No, computers and Brains are not "just" computational devices. One is a computational device, and the other is a computational device that also generates self-awareness.

If the brain is a computational device, is it not theoretically possible to reproduce the same computations with non-organic matter? If the brain is an extremely complicated way of firing electronic impluses, why should it not be possible to create a machine that does the same? I know we're nowhere near that stage yet, but I don't see any reason why it couldn't be done. And if one is a materialist when it comes to "thinking" and "awareness," then I don't see why such a machine would not be intelligent and aware.

And, for that matter, to be able to have perceptions such as redness. Though I have to say that I don't think perception is relevant to intelligence or awareness. There are plenty of people who can't see red, or can't see at all, and are perfectly intelligent and self-aware.

Douglas Hofstatder and Daniel Dennett's book The Mind's I <link> is a good collection of essays on consciousness and the AI debate.

The examples that were given of current "robot" technology are mostly remotely-controlled vehicles, not robots. That being said, there are systems in planning/development which exhibit autonomous behavior, such as the RQ-4 Global Hawk, a long-range surveillance UAV. However, the tasks that are being automated are still physical tasks and not higher-order reasoning.

When considering this "threat" I am reminded of what my grandfather used to say when passing a graveyard: "It's the live ones you need to worry about."

Computers already surpass us in many many tasks.

Eventually they may surpass us in nearly all of them.

But the brain is very very good at problems in human scales and terms (which is no surprise as that is what it evolved to deal with), and human minds should have inherent advantages on these scales for the foreseeable future.

As for the self/awareness sentience problem. I have a feeling that is a pseudo-problem, and that once we built a system with the internal complexity (and specific complexity not general) and power of a human brain, and feed it diverse and patterned enough inputs it will just sort of happen.

That is still at least decades off, and some of these problems and curves are getting harder not easier. (Think a bit of the people in the 50s and 60s predicting we would all be living in space and driving flying cars by now). The low hanging fruit is getting picked over, and the rate of functional progress will diminish.

I do think taking a look at the cog-sci literature, particularly things like "What is it like to be a bat?" would be of help to some of you. It is still very young, but there are bright people working on these questions.

Degsme, I understand that you think I'm over-emphasizing the single issue of Subjectivity/Awareness. I also see how, from your perspective, what this says about the ability of machines to think like us. But we're going to have to agree to disagree.

You write about Generalized learning, reasoning, perception, language, and problem solving as components of subjective consciousness. But in my view, these are not just components, but also qualities that emerge from consciousness. You seem to be saying that if you get a sufficiently advanced version of these programs working together in the same box, then you get awareness (or something close enough to that). I disagree.

Consciousness, as a 6th quality, creates feedback for those original 5 qualities and makes them work in a real-world fashion that machines have not yet been able to even inch towards. Take language. The human brain isn't just decoding symbols here - its giving them context within the understanding that another entity - one just like you - is saying them. That concept - that there is you, and someone else, and that you can coordinate or have the same or different goals, etc... - that is immense when it comes to true language usage.

You need self-identification to understand language fully, just as much as you need language to facilitate self identification. Its a chicken-or-the-egg kind of problem, and it most likely evolved in tandem in humans. But I don't see machines as being on the path to replicate this evolution, because as the skills and processing power that computers use to interact with the real world have evolved, there has been no parallel evolution of self-awareness, as there was in organic life.

You seem to argue that, yes there has. Well, what I see are advanced programs that create similar outcomes as self-awareness does, not actual self awareness. And I do think that there is a substantive difference.

The difference between exhibiting qualities of self-awareness and actual self awareness is whether the entity in question can leverage that awareness as its own central piece of information when evaluating other data. I see no indication that a machine has done so. I see no relation between Neural Networks producing financial predictions and pavlovian response. Pavlov's dogs didn't salivate because their brain forged a direct connection between audio stimulus and the need for salivation. The audio stimulus was processed, then took a detour via the animal's simply concept of itself and its relation to the objective world, and created the thought of food. The salivation is a response to their own thoughts, which were triggered by the bell, not to the audio waves of the bell itself. A "zombie dog" (or computer) placed in the same situation would not react that way, even if they did technically make the correlation between the sound and the food. The correlation would simply be filed away with no applicable use.

