Getting from cause effect to awareness

[LegionOnomaMoi]

As I've stated, the redundancy of the human mind is very beneficial to us but it isn't beneficial or even necessary if your trying to create cognition.

That's true. I haven't been limiting my discussion to human brains but to brains. Brains are the cognitive systems. Nothing else comes close.

Curious why it can't just be as simple as, machine sees car, machine receives several concepts of the car and understands it.

Because concepts are not definite. All of computing, and indeed all of mathematics revolves around making procedures extremely specific. If I define an object "car Volvo = car" then there is only one such object defined exactly like this in one specific place in the computer. That's precision. It's fantastic for calculating but it's terrible for creating abstract generalization. To make computers do things we define very explicit procedures. They must be mindless procedures because computers have no minds. They are not aware of anything and cannot contain in their memory systems the representation of any "thing". They represent states that drive procedures alone. They are big calculators. They are akin to a car in that you give input (with a car, gasoline) that ensures the computer or car generates certain specified routines.

This isn't what a human does? Sees car, brain interprets object based on what is in memory.

That's not what humans do. Because humans don't have any memory storage akin to a computer. Human memory is not only distributed but is part of "processing". That's why the computer metaphors fail so completely. What allows us to classify things abstractly is (at least in part; we don't know whatever else might be vital) the fact that we encode memories constantly through multiple nonlocal connections between active patterns. There is no place in your brain corresponding to "car" and if there were you wouldn't be able to understand "car". Because when you process concepts your brain relies on actively connecting patterns across multiple distributed regions that are not unique to "car" because "car" itself isn't unique- it is connected to multiple different related concepts. The human brain is not compartmentalized like a computer. While this means it cannot attain anything like the precision of a computer, it also makes conscious awareness and conceptual processing possible.

All the behind the scenes stuff doesn't get to the root of the issues of simply seeing an object and interpreting that object "correctly".

That's because it isn't simple. Why do you classify a car as distinct from the street it is on but not from the wheels it is on?

It's like your saying the computer is wrong unless it does it like a human, meh.

Computers don't do it. period.

 

[PolyHedral]

Here's my problem with the above- It is ambiguous. For example, a computable number can have infinitely many decimal places so long as there exists a procedure for calculating in finite time any particular decimal place (i.e., if we treat some real number as an infinite sequence we can compute any nth term in the sequence through finitely many computations). Naturally, for any number with infinitely many different decimal places (i.e., not a rational number), computing the number is impossible: there are infinitely many decimals and so there is no answer. But for any nth step in the computation, there is an answer. Hence, computable.

However, a problem that requires an infinite time to compute is not answerable. A computable number, function, set, etc., must be able to halt after finitely many steps. Otherwise, it is not computable. That's one of the important reasons behind the infinite tape and the ordering of arbitrarily many machines. He showed that with any arbitrary number of Turing machines with infinite tape, the problem of determining whether a procedure to find the answer to a problem for which no known explicit procedures exist is not computable. Infinite space mattered because he required arbitrarily long tapes, but infinite time means, in general, that there is no answer. The question is whether or not a problem that requires infinite time can halt at any desired step with an answer the way an infinite number can be computed.

As much as I could understand that at all, it didn't seem to explain why what I said was ambigious. :help:

 

[LegionOnomaMoi]

As much as I could understand that at all, it didn't seem to explain why what I said was ambigious. :help:

In general, no problem that takes infinite time (or spacetime, if you wish) is computable. that's because infinite time means there is no answer. Computability is all about decidability and being able to halt on the correct answer. When you allow for infinite space and time without specifying what must be finite in order for a problem, function, set, number, etc., to be computable, then it is ambiguous because strictly speaking it isn't. Computable numbers, for example, can be infinite but the computation cannot be.

 

[PolyHedral]

What allows us to classify things abstractly is (at least in part; we don't know whatever else might be vital) the fact that we encode memories constantly through multiple nonlocal connections between active patterns. There is no place in your brain corresponding to "car" and if there were you wouldn't be able to understand "car".

That, on its own, makes no sense. What you say next about "car" being related to many other concepts does make sense, and is very likely the mechanism behind how human memory works, but that association-graph doesn't require the thing's hardware to be distributed at all. If it did, graph databases wouldn't work at all.

 

[LegionOnomaMoi]

That, on its own, makes no sense.

