Robot consciousness

[Piggy]

Once again you don't understand computable, and you seize on irrelevant aspects. Computable does not mean deterministic, it does not mean rigid, it does not mean an inability to handle fuzziness. And certainly physical robustness has nothing to do with it. Finally, if neurons are computable, they are computable. We are talking about equivalence, not identity.

If there's something I don't understand, then by all means, please explain it and its relevance.

But what jumps out at me is your unwillingness to actually look at the brain.

 

[Piggy]

Still computable.

Ok, great. Relevance?

 

[Piggy]

I wonder - I read a book that I now forget, by a prominent language theorist, using arguments much like this. What a terrible book, because he understood nothing about computability. I wonder if you have been influenced either by him or the field in general. Because you have said nothing that is not computable. "Messy" "fluid" "open" are all ill-defined words in the space of information theory. More importantly, nothing you are describing is uncomputable. The book talked about things like Excel, and how it was exact, created the same result every time, and imagine if your taxes were computed differently each time. Sure, but only because the algorithms chosen were for computing taxes. What a misunderstanding of computing - the same misunderstanding you are showing.

Why do you think that an understanding of information theory is all you need to understand how the brain creates consciousness?

ETA: That bit about Excel sounds like horribly bad thinking to me, too.

 

[Piggy]

Piggy, as an aside, i think there is a lot of misunderstanding going on because Paul, I and others are referring to concepts that we are very familiar with, and that many books have been written about. When Paul or I say "pencil brain" we know and understand the 100 implications we both mean by that. When Paul says slow the brain down, we understand that we are not talking about the implementation domain, where for a specific implementation you cannot run slower than the impulse speed and duration of the signal. After all, what a boring question to ask - for any given substrate of course there is a speed to slow and a speed too fast. Can you imagine starting a JREF post - can I rev my engine too fast? Well, yes! Duh! Or "Can I run my car engine at 0.000001 rpm" - NO! But we could make an engine to do that, if we wanted.

So from my perspective you are arguing extraordinarily strange things - like comparing a pencil brain to a fart. But then I have at least 10 books by leaders in the field under my belt on this one topic alone, and dozens more on computational theory and the like. I guess I can see where you are coming from if you don't recognize the referents, but on the other hand, recognize we are talking in professional shorthand. Pencil brain for us is a UTM. It's a useful thought experiment because it challenges preconceptions - "how could a pencil think" type feelings. Of course, we aren't saying the pencil thinks, but the system produced by the pencil. It gets right to the crux of the matter.

A physicist might say "acceleration times time is velocity" - in that statement is the assumption that we aren't at relativistic speeds, that we are dealing with macro objects where Heisenberg effects are below our measurement accuracy, all kinds of things that don't need to be explained. A literalist JREFer, fresh from reading a bit of Einstein for the first time, hopping into the conversation, would be sputtering "but relativity states....", etc.

There are genuine misunderstandings about computation in this thread as well, but a lot of the argument is of this nature.

Then it would be courteous to do some explaining.

And btw, I have to confess that what I'm seeing from my POV is a lack of understanding about the brain, the only real-world example we have of a thing which actually creates consciousness.

Perhaps it would be best if we tried to make ourselves understood to each other.

If you would take the time to describe the details of this "pencil brain", then I'll be happy to explain whether it seems plausible that such a thing could do what the brain does when it generates conscious awareness.

 

[Piggy]

This is what we have been talking about all along, in the shorthand of 'pencil brain'. But since that is a sticking point for you, try super-big-blue-robot instead without worrying about Turing's math, or 'virtual' vs physical.

Great. Done. Now, regarding this SBB robot, what is the question?

 

[Piggy]

We were talking about a hypothetical world, right? In this world, it is actually possible to use billions of pencils and paper with one for each neuron, and each neuron is described in terms of when it fires, based on what input, and what output the firing results in.

Ok, let's drill down here.

In this hypothetical world, how exactly does any given pencil-paper pair (p3) work like a neuron?

Thanks.

(I know it's cumbersome, but we can't just gloss over that kind of detail if we're going to posit a simulation.)

