. In fact, reverberation in music is usually created nowadays with digital reverberators using TM-equivalent signal processors. They sample the sound at intervals and put that into an array, and then retrieve values from earlier-stored samples to create the effective delay. Single-stepping this process doesn't change the process, only its relation to the timing of external events.

 

[rocketdodger]

But we should be careful to always keep in mind that IP is an abstraction for what happens in pysical reality. It makes sense only to us (to humans) because it's our metaphor.

Wrong.

An abstraction is still physical -- it is a class of physical things. What we abstract as information processing is still, in every occurence, an actual physical process.

In objective physical reality, there's no IP going on, just chemical reactions, electric conduction, and other purely physical sorts of stuff.

And if that purely physical sort of stuff happens to behave in a certain way, it is information processing. Just like if a purely physical sort of stuff happens to behave like a crystal, it is a crystal.

I had this discussion already with westprog, and he failed miserably because of this: what is "crystalline" and "crystallization?" What is the difference between those equivalence classes and IP?

Similarly, when we look at what computers actually do, again, all we see is physics. A computer doesn't work because "information" makes it work. It works because of how it's built as a physical object.

But the equivalence class "information processing" is simply a class of behavior, one which the behavior of a computer definitely satisfies. So "how it is built as a physical object" is as an information processor.

Now, to human beings using computers, there's a layer of IP that we can discern that's useful to us. But that's entirely symbolic, and it's not what makes computers work.

You have it totally backwards -- the behavior of a physical system that makes computers work is what we humans label "information processing." It is what makes computers work, and our symbolism is just that -- our symbolism.

A sufficiently intelligent alien observer with enough access to the physical details of the system -- or a hypothetical omniscient observer -- could describe everything a computer does (such as causing pixels to light up on screens or causing printers to spray ink on paper or causing speakers to generate sound waves or causing trays to extend and retract or making discs spin and lasers lase) in terms of the materials and electricity. (If you believe that the behavior of a computer does not make sense entirely when viewed objectively in this manner, I'd like to hear why.)

Ditto for the brain. If you could know everything about the physical state of the system, everything it did would add up. From birth to death, there would never be a moment when you'd stop and say, "Whoa, why did THAT happen?! That goes against the laws of physics!"

And you know what... in both cases, you wouldn't need to know a single thing about the "information" that we humans talk about when we talk about brains and computers.

Ahh.. here you are very wrong, but you are also very close.

In both cases you wouldn't need to know a single thing about the human terminology and symbolism. You would still need to know everything about the actual physical entities.

And that is the crux of the matter -- information processing is simply a label we apply to a physical system that behaves in a certain way. Get rid of all humans, and the system still behaves in the same way. We just wouldn't be around to call it anything.

Let's get real simple. Take a man adding on an abacus. We can talk about that in terms of IP. But that only makes sense to the human mind. In objective reality, everything that's going on is consistent with the laws of physics, and explainable in those terms. Not just the movement of the beads, but also the movement of the man's body.

So what? The language you use to explain the behavior is irrelevant. The fact remains -- the behavior of a man adding on an abacus is vastly different from, for example, the behavior of a star.

As far as we know, everything in this universe is just a physical-energetic chain reaction, and that includes our brains.

Yet, we can partition all this stuff according to the class of behavior the systems exhibit.

Otherwise, everything would just be mush, and that clearly isn't the case.

So my question for you -- and I asked this of westprog as well -- is why the class "crystal" is acceptable but the class "information processor" is not?

So it is not possible that IP is what generates consciousness. IP is purely symbolic. It is our system of understanding and talking about certain things.

But there is no way in which our symbolic abstractions can cause phenomena in the real physical world.

Therefore, we can be certain that it is the physics of the brain, and not IP, which generates consciousness, however it's done.

Which people simply call IP...

Why would you go through the rigamarole of making a statement like "the physical behavior of the physical system we call a brain behaves according to the equivalence class of systems humans call "information processors," which makes it an instance of that abstract class," when instead you can simply say "the brain is an information processor?"