What part of what I said doesn't make sense to you?

but that association-graph

It's not an association graph.

If it did, graph databases wouldn't work at all.

That would only be true if graph databases did anything at all like what the brain does. They don't

 

[PolyHedral]

What part of what I said doesn't make sense to you?


It's not an association graph.

That would only be true if graph databases did anything at all like what the brain does. They don't

Graph databases do exactly that. They store the association between objects, whether those objects are categories or instances.

 

[LegionOnomaMoi]

Graph databases do exactly that. They store the association between objects, whether those objects are categories or instances.

I didn't say anything about associations between objects. Also, you continually mistake a graphical or schematic representation that is meaningful to humans, like the words in OOP or agent-based computing or (now) associations in graph databases or ontologies or any number of systems that are designed to make sense to humans as actually being what they are for humans. The fact that an NLP database like FrameNet can store information that is relevant to humans just makes it easier for humans to simulate language parsing by tracking relations that make sense to humans. It doesn't get us any closer to how humans do this nor is it at all similar to having computers actually understand anything. There is a reason nobody is making claims about how miraculous it is that OOP exists. It's because just because we can write code that seems to imitate the way we think about object association doesn't mean we can make computers actually associate between concepts. If it were that easy, we'd have made conscious machines ages ago. OOP has been around for a long time (in computer years, anyway).

 

[LegionOnomaMoi]

They store the association between objects, whether those objects are categories or instances.

What binds conceptual and perceptual entities (although the difference between the two in the brain isn't actually that distinct) isn't associations between objects. It's nonlocal nearly zero-lag synchronization between networks that binds features into a single "entity". See e.g.
Traub, R. D., Whittington, M. A., Stanford, I. M., & Jefferys, J. G. (1996). A mechanism for generation of long-range synchronous fast oscillations in the cortex. Nature, 383(6601), 621-624.

Kanter, I., Kopelowitz, E., Vardi, R., Zigzag, M., Cohen, D., & Kinzel, W. (2011). Nonlocal mechanism for synchronization of time delay networks. Journal of Statistical Physics, 145(3), 713-733.


Kanter, I., Kopelowitz, E., Vardi, R., Zigzag, M., Kinzel, W., Abeles, M., & Cohen, D. (2011). Nonlocal mechanism for cluster synchronization in neural circuits. EPL (Europhysics Letters), 93(6), 66001.

 

[idav]

That's not what humans do. Because humans don't have any memory storage akin to a computer. Human memory is not only distributed but is part of "processing". That's why the computer metaphors fail so completely. What allows us to classify things abstractly is (at least in part; we don't know whatever else might be vital) the fact that we encode memories constantly through multiple nonlocal connections between active patterns. There is no place in your brain corresponding to "car" and if there were you wouldn't be able to understand "car". Because when you process concepts your brain relies on actively connecting patterns across multiple distributed regions that are not unique to "car" because "car" itself isn't unique- it is connected to multiple different related concepts. The human brain is not compartmentalized like a computer. While this means it cannot attain anything like the precision of a computer, it also makes conscious awareness and conceptual processing possible.

I understand that but there is still a concept of a car in there that was stored because of past experiences. A car doesn't just pop out of thin air unless we managed to imagine the concept or something but I'm talking memory recall. There is a concept of a car stored in the brain, the redundancy and "nonlocal" connections don't change that fact. There are certainly different ways to tackle memory but it all stems to concepts of storage and retrieval which a computer does.

 

[PolyHedral]

It's because just because we can write code that seems to imitate the way we think about object association doesn't mean we can make computers actually associate between concepts.

So, executing instructions that have the structure and semantics of object association does not mean you're actually associating objects...

Why not?

If it were that easy, we'd have made conscious machines ages ago. OOP has been around for a long time (in computer years, anyway).

Consciousness itself isn't difficult. (Ohai, 'this' pointers!) It's all the background knowledge you need to understand what space is,

 

[LegionOnomaMoi]

I understand that but there is still a concept of a car in there that was stored because of past experiences.

We know that this is not true because of work with people with brain damage who are unable to recall episodic memories. Episodic memories are memories of past experiences. People can loose both the ability to recall past experiences AND the ability to make new memories of experiences yet they are capable of object recognition. There should be videos on youtube on people like HM and Clive Wearing which go over this.