 

[Piggy]

This huge paper machine will in principle be able simulate a brain complete with consciousness, but obviously each simulated millisecond will take a few centuries to finish in real time.

What does it mean to "simulate" consciousness?

That's a tough nut, because consciousness is a sense of experience.

Are you saying the machine would be conscious?

If not, how has consciousness been simulated?

 

[Piggy]

I can't believe you are asking this question - or I may have no idea what you are thinking about! Practically anything you do on a computer is a virtual simulation. A pencil/paper simulation could never be an actual simulation of anything that does not involve pencils and paper!

That's exactly why I'm asking.

Now, I had my own assumptions earlier when I said there could be no pencil brain with pencil thoughts.

When I said that, I was speaking entirely from the frame of reference of consciousness -- which is a pretty small part of brain activity.

By "thoughts" I mean specifically "conscious thoughts". I don't see other types of thoughts as relevant.

Of course you can make a pencil brain, a cog brain, an abacus brain.

But the OP stipulates a robot which is conscious in the way we are.

So what I'm trying to get at is specifically whether the pencil brain can be conscious.

To do that, it has to do something very specific.

 

[Piggy]

It is essential for the simulation to be able to simulate every single element that is part of the consciousness in a real brain. Neurons are fairly simple as far as we know, and they are governed by simple rules, which is excellent for simulations. However, as long as we do not know for sure how exactly to achieve consciousness, we also cannot be sure that we have got all elements right. For this theoretical paper machine to be certain to work, all neurons will have to be simulated, and if neurons have more sophisticated functions than we know today, these would have to be simulated too. If there are other cells that have a function in consciousness, these too will have to be simulated.

Once we know exactly how to achieve consciousness, ie, we have a working CTM, then we may be able to reduce the number of elements, both in types and quantity, and this is what is the goal for CTM, because obviously, a paper simulation, or even a super fast complete computer simulation of a brain is too impractical for us at this stage.

Well, let's start with what we know and stick with the thought experiment.

Could your pencil brain perform the kinds of tasks that we know the brain is required to perform in order to generate conscious experience?

Fwiw, I don't think the neurons matter much.

 

[Piggy]

We do not have to discuss the pencil brain thing. It really does not matter on what hardware the simulation runs. The important point is really that we are of course talking about a virtual simulation, and all attempts at simulating consciousness or intelligence have been virtual. One day we could probably do actual simulations on biologically simulated brains, but I think that virtual simulations are much easier to implement.

But the OP is not concerned, and neither am I, with a virtual simulation of consciousness. That's what I've been on about all this time.

The OP posits a robot which is conscious in the way we are.

So any questions regarding such a robot will have to assume an actual instantiation of consciousness.

 

[Piggy]

Some guys not so worried about tenure:

http://www.ncbi.nlm.nih.gov/pubmed/17285879

http://www.ncbi.nlm.nih.gov/pubmed/...nel.Pubmed_DefaultReportPanel.Pubmed_RVDocSum

Also interesting:

http://www2.computer.org/portal/web/...9/FBIT.2007.71

Very cool!

 

[Piggy]

roger:

To clarify a couple of points.

When I've been talking about a "brain" and "thoughts" on this thread, I've been meaning a conscious brain and conscious thoughts.

I don't doubt that a TM can be called a kind of brain, if you like.

You say that your pencil brain can do anything a human brain can do, because both are computable.

But that assumes that speed must be irrelevant. In other words, that nothing the human brain does is dependent on speed.

And that's what I've been trying to focus on.

I think we're butting heads because you're focused pretty much exclusively on information theory, and I'm focused on how the brain appears to create the phenomenon of consciousness.

 

[Piggy]

<Presses reset button>

Ok, after all that jabber (my apologies)....

Can we reframe the question to clear the tangents?

Consider a conscious robot with a brain composed of a computer running sophisticated software. Let's assume that the appropriately organized software is conscious in a sense similar to that of human brains.

Would the robot be conscious if we ran the computer at a significantly reduced clock speed? What if we single-stepped the program? What would this consciousness be like if we hand-executed the code with pencil and paper?

