Again, I don't see how your form of perception is any truer than a computer's. Past the eyes, vision is an entirely digital signal. Humans just have a more sophisticated functional relationship with their sensory input. We currently have a better capacity for making internal representations of external phenomena. Those perceptions are not real in any way. They just happen to be extremely useful in deciphering reality.

Here is a test I am curious about: Prove to me that you are perceiving something.

As for the idea of computers manipulating symbols, while humans just "perceive": Visual perception is actually very symbolic. If you look at the structures involved in the brain, there are specific bits for detecting straight lines, noses, motion etc. That information comes into (sub)consciousness in a symbolic way. For example, when you look at the edge of a table, your brain computes that there is a straight line and sends a "Straight line at position X" signal up to the next level of visual processing.

To say that humans don't manipulate symbols is actually a huge disservice to humans. Symbol manipulation is just too powerful of a system of information for the brain to not take part in. (off the top of my head) I can't think of any situation where not using symbol manipulation would somehow make a problem easier for an intelligence to solve.

Saying that computers must be different from humans because humans can't do math quickly is false. Universal Computation does not say all machines behave equally. The reason that computers can do math quickly is that math has some very specific algorithmic requirements. You can very trivially design efficient algorithms for a computer. In raw bits per second, your brain has a tremendous advantage over any computer, but there isn't the specific programming interface/circuitry required for that same efficient algorithm. Meditation is probably the best we can get as far a low level, algorithmic scale programming.

Computers routinely develop novel solutions to problems. In my spare time, I have written dozens of programs that have come up with solutions to many real world problems, such as locomotion in a 3D environment. Sure those solutions are in an abstract sense latent in the programming, but consider this. From maybe 200 lines of code, these programs can produce a nearly infinite number of unique solutions to a problem. Now, I don't doubt that a human could come up with set of solutions even more novel and numerous. But isn't that more of a difference of scale than quality?

I think there is a tendency to assume that a mysterious process must somehow be better than a conceivable process. Now, in an anecdotal sense, this is true. It's easier to conceive simple things rather than say the ingenuity of 4.5 billion years of evolution.

Well you missed the point totally. I didn't say that human brains don't do symbolic manipulation. I said that our conscious minds aren't involved in symbolic manipulation, that we are only witnesses to the results of what our brains do and that we have no apparent access to the computer like functions of our brains. When we solve math problems, we are working in ways that are nothing like computers. We're generally heaving concepts and words and images around, sometimes even outright delusions. If we had access to our brain functions, we wouldn't have to 'think' about math, it would just be done, as with a calculator.

All our conscious minds see is perception and perception simply isn't something that takes place inside a computer. Our brains may output straight lines in the visual cortex, but our concious minds may or may not notice that straight line.

Computers output straight lines to montiors or files, but they don't 'percieve' that output. We do, with our conscious minds. To suggest that computers 'perceive' their output is to suggest that telephones 'listen' to themselves or can talk.

As far as proving that I can 'perceive' something, all I can report to you is when I want to do a math problem, I have to laborously think though all the steps and then generally resort to pen and paper, if not my fingers and toes. Since I've done some programming, I know this process is nothing like what a computer program does. In the first place, none of the steps I might take in my mind are complusory, much unlike a computer program. If I add 2 plus 2 in my mind, it's not necessarily at all obvious what comes next.

Futhermore I have a sense of vision and I see things, but I can't take what I see and manipulate it in the same way that I can manipulate computer images. I can't play things back exactly, for instance. So again, I'm doing something that's quite different from anything that a computer does. So I believe I'm justified in calling what I do 'perception' and denying that it takes place inside a computer.

CareyAgimon,

Nerdnam did a good job of a explaining why brains are fundamentally different than CPU's. Asking someone to prove they are aware, or prove that a computer is not aware, is a rhetorical question. Its plainly obvious from how computers/people process information that these are different KINDS of processing systems, its not that one is more complicated than another.

I'm going to say it again - IF computers had any kind of perception, and it was based on how "advanced" they were (as you say), then a supercomputer today would be noticeably more "aware" than a pocket calculator from the 80's. Why shouldn't it be? Its exponentially more advanced! So much more processing power! Significantly more complicated software! It still all ads up to Zero awareness. Multiply zero by as much advancement as you want, its still zero.