A car doesn't just pop out of thin air unless we managed to imagine the concept or something but I'm talking memory recall.

Once the concept is formed it is no longer a memory of any experience. Also, we don't really "store" memories. This is rather vital to understanding the differences between brain and computer. Storage makes conceptual processing impossible because you cannot have the necessary binding of features into a concept.

There is a concept of a car stored in the brain, the redundancy and "nonlocal" connections don't change that fact.

There is still a concept represented in the brain, yes. Also, you keep saying redundancy. What redundancy?

There are certainly different ways to tackle memory but it all stems to concepts of storage and retrieval which a computer does.

Nothing in the human brain is about storage and retrieval. There is nothing to retrieve from and nothing to retrieve it. There is no storage. Storage entails constant states. We "store" things on a computer by changing voltage levels so that we can have binary states which underlie all storage- sequences of different binary states. In order to store things on a computer, these states must be unchanging. If they change, it is because something is being written over. Nothing in the brain is like this.

 

[PolyHedral]

Once the concept is formed it is no longer a memory of any experience. Also, we don't really "store" memories. This is rather vital to understanding the differences between brain and computer. Storage makes conceptual processing impossible because you cannot have the necessary binding of features into a concept.

So store the binding.

 

[LegionOnomaMoi]

So, executing instructions that have the structure and semantics of object association does not mean you're actually associating objects...

Why not?

Because there is no "object" and no semantics. All this is computer science 101. Computers are syntactic processors. We know how computers work. Programs are procedures. We store semantics and relations on a computer in a way that generates meaningful structures to us. You can draw these on paper. Does the paper then understand it? This is a rather fundamental attribution error.

We don't know our own code but we know that the brain doesn't associate concepts the way we picture concepts being associated. However, we do understand something of relations we have in mind regarding objects. Most of OOP structure is for the simplicity of the programmer. None of it matters to the computer. When we are interested in trying to capture features between concepts in programming we do so in ways that are meaningful to us to make it easier. The conceptual structure is in us. We can draw it on paper, label index cards, or program variable names with extensions. These are just ways to represent how we think about concept relations. To a computer, it's meaningless.

Consciousness itself isn't difficult.

Right.

 

[idav]

What binds conceptual and perceptual entities (although the difference between the two in the brain isn't actually that distinct) isn't associations between objects. It's nonlocal nearly zero-lag synchronization between networks that binds features into a single "entity". See e.g.
Traub, R. D., Whittington, M. A., Stanford, I. M., & Jefferys, J. G. (1996). A mechanism for generation of long-range synchronous fast oscillations in the cortex. Nature, 383(6601), 621-624.

Kanter, I., Kopelowitz, E., Vardi, R., Zigzag, M., Cohen, D., & Kinzel, W. (2011). Nonlocal mechanism for synchronization of time delay networks. Journal of Statistical Physics, 145(3), 713-733.


Kanter, I., Kopelowitz, E., Vardi, R., Zigzag, M., Kinzel, W., Abeles, M., & Cohen, D. (2011). Nonlocal mechanism for cluster synchronization in neural circuits. EPL (Europhysics Letters), 93(6), 66001.

Yes that makes sense which is why quantum computing and parallel processing can be helpful with that. In order to have any meaning worthwhile several processes must be going at the same time in a way that it can be still be seen as a single system. You could still have subsystems and subroutines that don't have anything to do with actual awareness part.

 

[LegionOnomaMoi]

Yes that makes sense which is why quantum computing and parallel processing can be helpful with that.

They can't. Because that's parallel processing. I'm not talking about what we normally think of as processing but of memory (in biological systems). The brain is a processor. There is not memory part to it that corresponds to memory in a computer. In order to get a computer to bind conceptual entities like the brain there would have to be no hard drive, no RAM, just processor.

You could still have subsystems and subroutines that don't have anything to do with actual awareness part.

These would have to be constantly running. You cannot have perceptual or conceptual entities bound together through synchrony by storing anything. Once you remove storage, you lose computers as they exist today or in the foreseeable future.

 

[LegionOnomaMoi]

So store the binding.

That would be ideal. The problem is that the binding is constant activity that is mathematically intractable. It's not so much that the mathematics makes simulation impossible as it is the fact that the essential feature in the binding is constant activity. The state is correlations between firing within and between neural networks. Even if we could simulate through difference equations and make it discrete, we can't store it.