 

[Piggy]

I do not think that anybody are suggesting that a physical computer like the brain will continue working if slow down activity to a point where the physical components fail. Even electronic computers cannot run at arbitrarily slow clock frequencies.

A TM is a theoretical construct that can run arbitrarily fast or slow.

If anyone is suggesting that consciousness can work at an extremely slow operating speed, then I submit they are indeed suggesting that the physical apparatus which makes consciousness possible will continue working if we slow down activity to a point where the physical components fail.

Again, the only way to produce conscious experience that we know of is the way the brain does it. I don't believe anyone has any other proposal to float.

 

[rocketdodger]

If anyone is suggesting that consciousness can work at an extremely slow operating speed, then I submit they are indeed suggesting that the physical apparatus which makes consciousness possible will continue working if we slow down activity to a point where the physical components fail.

But that point can never be reached -- relativity assures us of this.

As long as the relative activity of components that must interact with each other is preserved, you can slow the system to almost nothing and it will function just fine.

In other words if you jump on a spaceship that reaches 0.999999 C, your brain should function normally according to you because all the particles of your brain -- and all the particles it can possibly interact with -- are also slowed.

What you seem to want to say is that if you slow some particles, and not others, that things might fail. Well, obviously, and nobody has said anything to the contrary!

 

[Piggy]

But these speed limits are merely an accident of their particular hardware implementations. They are not inherent in the higher-level processes (i.e. keeping balls in the air, moving a vehicle). There are other ways to produce the higher-level processes where these particular speed limits are not a factor. Granted, some other speed limit will surely come into play, because physical processes happen in space-time and so are subject to the laws of physics. So a robot brain will surely be subject to *some* speed limit, below which it will cease to function--but the limit is an accident of the hardware implementation.

That being the case, how can you suggest that our hypothetical robot can generate conscious experience at any and every arbitrary operating speed?

If the robot produces consciousness in the same way the brain does (at a high level) then it, too, will have to sustain a "reverberant state of coherent activity" in real time.

How do you suggest that this can be done at pencil speed?

The point that I and others here have been trying to make is that it is provably possible to do all of the information processing a human brain can do on any hardware that can act as a TM. Furthermore, most of us who understand the concept of a Turing Machine probably also believe that consciousness just IS information processing.

And in that belief you are demonstrably wrong.

It's ironic that I've been accused of trying to introduce dualism, magic, and pixies into the "IP creates consciousness" hypothesis, when all I've been trying to introduce into it is the brain (biological or artificial), the physical mechanism that makes it possible.

I perfectly understand your point that IP is independent of substrate as long as the substrate can support the logic.

But when you move into the "IP generates consciousness" realm, you're ignoring the substrate altogether and assuming -- with no evidence or reason to support you -- that the logic can go it alone.

What generates consciousness is the physical activity of the brain, not merely the logic independent of that activity. In that regard, it is similar to every other real-world phenomenon we know of.

So if I were to wrongly assert that IP is what plays movies from DVDs, I would wrongly conclude that a computer can play movies from DVDs at very slow speeds, they'd just play them slowly.

In fact, the computer wouldn't play the movie at all, because it wouldn't be able to make the disc spin at speed and it couldn't operate the laser.

Current research demonstrates that something similar is happening in the brain when conscious experience is generated.

If we slow down the operating speed dramatically, I fail to see how we can expect the requisite biomechanics to operate.

 

[Piggy]

That's correct, for clear-cut cases. Do you have any evidence that would contradict that?

Your misunderstanding seems to hinge on your idea of consciousness as reverberation, and that reverberation couldn't be sustained by a TM-equivalent machine at low clock rate. I don't subscribe to the reverberation theory, but assuming it were true, there's no reason reverberation can't be done by a TM. In fact, reverberation in music is usually created nowadays with digital reverberators using TM-equivalent signal processors. They sample the sound at intervals and put that into an array, and then retrieve values from earlier-stored samples to create the effective delay. Single-stepping this process doesn't change the process, only its relation to the timing of external events.