You write: "Computers routinely develop novel solutions to problems. In my spare time, I have written dozens of programs that have come up with solutions to many real world problems, such as locomotion in a 3D environment. Sure those solutions are in an abstract sense latent in the programming, but consider this. From maybe 200 lines of code, these programs can produce a nearly infinite number of unique solutions to a problem."

First, who did you say wrote those programs? You? Not the computer? Exactly. If you did not set specific parameters for computation, the computer is not going to have any wherewithall to do jack.

"Sure those solutions are in an abstract sense latent in the programming..." Thats entirely my point. Whether or not it gives you 200 solutions or 200 million, it won't know which one is best, unless you also program it to compute that as well. Its not about quantity or quality, its about whether decisions can be made independently or not. You are the decider.

The only thing thats "mysterious" about awareness is what causes it. Its existence and functional utility are very apparent. Plans, ambitions, and (most importantly) decisions are all dependent on having self-awareness and perception of an objective outside world. And unless you are arguing that a supercomputer exhibits exhibits these qualities more than a calculator, then you can't say that its all just a question of programming complexity.

I'm seeing a lot of fancy hand-waving: "Well, we 'Perceive' things because we are 'Conscious', we are conscious because we are 'Self-Aware', and we are 'Intelligent' because we can do things 'Independently'." Those words don't mean anything, they are just clap-trap from bored philosophers. I dare you to first define them and then prove that computers can't do them.

Nerdum, you should reread my previous post. Your argument is like saying "A PC is a computer and PC's have WinAmp, Macs don't have WinAmp, therefore, they aren't computers. I mentioned specific technical reason why we can't just send "array.Sort()" to our brains. Those technical differences are completely explainable in the context of the theory of Universal Computation. You have only proved that brains are different forms of computers, not that they aren't computers at all.

Schmidty, your analogy of calculators to super computers is false. Unless something is being specifically designed for AI, it's not going to move towards it. Most super-computers are running things like nuclear physics simulations, applications with little need for AI. The computational power of a super-computer compared to a human is still just a drop in the bucket. Difference in scale isn't the same as difference in inherent quality.

If your DNA didn't build your brain, you couldn't be intelligent either. You base your decisions off of reason rather than ignorance. You can trivially describe almost all of your intelligent daily decision making in words and rules. At what point does your brain become this fabled "independent" and "therefore" intelligent entity? How does determinism decrease intelligence?

What you are talking about is humans being able to respond to a very wide variety of situations. I am saying that computers can respond in the same intelligent way, just to a (currently) much smaller variety of situations.

There isn't a metaphysical property of awareness. Everything you are conscious of is a simplified version of the real thing. Therefore, all awareness is ultimately symbolic. Symbolic interaction is the only form of intelligence. Whether the rules for symbolic interaction were written by evolution, person experience, personal reflection on internal symbols, or a person writing code, the rules are just as effective. Why would you use anything but effectiveness as a measure of intelligence?

Consciousness doesn't mean anything? Ok Careyagimon, if you are going to base your argument on the assertion that you yourself are not conscious, then I guess there's nothing I can say to that.

"How does determinism decrease intelligence?"

The answer to this should be obvious - by allowing independent decisionmaking. A machine can not and will not make a decision that it is not programmed to. You can program as much randomization into it as you want, it still isn't going to be making a decision based on its own will or ambitions. Nor will it be aware that there is a decision to be made.

You point about supercomputers not being programmed with AI programs is well taken. But there, nonetheless, powerful computers running AI programs somewhere. The crux of my argument is that they posses no greater intelligence than a simple machine, they are merely programmed to exhibit it. If you are saying that, no, this means that they ARE more intelligent, then I would say to you; in the presence of a nail, is a hammer suddenly "more intelligent" than a screwdriver?

No, because real intelligence is not the ability to present the right solution to a situation. That is just the PRODUCT of intelligence. Intelligence is the ability to understand a situation fully and evaluate it. You say that AI can do that - it pulls in the info, processes it, and gives a response. I'm saying that that is not enough, if you want to get to the point where an AI can have a will of its own and make independent decisions.

In other words, I don't see how anything can progress beyond stimulus-response without self awareness. If you think that all humans are are advanced stimulus-response machines, I've done my best to explain why I disagree with that notion. If you think that Computers will eventually gain a self-awareness similar to ours, I say that it is possible, but I haven't seen any evidence that AI is moving towards any level of it.

Proof of self-awarness IS NOT a rhetorical question. It is the key differentiator between 'strong intelligence" and "weak intelligence". And like anything else it comes in degrees.