Point 1: This robot's brain, regardless of what it is built of, must be assumed to produce consciousness in basically the same way the human brain does, because that's the only way we know that it can be done. (And it would make no sense to ask: "There's a robot with a brain that produces consciousness by unknown means -- what happens to its consciousness when we slow the processing rate?")

Point 2: The question becomes equivalent to: "If we slowed the human brain's processing down to the equivalent of pencil-and-paper speed, would it still produce consciousness?"

Point 3: We can imagine building a brain using other materials with each neuron replaced by some other (much slower) mechanism that takes equivalent inputs and produces equivalent outputs, all arranged in the same sequential configurations we would expect to find in a human brain. We feed appropriate input to this brain to simulate typical input into a human brain.

Point 4: Both systems are computational.

Point 5: The mechanical brain can therefore do everything the biological brain does iff nothing done by the biological brain is dependent on being performed at a speed greater than that achieved by the mechanical brain.

Point 6: If consciousness in the biological brain somehow depends on a signaling speed higher than that achieved by the mechanical brain, then the mechanical brain will not be conscious. If not, the mechanical brain will be conscious.

Agreed?

If not, why not?

Thanks.

 

[roger]

roger:

To clarify a couple of points.

When I've been talking about a "brain" and "thoughts" on this thread, I've been meaning a conscious brain and conscious thoughts.

I don't doubt that a TM can be called a kind of brain, if you like.

You say that your pencil brain can do anything a human brain can do, because both are computable.

But that assumes that speed must be irrelevant. In other words, that nothing the human brain does is dependent on speed.

And that's what I've been trying to focus on.

I think we're butting heads because you're focused pretty much exclusively on information theory, and I'm focused on how the brain appears to create the phenomenon of consciousness.

Well, to my point of view speed is an implementation detail, and thus pretty uninteresting.

To expand on that, no, a pencil brain, actually and really implemented by a person sitting at a desk with a pencil, or by a TM, will not be conscious, or even work for no other reason that in a person's lifetime they won't even be able to compute one state of the brain, let alone have it do any processing. It'd take more than your life just to write the starting condition on paper.

So, yes, on a practical level a real pencil brain wouldn't work, especially if you were trying to process data in our world, real-time.

But, this whole thread is about thought experiments. I'm free to assume that I have an infinite life, essentially infinite paper (you'd need a very large, but finite amount of paper to simulate the brain) and the speed of inputs would have to be very, very slow compared to you. So, accelerate to relativistic speeds, tunnel to a multiverse, whatever it takes to get that time dilation.

What we are trying to get to here is whether the brain and consciousness is nothing more than a consequence of matter - a completely non dualistic position. Sure, in this world a pencil brain is ridiculous - it'd never process data fast enough, but we are interested in principles. So, the chain of reasoning goes like this:

1) the universe at a macro level is computable
2) everything in the brain that we have observed - neurons, chemical soup, the networks, are computable, and not susceptible to noise that would make them non-computable
3) consciousness comes from the brain ad it's coordinated activities
4) So, consciousness is computable, or there is something 'extra' in the brain we have never observed
5) hence, it's a very strong position(meaning likely) to state that a computer could be conscious, so long as it was structured the same as a brain

Note there is no dualism in 5. We don't require a person to interpret the symbols in the brain, or the computer. The organizational structures in the brain, their self-referential way of computing, is the consciousness (I have trouble talking about 'creating' consciousness, as I don't think of it as a 'thing', to me consciousness is just the brain's machinery doing its thing).

Now, information theory tells us a TM - a pencil, a paper tape, a motor, and just a tiny handful of operation, can do anything computable. So, a person with a pencil can do the same, following rules blindly, neither understanding what they are doing, why they are doing it, or the meaning of all the 1s and 0s they are writing down.

When we talk of a pencil brain, we mean formalizing all of the systems, components, and data in the brain as a set of computable functions, and then implementing it as a TM and associated program. Realistically, TMs are really slow, and we'd really write a program on a supercomputer, but there is nothing stopping you from doing a TM or a person and a pencil.