If you don't agree with the research, why not? And what other mechanism do you propose?

As far as the conflation of consciousness and recollection, that's easy to refute.

The subliminal studies I've cited, for example, show that information which is not consciously processed is recorded and can be used later (recollected). And we all know that information which is consciously processed can be recorded and used later (recollected).

On the other axis, not all information processed by the brain, consciously or unconsciously, is moved into long-term memory (dreams most notably).

In addition, when we look at the studies on subjects like Marvin, we see that the neural pathways for conscious awareness and encoding into memory are not identical.

Conscious experience is not a category of retrievable information.

ETA: In addition, the studies on how the brain "fills in" incoming stimuli with stored schema and associations demonstrates that the raw materials of conscious experience are a mixture of real-time and stored data.

 

[Piggy]

In both cases you wouldn't need to know a single thing about the human terminology and symbolism. You would still need to know everything about the actual physical entities.

And that is the crux of the matter -- information processing is simply a label we apply to a physical system that behaves in a certain way. Get rid of all humans, and the system still behaves in the same way. We just wouldn't be around to call it anything.

We agree about that.

 

[Piggy]

Why would you go through the rigamarole of making a statement like "the physical behavior of the physical system we call a brain behaves according to the equivalence class of systems humans call "information processors," which makes it an instance of that abstract class," when instead you can simply say "the brain is an information processor?"

I have no problem with calling it an information processor. My problem is with making logical errors based on that label, specifically the bizarre hypothesis that the logic of information processing is what causes the real-world phenomenon of conscious experience, and drawing conclusions about the effect of operating speed without reference to the requirements of the physical system that must perform the task.

 

[rocketdodger]

We agree about that.

Good.

So now when we talk about it, you need to understand that we really mean "a physical system that behaves in a certain way."

And it is that behavior that is important.

 

[Piggy]

But that point can never be reached -- relativity assures us of this.

As long as the relative activity of components that must interact with each other is preserved, you can slow the system to almost nothing and it will function just fine.

In other words if you jump on a spaceship that reaches 0.999999 C, your brain should function normally according to you because all the particles of your brain -- and all the particles it can possibly interact with -- are also slowed.

What you seem to want to say is that if you slow some particles, and not others, that things might fail. Well, obviously, and nobody has said anything to the contrary!

Oh, but they have. Repeatedly. Because we're not talking about putting Joe on a rocket ship.

We're talking about slowing the operating speed down in real time.

 

[Pulvinar]

If you don't agree with the research, why not? And what other mechanism do you propose?

How about a computer running a particular program? If we want to know just what that program is, we need to do more research, and come to an agreement on just what to label as conscious.

As far as the conflation of consciousness and recollection, that's easy to refute.

I agree. I sure wouldn't equate consciousness with recollection, as I've pointed out a few times now.

The subliminal studies I've cited, for example, show that information which is not consciously processed is recorded and can be used later (recollected). And we all know that information which is consciously processed can be recorded and used later (recollected).

On the other axis, not all information processed by the brain, consciously or unconsciously, is moved into long-term memory (dreams most notably).

In addition, when we look at the studies on subjects like Marvin, we see that the neural pathways for conscious awareness and encoding into memory are not identical.

There are always exceptions and fringe areas in something as complex as the brain, especially in pathological cases. Blindsight, for example, can be seen as not a failure of recollection per se, but a failure to recognize that some information is being recalled. That's the sort of thing that makes it hard to categorize.

Conscious experience is not a category of retrievable information.

You still haven't given me a clear-cut counterexample of this.

ETA: In addition, the studies on how the brain "fills in" incoming stimuli with stored schema and associations demonstrates that the raw materials of conscious experience are a mixture of real-time and stored data.

