Hi,

Replication
This is not simulation, emulation, mimicry, modelling (analog or digital). IBM did not replicate. Those who are most imbued with the phlogistonised thinking are in the computer science field. They are completely unable to see the difference because they are immersed in what they do.

"To he that only has a hammer, all the worlds problems look like nails" (Maslow)

Semiconductors?
No. The chips are not based on semiconductors. There are parts of it that can use standard digital electronics, but this is not the important part that does the signaling: = Leaky capacitors.

Rich,
To complete the answer to your question, I will start by replicating an ant-sized brain. That, as AGI, would be a major breakthrough. It's smarter than any existing AGI (handling novelty).

The key to understanding what the chips do is that the signalling in a brain is LITERALLY happening in the chips, but without all the biological overheads. The chips will have and EEG and MEG signature like the brain does.

cheers
colin

once you've explained how a leaky semiconductor does signalling, it should be straightforward to simulate it on a computer.

I did not say that a "leaky semiconductor" did signalling. I said that semiconductors have nothing to do with the signalling physics.

I said that leaky CAPACITORS do the signalling.

Yes, you can simulate all of it on a computer.

I am saying that the simulation is NOT the science of cognition.

I am saying you build, like artifical fire, artificial cognition... then you explore it and you find out about the science of cognition.

can you see this?

Exploration of the physics of cognition is not the computation of assumed physics. You cannot claim to have accessed all the properties of the natural original by replacing it with the physics of a computer.

Colin,

Thanks for the clarification. I don't think I will understand the signalling in the near future, but for now I can say it is very interesting and exciting research you are doing.

Also, I know we're all dissing your analogies, but your CERN one hangs together nicely because it compares physical exploration to simulation directly.

Thanks for posting. If I run across any investor types that are looking into AGI (sadly unlikely, but you never know) I'll be sure to mention your project. If you have a website for your work you should mention it here!

If you are looking to present forward looking omnibus talk on this, you might mention this to the folks who run the singularity summit - if nothing else it would be a fun way to meet people who are sure to be very interested in the topic.

Colin, what is the procedure for replicating an ant's brain?

We need its connectome for that, I think. Once we have the brain, we also need the body, so that it can be embodied and act like a normal ant. It will need mechanical actuators to move, photo-sensitive elements to see, transducers for its antennae, chemical sensors. I don't think we can miniaturise all that, so we're talking about an ant the size of a dog at least. Then we can start studying the computational maps in its brain. That's the plan, if I read it right?

Back to step 1. Can we get an ant's connectome and how do we know we got it in enough detail? Especially, how can we tell the strengths of its synapses from scans?

Hi Evgeni,

You have intuited through to the nitty gritty! If I could hire you to help I would!

Yes ANT Version 0.0 will be the size of a dog, with all the attendant input/output and power needs. Ugly ant-dog indeed! Current robotics has the bodies and the instrumentation is available. I am not too concerned about this end of it.

As to connectomes: Once the basic chip topology is created, we don't 'program it'. We learn about learning adaption, and hard-wire the dynamics of learning adaptation, not knowledge.

i.e. It sets itself up.

Yes it'll have to be motivated somehow. We'll have to start with basic reflex activities and build up to goal-directed behaviours. Experimentation with 'immature (hyper-learning)' and mature 'phenotypes (hypo-learning)' will be required. Whole new areas of expertise need to be established.

The final architecture may end up looking like an ant brain.

Or not.

Several hundred thousand to a million 'processing elements' should do it.

I won't know how to do any of this this unless I start.

And I won't be able to start unless somebody else's penny drops (actually lots of pennies!) about the power of replication in this endeavour, and the notable lack of it in this endeavour.

So I'll just keep plugging away!

Fascinating stuff, eh?

Cheers

Colin

Nothing! That's the point!

Despite the report pointing out problems with AI approaches using computers....absolutely _nobody_ has done actual replication.

In a wet neuroscience sense, things have recently changed. We finally have empirical confirmation of ephaptic coupling. Now we have solid science on what to replicate.

Anastassiou, C. A., Perin, R., Markram, H. & Koch, C. 2011 Ephaptic coupling of cortical neurons Nature Neuroscience, 14 217-223

Frohlich, F. & McCormick, D. A. 2010 Endogenous Electric Fields May Guide Neocortical Network Activity NEURON, 67 129-143

cheers
Colin

Creating an artificial general intelligence is measured by its capacity to perform functions.

NO.

It is measured by its capacity to behave like humans (biology). This includes an ability to handle novelty.

One perfect benchmark is "the scientist". Scientists are modellers of the unknown. A computational model of a modeller of the unknown is an oxymoron because you have to provide a formal model of the unknown.

So the answer to all this is that YES, you can simulate a scientist, but to do that you'd have to know everything and simulate the (unknown) environment as well. So it's quite possible, but completely useless from the perspective of incomplete knowledge of the natural world.

"The Scientist" is my benchmark AGI.

Confont your computed AGI with something that nobody (including the AGI designers) ever saw before .... viola.... underperformance.