Human levels of self-awareness are present in only a handful of animals (dolphins, some primates, elephants), but partial levels are present in other animals like dogs (can recognize self from concrete symbols like urine/musk but not from abstract symbols like images).

The test for computer intelligence is known as the Turing Test But the problem you are running into is that you keep defining "intelligence" as some sort of monolithic concept. Yet even for humans it is not seen that way, instead it is seen as an aggregation of various cognitive functions.

First, who did you say wrote those programs? You? Not the computer? Exactly. If you did not set specific parameters for computation, the computer is not going to have any wherewithall to do jack

No more or less than the human brain. Much of what is known as "child development" is in fact "programming the human brain". So a human brain is no more capable of "doing jack" without training than a neural net computer is capable of it. But BOTH are capable of computation once they have been programmed.

You are putting way too much differentiation into the programmatic aspects of computers vs. brains. Both are computational devices. Currently the human brain is a couple of decimal orders of magnitude more powerful, but it is still just a computation device.

It was a rhetorical question the way it was used, as in, "c'mon, you can't even prove to me that you're really aware"

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. That is not a trivial difference. That is the defining difference between programming and intelligence.

"Much of what is known as "child development" is in fact "programming the human brain""

In analagous terms, thats accurate, but its not an accurate descriptor of what is happening. First of all, no technician is programming a child's brain with a limited set of routines or algorythms. The "programming" is the sum response of the brain (which is already, itself, "programmed" to learn and grow) to the CONSCIOUS EXPERIENCE of the world. Right off the bat, thats 2 momentous differences between humans and computers.

1. The brain comes pre-set to learn practically anything. Through simple excercise of brain function, different regions become more adept and healthy. Not only do we not need to "install" anything in us to learn, we practically can't stop learning at any given moment. This is because we are consistently referencing memories to the present, and drawing lots and lots of minor conclusions about the immediate future from them. Computers do not learn in a fluid manner like this, and the reason is because they are not aware.

2. The "programming" in humans takes place in real-time, in RESPONSE to events that are PERCEIVED, and is more or less indistinguishable from simply living and existing. Computer programming, or AI "learning" is nothing like this.

Here is a turing test I want to see - can an AI experience a pavlovian response? Give an AI robot ears and program it to open it's mouth whenever food is present, and ring a bell when the food is on its way. Will this robot ever, ever open its mouth out of anticipation of the food arriving? Will it ever be able to create a correlation between the ringing bell and the incoming food (without that possibility being programmed in ahead of time, of course)? I will be very, very impressed if I ever hear of such a thing, because that is an example of experiential learning. Its also the basis for real intelligence.

In comp-sci there is a theorem/problem known as the Turing Machine Halting Theorem. This has been shown by J. Hartmanis as equivilent to Goedellian Incompleteness for which Hartmanis was awarded the Turing Award (full disclosure Hartmanis was one of my favorite profs).

Say HUH? For those who aren't Comp Sci folks a Turing Machine is an abstracted model of any and all Von Neuman (stored procedure, computational machine) computers. it consists of an infinite tape on which you can both store its programming as well as any input or output data, a stored program or CPU and a temporary store called a "stack". Mathematically it can model any and all possible computers that are not based in Quantum probability.

The "Halting Theorem" is deceptively easy to understand but rather hard to prove mathematically but also an increadibly powerful application of Godelian Incompleteness. Consider the following problem. On this tape is a starting point From that starting point onwards a random pattern of 1,and 0s has been pre-recorded. The "stored program" knows how to recognize '1s' and every time it sees a 1, it stores on its stack an incremented counter. When that counter reaches 5 it lights a green light, if it reaches 6 it lights a red light and Halts. And if at any point in time it recognizes that there are exactly five '1s' it halts.

So what sort of questions can we answer with this computer?

• Are there more than five '1s' on this tape? If there are, then YES we can answer that because as soon as we hit 6 the machine halts.
• Are there Less than five 1's on this tape? No we cannot answer that because the tape is infinte
• Have we found five '1s' on this tape? Yes we can answer this, it is NO if the green light is not lit and YES if the green light IS lit
• Are there EXACTLY five 1's on this tape? No we cannot answer that EITHER because the tape is infinite.

What this means is that there are some things that are "true" (exactly five '1s' on the tape) that the computer cannot determine as true.