The fact that we have active and passive imagination that can conceptualize, play with, manipulate, and otherwise invent different scenarios inside of our own heads is the central point of human thought. Our head runs its own predictive experiments not based on pieces of data, but based on discrete thoughts. Or, for lack of a better term, platonic ideas. Working with ideas and concepts, wholesale, like we do is what makes intelligence real. It provides the ultimate processing shortcut. And doing so is more a product of self-awareness than it is an ingredient of consciousness.

These ideas and images and thoughts are not simply by-products of the stimulus-response process. They are additional stimuli and responses that we create for ourselves to give context to our situation. Its an experience. Nothing I have seen indicates that machines can do this, or have ever had the briefest, most fleeting brush with experience.

Same thought I had in a short I published years ago. We will destroy ourselves..

<link>

If anyone is interested.

larry

How are nanites different from viral plagues?

Nanites aren't different from viral plagues.

Lots of forms of life are non-conscious, and operate merely according to preset programming and stimulus-response. Single-cell organisms would fall into this category. Viruses may not even be alive.

You write about Generalized learning, reasoning, perception, language, and problem solving as components of subjective consciousness. But in my view, these are not just components, but also qualities that emerge from consciousness. You seem to be saying that if you get a sufficiently advanced version of these programs working together in the same box, then you get awareness (or something close enough to that). I disagree

You are not just disagreeing with me, but with most modern cognitive scientists. The question of how consciousness emerged has been a long researched and debated one and as yet unconcluded. While it is possible that a radical new conceptual theory could yet come forth, the current line of research indicates that instead of Learning, reasoning, perception and language emerging from consciousness, instead, consciousness emerges out of the aggregation of progressively more sophisticated versions of these component aspects. Essentially how a landslide and the physics/mechanics of landslides emerge out of the combined properties of billions of grains of sand.

IE it is CONSCIOUSNESS that is the emergent property, not the trappings of consciousness.

. I see no relation between Neural Networks producing financial predictions and pavlovian response. Pavlov's dogs didn't salivate because their brain forged a direct connection between audio stimulus and the need for salivation. The audio stimulus was processed, then took a detour via the animal's simply concept of itself and its relation to the objective world, and created the thought of food.

Neuro physiologically and hence cognitively, the two are indistinguishable. You are making distinctions without a difference.

The fact that we have active and passive imagination that can conceptualize, play with, manipulate, and otherwise invent different scenarios inside of our own heads is the central point of human thought.

Again you are conflating the emergent with the environmentally driven. Scenario exploration (ala DeepBlue) is something that nature would select for. Abstract conceptualization would be an emergent property that is a side effect of scenario exploration - something the idle brain does on its own when there isn't a pressing decision driving it.

But more importantly, you are asking a computational device capable of cockroach functionality to act like a human brain, and then claiming there is no evolutionary path forwards for it because it doesn't have the horsepower to do so. That's just not a valid line of analysis.

The very fact that we ARE capable of fully mimicing a cockroach, as well as some of the cognitive decision making processes (financial neural networks) of higher order thinking, suggests more than strongly that its just a matter of computational horsepower, not anything inherent in human cognition.

These ideas and images and thoughts are not simply by-products of the stimulus-response process. They are additional stimuli and responses that we create for ourselves to give context to our situation.

Arguably the two are indistinguishable except through the validation of others. Edelman proposed this in his book Bright Air Brilliant Fire almost 2 decades ago when he hypothesized about "neuronal sheets" being how we "conceive" and even "remember" things. More recent work by the likes of Calvin validates and extends this concept into "neuronal choruses" that can be and sometimes are recursive.

Again, nothing you describe is particularly unique to the human brain except the processing horsepower.

Oops I forgot to add the thought that came to mind as I was writing the above.

Namely that I think you are holding computer congition to too high a standard. I suspect this comes from the perception that "computers think much faster than humans" - which is patently NOT the case. Computers do very concrete operations very quickly. But when you want to layer on more complex and abstract processes and build up these processes from the very low level concrete operations, you end up with a computer that is much much slower than the human brain.