 

[idav]

We know that this is not true because of work with people with brain damage who are unable to recall episodic memories. Episodic memories are memories of past experiences. People can loose both the ability to recall past experiences AND the ability to make new memories of experiences yet they are capable of object recognition.

Yes I've seen studies like this. You like missing my points completely lol. By experience I mean we need a chance to learn what the object is.

Also, we don't really "store" memories. This is rather vital to understanding the differences between brain and computer. Storage makes conceptual processing impossible because you cannot have the necessary binding of features into a concept.

An object can have multiple attributes and this can easily programmed. Storage doesn't make it impossible it makes it more concise.

There is still a concept represented in the brain, yes. Also, you keep saying redundancy. What redundancy?

So if the concept exists in the brain how is that not storage and how is it not retrieval when you try and remember the concept of a car?

I call your "non-local bindings" redundancy. I call it that because the brain has the ability to lose cells but never actually lose the concept of the car because of the memory redundancy.

 

[LegionOnomaMoi]

Yes I've seen studies like this. You like missing my points completely lol. By experience I mean we need a chance to learn what the object is.

We can't define what an object is. That's the problem. We cannot program what we cannot formalize (not that we can program everything we can formalize).

Storage doesn't make it impossible it makes it more concise.

You are assuming it can be more concise. Why?

So if the concept exists in the brain how is that not storage and how is it not retrieval when you try and remember the concept of a car?

Storage means permanent states. That's fundamental to computer memory.

I call your "non-local bindings" redundancy.

How? Why?

I call it that because the brain has the ability to lose cells but never actually lose the concept of the car because of the memory redundancy.

That's because the cells aren't what corresponds to cognitive processes. Constant firing patterns are and these do not require specific cells.

 

[nazz]

They can't. Because that's parallel processing. I'm not talking about what we normally think of as processing but of memory (in biological systems). The brain is a processor. There is not memory part to it that corresponds to memory in a computer. In order to get a computer to bind conceptual entities like the brain there would have to be no hard drive, no RAM, just processor.

This is a point I try to drive home in discussions about the brain and consciousness (usually to no avail). So this being the case how is that we have memories? Can you explain this in non-technical terms please?

 

[LegionOnomaMoi]

So this being the case how is that we have memories? Can you explain this in non-technical terms please?

Thanks in no small part to the problems of the computer metaphor, people tend to think of memories as "stored" like they are in a computer. A memory for "dog" is stored in this sequence over here, a memory for "car" over here, etc. That's not the way that the brain works. Storage relies on a steady-state. A bit that is part of the representation of some variable must be either 0 or 1. If it changes, then that is no longer the same variable. In a computer, data is represented by the permanent (so long as the representation doesn't change) state of specific bits. In the brain, there is nothing akin to bits. That's because the representation is patterns of firing.

You may have heard of the Stroop test. Usually, it involves looking at words like red, blue, green, etc., with each word being a particular color:

The task is to give the color of the words not what the word says.
Let's imagine that we simply had some place in the brain where "blue" was stored. Then all we'd have to do is call up that memory. It shouldn't matter what the word spells. Only it does. That's because our information about the color depends on neural activity which corresponds to the perception of the color, not just the concept of the color or the color's spelling. In reaction time tests, people process abstract words faster if they just saw a corresponding item first such as first seeing an upward pointing arrow and then seeing the word "hope" because part of the way we represent "hope" is upward direction. By overlapping concepts like this, we basically "overlap" memories of concepts. The concept of "hope" is represented by certain patterns that also represent "up". At a very basic level, when certain features are presented together or represented in the brain by sensorimotor activation that occurs at the same time, they get represented by correlations between the firing of lots of neurons. In a very real way, we can store an unlimited number of concepts because of the way we are able to overlap the representation of information about these concepts. Mental connections are active neural connections.

To simplify, think of the way that conditioned learning works. A bell rings when the dog is presented with food. After a while, the same physiological response that the presentation of food results in occurs with the bell. There is "overlap" between how the brain represents the bell and food.

This is kind of the opposite of redundancy, which is why I'm not clear on what idav means by this.

That's probably a terrible explanation but usually it takes me several to get sufficiently non-technical and yet informative.