The problem is in where to draw the line between real-time and stored. Is information held in the global workspace stored or not? Or, how many milliseconds has to pass before it's not real-time? These are judgement calls. Hard to make arguments one way or the other based on them.

 

[Singularitarian]

If anyone is suggesting that consciousness can work at an extremely slow operating speed, then I submit they are indeed suggesting that the physical apparatus which makes consciousness possible will continue working if we slow down activity to a point where the physical components fail.

Again, the only way to produce conscious experience that we know of is the way the brain does it. I don't believe anyone has any other proposal to float.

A Speed Limit on Thought? | Germany | Deutsche Welle | 15.03.2004A group of German neuro-physicists say they've discovered an upper limit on how fast human beings can think. It all depends on how closely connected the ...
http://www.dw-world.de/english/0,336...90_1_A,00.html - 34k - Cached - Similar pages



''Germany | 15.03.2004
A Speed Limit on Thought?

Großansicht des Bildes mit der Bildunterschrift: The brain's neural network imposes a natural speed limit on thought.

A group of German neuro-physicists say they've discovered an upper limit on how fast human beings can think. It all depends on how closely connected the brain's nerve cells are.''

...............................................................................................

Personally, even without the additional information above, i would have personally believed that consciousness requires a certain amount of information processing which would require an upper limit on the speed of consciousness. Consciousness would fail if we where designed any much slower!

 

[Piggy]

How about a computer running a particular program? If we want to know just what that program is, we need to do more research, and come to an agreement on just what to label as conscious.

Since there has never been computer generated consciousness, we have no examples of that.

As for what consciousness is, I've already offered a working definition.

Take the subliminal experiments -- those define a boundary between conscious experience on the one hand and input that is processed by the brain but never experienced consciously on the other.

There's also the cocktail party effect. Streams of nearby (but un-attended-to) speech are processed by the brain but not experienced consciously, until the string of sounds that is your name is part of the input, at which point it triggers enough strong associations that the brain feeds it into conscious awareness and it "jumps out at you".

People who view the "gorilla on the basketball court" film and don't report seeing the gorilla have not been consciously aware of it.

On the other hand, when we look at a wallpaper pattern, we are consciously aware of seeing much more than our eyes actually perceive -- the rest is filled in by schema. We can be consciously aware (and usually are) of things which we're not actually perceiving.

 

[Piggy]

There are always exceptions and fringe areas in something as complex as the brain, especially in pathological cases. Blindsight, for example, can be seen as not a failure of recollection per se, but a failure to recognize that some information is being recalled. That's the sort of thing that makes it hard to categorize.

There's nothing fringe about this. But of course, if we want to put it into sharp relief we have to probe the edges.

As for blindsight, your description does not make sense. What do you mean by "failure to recognize"? By that, if you're referring to the obstacle course experiment, you must simply mean that the brain is using information which is not made available to consciousness, which demonstrates that conscious awareness is qualitatively distinct from non-conscious processing.

 

[Piggy]

You still haven't given me a clear-cut counterexample of this.

Of course I have.

Please go look at the subliminal experiments again, including the one showing the physiological signatures of conscious awareness, which clearly show that the brain processes and uses information which is not served up to consciousness.

Then look also at the studies showing that the brain uses stored patterns to fill in incomplete sensory data to generate a complete conscious experience.

Then look at the studies demonstrating that conscious experience is a downstream process.

All this has been covered and cited.

Conscious experience is clearly formed from a blend of (slightly delayed) real-time input from our internal and external senses associated with stored patterns (memory, schema).

Take a look again at the experiments done with the brain probes which clearly show specialized brain activity coordinated across the brain for events that are consciously experienced but which is absent for events that are not consciously experienced, even though we know from other experiments that the latter (subliminal) events are processed by the brain, recorded into memory, and used for subsequent decision-making.

Consciousness is a function, a behavior, with its own dedicated physiological processes that differentiate it from other types of processing.