Consider the simple task of adding two numbers. There is CURRENTLY no conceptual abstraction for the computer of those two numbers. The HUMAN takes the role of mapping the conceptual abstraction of a "number" (remember that at one point, humans could not conceive of the number "zero") to a very precise and concrete form. This is then fed to the computer which in turn is able to process this disambiguated data very quickly.

But the human brian has to do this conceptual disambiguation EVERY TIME it recognizes a number much less operates on it. As a result it is slower to do the actual addition from input to output, but its actually doing a lot more processing than the silicon machine is

The hypothesis btw about savants, is that their brains somehow get wired to bypass this abstraction, hence they can do the numbers very fast, but have no conceptualized recognition of their meaning.

"Consider the simple task of adding two numbers. There is CURRENTLY no conceptual abstraction for the computer of those two numbers. The HUMAN takes the role of mapping the conceptual abstraction of a "number" (remember that at one point, humans could not conceive of the number "zero") to a very precise and concrete form. This is then fed to the computer which in turn is able to process this disambiguated data very quickly."

Now you're talking my language.

So I take it you are saying that, eventually, there will be a "conceptual abstraction for the computer of those two numbers"? And that processing power is the only missing ingredient which will lead to this understanding of discreet concepts in computers?

That would be the basis for consciousness, in my view. But from what you have been telling me sounds like it has all been on the "disambiguated data" side of thinking. All these prorgrams and experiments you are linking to, while extraordinarily impressive, do not seem to be breaching the gap between these two types of thinking. Certainly they are attempting to.

I'll admit that I myself may not have the abstract mind enough to understand how one thing will lead to another here, but it still seems to me like adding more processing power will simply lead to faster disambiguated processing. I see that we have a number of research initiatives going on to try and get to that destination, but I have yet to see any results that indicate we know the process by which a mind - man or machine - can understand such discreet things. The answer is out there, and when we figure it out we may or may not be able to program it into a machine.

Ultimately, the problem is seeing the difference between something that appears to us to have these abilities, and something that actually does. At this stage in the game, I think it would be foolish to say that we know, on some level, some of these AI programs have a teensy bit of experiential consciousness.

I'm not one to say I'm definitely right. But I will set the threshold for what I consider to be definitive proof, which is precisely that. When it is proven that some AI program, somewhere, can relate and compute in terms of abstract, discreet concepts, like "one" and "two." Until then, I see no way to differentiate between mimicry and actual re-creation of intelligence. You seem quite confident that we will hit that point eventually, and you may be right. You certainly know more about the science of this than I do. But it seems that we are discussing this at a time when the evidence necessary to reach my preferred level of proof has not yet been hit.

We already have some level of conceptual abstraction in computers. So called 3GLs and 4GLs provide some of that abstraction. Even something like Visual Basic does. Today's Visual Basic, were it to be run on a TRS80 or Apple II would add very very slowly because not only is the number "2" polymorphic (Integer, text, float) but the addition operator "+" is "overloaded". As a result the computer at run time has to determine what form of "2" it is dealing with and decide which addition algorithm to apply before handing off the actual addition to the hardware.

Faster disambiguated processing IS THE KEY. We currently shave off certain types of abstractions (Natural lanugage interfaces) because they are too compute intensive for our current hardware. But as hardware power doubles every 18 mos, we can - and are - adding those capabilities in. Consider the U/I you are using. 20 years ago NO COMPUTER was capable of that level of graphics at that rate of speed. SOME specialized graphics could do this, but not at this speed.

Ultimately, the problem is seeing the difference between something that appears to us to have these abilities, and something that actually does.

This is what the Turing Test is all about. You are defining "experiential consciousness" in an "all or nothing" manner rather than a composited manner. And it isn't very meaningful because

When it is proven that some AI program, somewhere, can relate and compute in terms of abstract, discreet concepts, like "one" and "two."

That is EXACTLY what AI systems like the ones used in Chemistry have been doing for years. This itself is not that difficult a task.

I think we're talking past each other at this point.

The examples you are giving here do not sound in any way like "comprehension" or "awareness" of the concept of integers. Instead, they sound like additional lines of code which allow the program to compute

At least we agree that disambiguation processing is the key element. But I'm not sold on the idea that we've made headway into awareness and actual intelligence just by increasing the speed at which these things are processed.