"It is measured by its capacity to behave like humans (biology)"
Well it is obvious you are of the opinion that we must replicate the biological functions, but incorporating it into the definition of AGI just begs the question, if that is your intention. I'm not sure why "no" was capitalised, was it to emphasise that I am wrong? If, within your definition, AGI is measured by its capacity to behave like humans, then is that not the array of functions it is tasked to perform?

Is it your assumption that a model can't incorporate an ever evolving model of the external world into its own model? Are you suggesting that by definition models can't learn? Are you suggesting that the only possible way in which such a creation can learn is through a strict replication of the biological functions?

Hi Ben,

All your comments are valid. But, as I just posted above (modified slightly), none of it justifies the total lack of replication approaches in the face of 60+ years of AGI failure.

***************
What I am saying is that I do not know for sure. You do not know. _Nobody_ knows because the basic exploration has not been completed because the necessary replication (as practised everywhere else in science) has not been carried out yet.Until we do the exploration properly we cannot model it. We cannot know what cannot be modelled and what can.

60+ years ago, when the magic of computers was born, we started simulating models and somewhere along the way we forgot that the usual way to explore the natural world is to replicate phenomena to get the right model. Then we know the simulation works.

We started simulating too early, kept going generationally, and now we habituate simulation. We need to return and complete the exploration.

60+ years of AGI failure, an obvious and unique anomaly in our exploratory behaviour, and a raft of oddities established on a discipline-wide basis that reinforces the belief system that keeps it all going despite no scientific confirmaiton of the premises inolved.

What have we missed?

Well it must be something _really important_ because we keep on simulating and we keep on failing.

How many more decades of failure will it take to make people realise that maybe, just maybe, the glaring lack of the replication approach might be a good bet? What manner of strangness is involved that we keep going, year after year after year on the basis of an uproved hypothesis?

It doesn't take Sherlock Holmes to see that maybe non-zero $ aimed at replication might actually be a massively leveraged route to some level of explanation for the failure.

I am not saying abandon AI projects. I am saying that replication could benefit the entire arena by having some level of investment, rather than zero.
*********
BTW
If anyone wants to see the status of the Church-Turing Thesis as a presupposed (unproved) hypothesis Look here at the very first paragraphs.

http://www.worldscinet.com/ijmc/04/0401/free-access/S179384301240001X.pdf
*********
The kind of chip architecture I plan is more possible (indeed easier) than ever before. Once I have the basic device sorted out, I should be able to do a whole ant or bee brain in the first few scaled up versions. The numbers are simply not that great. One of the useful aspects of the lateness is that chip fabrication techniques are developed beyond what is needed.

This could have been done in the 1970s!

The most massive brain simulator ever is planned. $Billions. I predict that simulator will be a very useful tool to explore models of human cognition, but that it is not an example of cognition. It will underperform in the mysteriious way that all these projects have underperformed: in the face of novelty the dynamics of learning will be fragile and poor.

I estimate $200,000 spent on actual replication could give us the missing AGI ingredient, and prove the $billions wasted.

If I was assesing risk, I'd spend the $200,000 to at least make sure I was fully informed.

The people at CERN have spent their billions replicating things in the supercollider to understand what is going on because they admit their models are frayed around the edges, and computing can't get at them.

What craziness is it that justifies a case for that the most complex thing in the universe can be understood without ever having properly tried replication?

cheers

Colin

Knocking others? Maybe I should take it as a positive sign that my logic is touching some nerves. If that gets people thinking then great, I have done my job as a communicator.

The point of the article is that when you try to get $ you run across the "received view" and encounter difficulty ....so much difficulty that in the entire history of AGI _nobody_ has done replication, and the reasons for it do not stand up to scrutiny and deserve a poke in the eye.

I have a viable design for the basic device. I have some of the parts in my office. I have worked out how to test it all and have an appropriate prototyped 3D field measurement system at my disposal. I have a faraday cage at my disposal. I have a PhD in the physics of it.

I would be delighted to prove it or disprove it!

All I need is $$$$.

BUT ....

between me and the $$ is a whole lot of broken logic and dogma. I am doing what I can to make it visible for what it is.

So don't give me the 'put up or...' line. I have been trying _for years_ to do exactly that!

Colin, thanks for the article. It was very interesting. I agree model simulation is (generally) incapable of replicating that which it does not first understand or assume (except by accident) - although sometimes one can obtain astounding surpriseful results based on simple rules when modelling dynamically and detailed complex systems (e.g. bird flocking)

At the end of the day it appears your main gripe is that all funding is directed toward a single approach and that the "failure" to demonstrate any real progress in AGI (vs AI) warrants consideration of funding directed toward an alternative approach? - i.e. yours - e.g. the ant brain (& body - to avoid the disembodied mind problem) artificial replication. If that's the case I agree.

Allow me to suggest a strategy if I may be so bold.

Some of your language in this discussion (and on consciousness) might appear to some to be a little way out and impractical, not to mention challenging (and to some "crazy"?). Is it possible the funding powers that be have a lot of trouble understanding? Do you think that pointing out their lack of understanding will persuade them?