And this sort of tests your convictions here, because it is very counterintuitive. Because my claim is that form of a pencil brain, a human with a pencil, would indeed be conscious, though that consciousness would stretch out over eons. Certainly as the person sat there and wrote 1s and 0s, there would be nothing you'd call consciousness, but then are you 'conscious' during a single planck time interval (~10^-44 secs)? I'd say no more than a pencil brain would be conscious over a year period. But that doesn't mean you are not conscious over a second, and that a pencil brain is not conscious over many eons. Consciousness is nothing more or less than the processes running.

So, going back to the OP, you can see I hope that Paul was asking questions about the pencil brain. Remember, computationally, the pencil brain is exactly the same as the biggest baddest supercomputer you can imagine, just a heck of a lot slower. So, if a pencil driven by a TM, or a human blindly following a program, is doing exactly the same thing as 1) a supercomputer, and 2) your brain, does the fact that the pencil goes a lot slower mean no consciousness will form? Again, this is a thought experiment, no bringing up the speed of impulses, or the indubitable fact that when trying to simulate 100billion neurons by hand the pencil marks for neuron one will fade long before you reach the 100 billionth neuron.

I say yes, the pencil brain will be conscious, measured in appropriate timescales (millions of eons, not seconds). Our year will be it's planck time. But that is just a scaling issue.

If you say no, and not for some practical concern like who has enough paper to write the state of 100 billions neurons, or pencil fade over several hundred years, then you must be introducing something extra - something noncomputable - in the brain that creates consciousness.

Is this clearer?

edit: since we've strayed so far, I'm going to quote the OP, and give my answer, with all of the above informing it. perhaps that'll add nothing, but perhaps it'll make it clear why pencil brains have been such a topic for us

 

[steenkh]

So there's this notion that consciousness "emerges" simply by virtue of having a bunch of neurons in one place.

But that's not the case. Consciousness is something the brain does, like vision or motor coordination or regulating breathing. It's a specialized function.

So nothing but brains can have consciousness, by definition? Is this in the same way that only eyes can have vision because it is a specialised function?

 

[roger]

Consider a conscious robot with a brain composed of a computer running sophisticated software. Let's assume that the appropriately organized software is conscious in a sense similar to that of human brains.

Would the robot be conscious if we ran the computer at a significantly reduced clock speed? What if we single-stepped the program? What would this consciousness be like if we hand-executed the code with pencil and paper?

I can't take credit for these questions; they were posted on another forum. The following paper is relevant to this issue:

http://www.biolbull.org/cgi/content/abstract/215/3/216

~~ Paul

Interesting question, Paul.

I start from the assumption that the brain is purely physical. There is nothing 'extra' that provides consciousness. I'm assuming you make the same assumption, as you are positing a computer that is conscious the same way we are.

All of our studies show us that brains are comprised of macro structures which perform computable functions - we have never identified a neuronal behavior, for example, that was not computable.

The Church-Turing thesis tells us that any algorithm has an equivalent TM or lambda function associated with it. Sure, this has not been proven, but name me a serious researcher that does not accept Church-Turing. Certainly we know of no counterexamples.

Now, there are two features of a TM that are applicable to your question. First, there is no term for execution speed in a TM. A TM could perform 10^66 operations a second, or 1 an eon, and it would still be a TM, and the algorithm would still be computable. Naturally, for any given TM implementation you have to make the speed applicable for that medium - no second long brain impulses, no silicon impulses that are required to be FTL, etc. But there is nothing intrinsic about computation speed in the abstract.

The second feature that applies is the fact that a TM is nothing more than a pencil pusher. It consists of a paper tape, a pencil to write 0 or one on the tape, and a few instructions to move the tape, read it, write on it, and perform a few conditionals. We could easily substitute a person to push the pencil around according to a program, if we liked - give them a description of a TM, and tell them to execute the following instructions based on the data stream to be provided.