So where does this apply to self awareness? OK we reprogram the Turing machine to recognize the design of a Turing machine (DTM). It can validate that DTM has a finite stored program set in that machine, and it can verify that there is a stack, and it can read DTM's tape.

But CAN IT say "YES this is a Turing Machine" - ie halt when it has recognized the DTM as a Turing Machine? the answer is NO.

Why? Because the DTM has an infinite tape associated with it, and the Recognizer will never reach the end of the tape. If it reaches an end, then the DTM is NOT a Turning Machine, but if it is a Turing machine the end will never be reached and thus the Recognizer will not halt.

What this says is that a Turing Machine - aka a Stored Program infinite input and storage computational device CANNOT recognize itself without intervention by a "more powerful device". IOW it cannot be truly "self aware".

And since the human brain is simply a computational device with essentially infinite input and storage capacity (reading and writing), we TOO CANNOT actually self-recognize.

Let me say this again, WE CANNOT ACTUALLY SELF-RECOGNIZE. IE Humans really ARE NOT "self aware". What the human brain is designed to do, unlike the Turing Machine, is that we have complex heuristics built into our brains. Some of these are neuro-biological, some of these are learned, but what we do is GUESS the outcome based on probability.

Decisions are made by neuronal choruses (see How Brains Think ) and the "chorus" that is "the loudest" in a particular circumstance is the one that wins. Escherian drawings take advantage of this and depending on how you focus your attention, different "choruses" become dominant in the recognition processing of the visual input.

The same applies to our "self-awareness". We are only "kind of" self-aware. we are "self aware" to the extent that we BELIEVE ourselves to be "aware". But as someone else pointed out, we are not fully aware of how our neurons are working. Sure we can abstract them into models, reason about those models etc. but we cannot really be fully "self-aware" in "real time".

So then the question of "sentience" becomes rather one of

"How Self-aware is ENOUGH" rather than "

is it self-aware. A dog is self-aware to a limited extent. Less so than a chimpanzee or dolphin, but more so than a fish. Furthermore the level of "self-awareness" is predicated on multiple types of cognitive capability. Can we differentiate between
ourselves and the rest of the world
ourselves and other animals
ourselves and other humans
our current self and our past self
our immediate actions and how we would like to have acted.

Same applies to computers. the level of self-awareness and sophistication of that awareness is highly variable.

And in fact there ARE attempts to write programs that "self-recongize" ie "self validate" for correctness. They are subject to the "Halting Lemma" but by coding in heuristics (ie probablistic decision making aka "fuzzy logic" ) we can get around that the same way Human Brains do.

Just as a child's brain is not "programmed by a technician" neither is a "neural network". Both are programmed experientially. Both learn in the "fluid manner" that you describe. And no, it is not true that human brains are programmed in "real time". In fact repeated feedback, some of which can be done by visualization (ie re-running the same inputs) is necessary for complex programming tasks.

Here is a turing test I want to see - can an AI experience a pavlovian response? Give an AI robot ears and program it to open it's mouth whenever food is present, and ring a bell when the food is on its way. Will this robot ever, ever open its mouth out of anticipation of the food arriving?

Yes. Neural Networks do this all the time. In fact Neural Nets are used in EXACTLY this way by the finance industry

And mathematically, Neural Networks have been shown to be no more powerful than Turing Machines. And every non-probabilistic computer (ie non-Quantum computer) is a Turing Machine.

You are simply wrong on the science of self-awareness and learning here.

Depending on the definition of Sentience, Global Hawk already has an element of it. It has senses, upon which it relies to make decisions about keeping itself from harm (crashing) and it reacts to a host of inputs to do so in real time. It similarly detects and reacts to attempts to kill it by deploying countermeasures (jamming, chaffing) and evading

Now that's the sentience level of about a cockroach, but with an even more limited conception of what is and is not a threat. But it IS an evolution of sentience.

I did not mean it as a rhetorical question. What I was getting at was that you haven't put forth any description or meaning of consciousness. When I said that consciousness is meaningless, I mean that it is meaningless as an argument. Obviously it is of huge importance to AI etc. But it keeps getting thrown out like "Well, computers can't be intelligent, because of uh... consciousness." No matter how many times you simply repeat the word, that's not an argument or proof, it's just tired old philosophical back-patting. Capitalizing it doesn't do anything either. It's the same thing for "perceived" and "independent." Put forward a good functional description, and you can start to say things like "Computers can do this, but not that."