Are you really arguing that a computer that can utilize integers are "more aware" than those that cannot?

Because what I am seeing here is increasing proof of a disconnect between processing ability and intelligence. For all the advancement that has been made in getting computers to process in terms of things that we consider to be discreet objects, computers don't seem one iota more capable of comprehending them than they did 20 years ago.

Computers may be speaking an internal language that uses integers, but that offers nothing in the way of comprehending integers.

Today's Visual Basic, were it to be run on a TRS80 or Apple II would add very very slowly because not only is the number "2" polymorphic (Integer, text, float) but the addition operator "+" is "overloaded". As a result the computer at run time has to determine what form of "2" it is dealing with and decide which addition algorithm to apply before handing off the actual addition to the hardware.

This doesn't sound like comprehension of integers to me, or even the basis for comprehension of integers. It sounds like additional programming which adds additional data to representations of integers. If anything, it sounds like it is adding further, unrelated data to the situation for the sake of displaying more accurate screens to the human user. Even if you had an AI program that could deal with these bundles of data in compartmentalized forms, that has nothing to do with understanding that "one" means singularity, or "two" means duality, etc. All this additional programming info doesn't seem to bridge the gap between content and context.

So again, it sounds like mimickry, not intelligence.

I don't intend to describe intelligence as monolithic, but I do see it as having a basic requirement; that of subjective awareness, and its accordant perception of an objective outside world. The degree to which that perception is present in a mind can vary from minimal to human-level, so I'm all for a diversity of intelligence types. But if you aren't thinking, you aren't thinking.

So as for the Turing test, well, spending time on that leads to the conclusion that one can never really know with 100% certainty whether anything is actually self-aware.

But hm... my own standard that I've set is that I want to see a machine that can work in concepts, like integers, or identities. I can't think of a turing-style test that would work for that off the top of my head. Thats a ponderable for me. I'll get back to you if I think of one.

Thinking may be computing, but computing is not thinking. There is a difference, and it is in the understanding of discreet concepts as things-in-and-of-themselves. You and I know what "1" means, and how it is different from other integers. No matter what version of visual basic you're running, there is no understanding of the concept of singularity, or any other concept, going on there. Just because it has been programmed to react to compute integers as unique pieces of data does not mean there is any further comprehension of what an integer is.

For all the advancement that has been made in getting computers to process in terms of things that we consider to be discreet objects, computers don't seem one iota more capable of comprehending them than they did 20 years ago.

Absolutely not the case. Huge advances in how objects are defined, recognized etc.

This doesn't sound like comprehension of integers to me, or even the basis for comprehension of integers.

Except it is, Numbers not Integers. You were taught to add integers as a kid, but not floating point or imaginary numbers. As you advanced your mathematical cognition, the level of polymorphism on what is addition increaseed. So too with computers.

I don't intend to describe intelligence as monolithic, but I do see it as having a basic requirement; that of subjective awareness,

Well tha's a definitional arguement. One that most cognitive scientists do not agree with. As I pointed out with Dogs vs. dolphins, subjective awareness is itself incremental.

So as for the Turing test, well, spending time on that leads to the conclusion that one can never really know with 100% certainty whether anything is actually self-aware.

Wrong Turing concept. Turing test involves communicating with 2 entities via a computer. One is a computer, one is human. when you cannot distinguish which is which "The singularity" is upon us. Today it is possible to write a program that would convince the majority of people that the singularity is upon us.

Thinking may be computing, but computing is not thinking. There is a difference, and it is in the understanding of discreet concepts as things-in-and-of-themselves.

No. Actually thinking is simply decision making. Not all computing is decision making but most of it is.

And... definitional disagreements bring us to this end.

At this point, we would have to parse out what consciousness is, how it differs from computing, and what does or does not constitute a "decision." From what I gather, we simply have different definitions of these heady concepts.

I imagine if we were to both examine a highly advanced AI, and were both up to speed on its capabilities, we would still disagree as to how "truly intelligent" it was.

I've been interested in going up against an advanced turing test program for some time now. If you know of any way I might, please do tell.

Fair enough. Thanks for the interesting conversation.