So, combining those two points, a person pushing a pencilaround on a piece of paper according to specific directions can do anything computable. If, as I argue, the brain is computable, and that consciousness arises from computable processes, we must conclude that consciousness is speed invariant within the limits imposed by the substrate hosting the computations - brains gotta keep the neuron impulses going.

While our brains don't appear to be synchonously clocked, we can impose the planck time interval on our brains - roughly 10^44 seconds. We could take a snapshot of the brain for every interval, and say that is our clock speed. During any one of those intervals, surely there is no 'consciousness'. I introduce scare quotes because consciousness is not a thing if it is computable - it is the result of the brain's systems interoperating recursively and self-referntially. yet, over say 10^44 of those intervals, we have consciousness.

So, even though it seem intuitively absurd to talk about paper and pencil having consciousness, I don't see how we could reach any other conclusion if we grant my assumptions. No, it won't be conscious for any small time duration, such as a single human life, but over many of them, say 10^44 of them, it would be.

Of course, this is the point where many people balk - how could a pencil, paper, and about 10 simple rules produce consciousness? it does seem crazy, but there it is, since we know there is nothing a supercomputer can do that the paper, pencil, and handful of rules cannot do, and vice versa. Our brains are matter, and so far as we know everything the brain does can be described as an algorith, so, a computer should create consciousness, and thus so should pencil and paper (given appropriate time scales).

Similarly, if we single stepped the computer consciousness would still be there. Sure, the time scale of that consciousness would be different. Say you stepped it at 1 instruction/second. It would be incapable of being aware of that time scale, just like our brains are unable of perceiving 10^-44 of a second. But the computer would still be conscious at a much longer time scale - eons perhaps, or millions of eons.

Of course, this is all off if it turns out the brain is not computable. Physicists argue whether the universe is ultimately discrete or not. I fall into the discrete camp, but oddly enough I can't find my name in the Nobel prize register. I'll have to ring them up. In any case, by the time you reach macro structures in the brain (and by macro I'm talking molecules and such) we can deal with it discretely, and thus everything in the brain is algorithmic. Finding out otherwise would invalidate a heck of a lot of physics, chemistry, biology, and medicine that we currently do.

 

[steenkh]

In this hypothetical world, how exactly does any given pencil-paper pair (p3) work like a neuron?

By simulating all the functions of a neuron.

(I know it's cumbersome, but we can't just gloss over that kind of detail if we're going to posit a simulation.)

I think you do not really understand that I am not saying that we know all of the functions of a neuron. Obviously, such a simulation can only be performed once we are sure that we know how neurons work down to the last detail.

I never claimed that the paper machine was something we could make right now, or ever.

It is not even certain that the brain works like a Turing Machine.

When I said that, I was speaking entirely from the frame of reference of consciousness -- which is a pretty small part of brain activity.

I am not sure that it makes sense to distinguish consciousness from other parts of brain activities. If consciousness really is an emergent function of masses of neuron activity, then you need all of that activity to get consciousness.

By "thoughts" I mean specifically "conscious thoughts". I don't see other types of thoughts as relevant.

I do not think it is meaningful to distinguish conscious thoughts from other thoughts, whatever they may be.

There has been some experimentation to show that people "make decisions" before they are "aware" that they make a decision. This sounds rather futile for me. Unconscious thoughts making decisions are as much part of consciousness as any other kinds of thoughts, so it really tells more about how we experience consciousness than what consciousness consists of.

Of course you can make a pencil brain, a cog brain, an abacus brain.

But the OP stipulates a robot which is conscious in the way we are.

What does "conscious in the way we are" mean? According to you, only brains can have consciousness, so per definition, robots cannot be conscious, and this whole thread is futile.

So what I'm trying to get at is specifically whether the pencil brain can be conscious.

Yes. And I have given my opinion. As long as we do not know exactly how the brain achieves consciousness, it remains an opinion.

To do that, it has to do something very specific.

I fail to see why simulating every neuron and other elements in the brain that contribute to consciousness would not be "something very specific".

Well, let's start with what we know and stick with the thought experiment.

Could your pencil brain perform the kinds of tasks that we know the brain is required to perform in order to generate conscious experience?