"1. The brain comes pre-set to learn practically anything." How is that different from the example I gave earlier? Several lines of code being able to produce a variety of responses to a diverse set of experiences. "First of all, no technician is programming a child's brain with a limited set of routines or algorythms." In my example, no technician is explicitly programming each of the responses either.

I think you are confusing two different meanings of the word independent. No computer or person has any sort of metaphysical independence. The meaning that you are looking for is the ability of an intelligence to respond to a wide variety of situations in a more positive than negative fashion, without an intervening outside intelligence. Independence is only important in the sense of not requiring something else for continued effective operation.

A good indicator intelligence, although by no means a requirement, is when a seemingly small "program"(or brain) can react successfully to a much larger variety of situations than are explicitly programmed for. For example, a human is intelligent because one unique human can learn to climb mountains, play a trombone, recognize symbolism in a film etc. If a computer program can do all of that without further/explicit programming, why wouldn't you consider that to be intelligent?

If we take a human, poke out his eyes, chop off his ears, nose, and tongue, and burn off 99.9% of his skin. Now, the only path for information into the resulting hulk is morse code tapped onto the tiny patch of remaining touch receptors at the center of his forhead, how perceptive is that unit? How do we know how perceptive it is? Computers can percieve just as humans do if they had the appropriate sensing apporatus. A binary in and out feedback system doesn't really help to establish human perception.

Wow, talk about a fantastic lack of empathy.

Imput that brain into a robot frame and you really are going to find a killer robot on your ass.

Degsme, I understand what you are saying in that human consciousness is really a series of predictions, and we are not literally "self-aware." But we're aware enough, as you point out. You seem to indicate that there exist, today, certain machines that have a level of self-awareness, however small. I disagree, and having looked at your evidence, still am of that opinion. What you have shown is that there are very sophisticated programs and computational designs which mimic human thought and can produce, in some ways, similar results. This is not the same as intelligent thought.

Intelligent thought is not just an objective appraisal of data, even if you do incorporate learning ability. It requires a subjective stance which can recognize WHY one is computing a certain task. That piece of information, the "why," is extraordinarily useful in helping organisms make choices and plan ahead.

It seems like our own experiences here have led us to see the facts differently. I'm not sure I can add anything further to my point other than to say that your evidence, while very impressive, seems to be lacking any "road to subjectivity" in computers. If I'm wrong, I'm wrong, and skynet wins. Time will tell, and this is a subject that I remain skeptical about, but interested.

Again, computers are not at all incapable of "planning ahead" in some motivated (why) cause. Consider the quintessence of "planning ahead" - Chess. Already a computer has beaten the best "planning ahead" a human mind can generate.

I realize I'm probably not going to convince you, nor am I saying that the level of "intelligence" in computers todate is ANYHWERE near that of humans, largely because the computational capabilities aren't there yet. But they are there sufficient to drive a cockroach and act like a cockroach. And according to that estimate the world's supercomputers are within 6-10 years (24 = 32) of being able to simulate a human brain. Which means that when MRI has the resolution to scan a human brain - though not in real time, - an attempt WILL be made to simulate it.

Note kurzweil's predictions aren't doing that badly ten years down the road

I think you are over-emphasizing the perception of subjectivity. It isn't ONE process. It is a series of processes, some of which computers are fairly advanced on, and others where they are not. But what is necessary for its emergence at Human or at least primate levels, is the aggregated advance of these components

• Generalized Learning is already starting to happen with computers. You see this in "self-healing" systems, and bayesian infererence engines (which are included in mainstream software like Microsoft's "Sharepoint Server"
• Reasoning. As demonstrated before, Deductive reasoning has already demonstrated quite a bit of success. Relevance filtering is still an issue, but work on this is crucial to visual navigation systems, so this work will move forwards as well
• perception - Again, as Global Hawk and the Financial Forecasting models demonstrate, perception and integration of that perception into a comprehensive world model that is used for decision making demonstrates, work is moving forwards very quickly here.
• Language Understanding - what this really means is AMBIGUOUS language understanding, aka the "world scope" problem. and this too is making fast progress and has lots of research by companies like Google, IBM and Microsoft.
• Problem Solving and this is the last and weakest area. Currently only specialized problem solvers (knot topology for example) exist. But again, work is going forwards on this.

When these all get put together, your "subjective stance" is largely met. And that is just a matter of time and hardware.