By definition, yes. This means that all elements are simulated properly, and that there are no non-computational elements involved in real consciousness.

Fwiw, I don't think the neurons matter much.

OK. As long as the paper machine simulates all relevant elements, it would still generate consciousness.

But the OP is not concerned, and neither am I, with a virtual simulation of consciousness. That's what I've been on about all this time.

The OP states: "Let's assume that the appropriately organized software is conscious in a sense similar to that of human brains." Where do you read that the OP is not concerned with a virtual simulation of consciousness? How do you interpret the word "similar"?

The OP posits a robot which is conscious in the way we are.

No. and by your definition that would be impossible anyway.

 

[Piggy]

What we are trying to get to here is whether the brain and consciousness is nothing more than a consequence of matter - a completely non dualistic position.

Ok, I'm sorry we got off on that tangent, then, because of course that's true.

So that was totally off the radar for me. I was only concerned with the question of speed, which is what the OP appears to be asking about.

 

[Piggy]

1) the universe at a macro level is computable
2) everything in the brain that we have observed - neurons, chemical soup, the networks, are computable, and not susceptible to noise that would make them non-computable
3) consciousness comes from the brain ad it's coordinated activities
4) So, consciousness is computable, or there is something 'extra' in the brain we have never observed
5) hence, it's a very strong position(meaning likely) to state that a computer could be conscious, so long as it was structured the same as a brain

Agreed there, too. I don't see why it will not be possible for consciousness to be manufactured one day.

In fact, the thought experiment with the robot assumes it. We never disagreed on that.

 

[Piggy]

Because my claim is that form of a pencil brain, a human with a pencil, would indeed be conscious, though that consciousness would stretch out over eons. Certainly as the person sat there and wrote 1s and 0s, there would be nothing you'd call consciousness, but then are you 'conscious' during a single planck time interval (~10^-44 secs)? I'd say no more than a pencil brain would be conscious over a year period. But that doesn't mean you are not conscious over a second, and that a pencil brain is not conscious over many eons. Consciousness is nothing more or less than the processes running.

Ok, this for me is the crux of it.

There's an easier (for me) thought experiment that can be done, which is just to posit that, somehow, neuronal activity in a human slowed down to that speed with no adverse affects to the body.

Same thing, right?

So to answer the question "Would the subject still be conscious?", we have to examine how the brain creates that effect and ask if it would work at that signaling speed.

That's what I've been on about.

You've been asserting that the function of consciousness can work if the singaling slows down to glacial levels, and positing a pencil brain that is somehow "conscious over many eons". But can it?

To answer that question, it's not enough to note that both the human and pencil brains are computable. We have to look at how the brain creates the effect and ask "Can this happen at extremely slow signaling speeds?"

So, going back to the OP, you can see I hope that Paul was asking questions about the pencil brain. Remember, computationally, the pencil brain is exactly the same as the biggest baddest supercomputer you can imagine, just a heck of a lot slower. So, if a pencil driven by a TM, or a human blindly following a program, is doing exactly the same thing as 1) a supercomputer, and 2) your brain, does the fact that the pencil goes a lot slower mean no consciousness will form? Again, this is a thought experiment, no bringing up the speed of impulses, or the indubitable fact that when trying to simulate 100billion neurons by hand the pencil marks for neuron one will fade long before you reach the 100 billionth neuron.

I say yes, the pencil brain will be conscious, measured in appropriate timescales (millions of eons, not seconds). Our year will be it's planck time. But that is just a scaling issue.

If you say no, and not for some practical concern like who has enough paper to write the state of 100 billions neurons, or pencil fade over several hundred years, then you must be introducing something extra - something noncomputable - in the brain that creates consciousness.

I say no, and for entirely practical reasons.

The step which, from my point of view, you are failing to make is to actually go look at how consciousness is created and ask, "Is this a function that can be maintained by something that works as slowly as a pencil brain?"

I say no, because you don't have simultaneous coordination of large enough amounts of coherent data over short enough continuous spans of time to achieve it.

That's why I say that the pencil brain cannot mimic that particular function.