These powers may well have "a whole lot of broken logic and dogma". But if you have been trying for years - perhaps the way you have been trying needs to change? Given that your approach meets a perceived "mindset" blockage then you either need to appeal to a different funding source (not one steeped in computational AI) or wrap your appeal up in different language that does not "challenge" the received view but appeals to it or helps it understand that you are not "odd" or a "threat"

Are you familiar with the story of the 3 situations where we have a block of wood that we let go? In scenario 1 when we let go of the block of wood it falls. In scenario 2 when we let go it remains where we have left it (doesn't fall). In scenario 3 when we let go it flies upward. The only way to explain these apparently anomolous results is to realise that in scenario 1 we are standing in a normal environment on the ground. In scenario 2 we are in space or in free fall (zero g). In scenario 3 we are under water. Normally the other environmental cues avilable to us explain these different scenarious and are readily apparent - but without these cues the results of scenario 3 are inexplicable to someone wrapt in scenario 1 and unable to perceive the possibility of scenario 2 or 3

Currently (based on limited insight from this article) your "problem" is a little like the issue of these scenarios. I apologise this analogy is clumsy and imperfect - but imaging trying to explain to someone in scenario 1 (normal g, air) the odd behaviour of the block of wood in scenarios 2 & 3 if that "someone" was unable to conceive of being underwater or in space. They just "won't get it".

No matter how much you try to "make it visible" you will be unlikely to succeed.

So, how to progress? Are you able to break up your ant brain/mind/body replication project into discrete "chunks" - each of which have a small but useful "goal" explicable within the paradigm of existing funding powers but not obviously outside it or challenging it but which, together, start you toward the overall goal? A bit sneaky maybe but worth a try?

Good luck and keep trying :)

Thanks Mark.

I am aware of my curmudgeon tendencies, and I have tried a plethora of approaches. You shoudl have seen the first version of the article!

My formal funding approach involves a very modest goal: a smart sensor achieved in a very structured manner, with mere pointers to the various downstream potentialities. AGI is not even mentioned, nor is consciousness or any of the stuff. It doesn;t need to be. I fact the work looks like standard brain electromagnetism explored by replication. It will appear like that in the literature because that's what it is. I have learned a lot about the sensibilities of the science funding system over the years.

This is a 10-15 year (from now) process. I will probably not be the one that builts the first robot. I have no illusions about how difficult this is.

But right now, the underlying issue for science (lack of replication) remains and it has to be addressed by someone, somewhere. If the process broaches ideas that look a bit wacky to some, then so be it!

cheers
colin

I did not make those claims about what will be in 2019. I don't know anything about the technology of 2019. If you read what I have actually written it is simply a summary of an article where IBM researchers make claims about 2019.

I'm not sure what you are getting at with your questions. My own position is that we don't know either way whether human brains and AGI can be simulated on computers, or not. I am concerned that proponents of "brains are computers" and proponents of "no they're not" are equally dogmatic and hostile towards each other for no good reason as neither side has an explanatory theory of consciousness with which to settle the matter.

It is not possible to predict what will or will not be in 2019. That would be prophesy.

I have two comments in reply:

1) The term _thinking_ can refer to linguistic-type thinking, with symbols and production rules. Even visual imagination.
Early AI attacked those type of problems. But these types of thinking are probably a corollary of the _other_ activity that the brain performs, and that's creating the signs and symbols out of the input data in the first place. That activity may or may not be best described as thinking. This is the forming of words and phrases out of sound, and interpreting a visual scene as objects and their properties. I think this is the first and most important set of problems blocking our way to building machines that we can relate to more easily, and program more intuitively.

2) We don't have the right hardware. The core of the issue is that the brain can simultaneously attempt to match maybe a few thousand symbols it knows against the input, and also do this hierarchically. This kind of thing is cheap in the brain, namely trying to do everything you know to the input simultaneously. It's not cheap in conventional computers. Current parallel computing paradigms are in their infancy in the sense that as an ordinary software developer I would have to bend over backwards in order to recruit the graphics card or FPGA to help me with a calculation that I know can be parallelised. Either way, there are always these constraints from the available architecture and consequently languages, and they are very distracting for the signal-to-symbols problem research because they are not the right tools for the job. So the solution is probably to work on designing _hardware_ to solve the problem directly, and emulating it in software to start with. Actually, FPGAs might be flexible enough, I'll need to read more about them. Do they have memory distributed across the area of the array? I mean, can I make a million little units in an FPGA, each having its own private memory?

So I agree that thinking in terms of hardware is moving in the right direction. But I don't see why we should skip the simulation step. A wind turbine simulation of an airplane, for example, isn't the same as the real thing, but it's sure a cheaper way to optimise its aerodynamics than building lots of versions of the real thing.

I'd like to point out that the required hardware ( that does actual action potentials coupled spatially with the actual EM fields produced by the action potentials) does not exist. It's not an FPGA or any other kind of semiconductor model of anything. No memory of the kind we know. No bus suystem of the kind we know. No register system of the kind we know. It's not analog computing. It's not manipulatinf symbols of the kind found in a computer program.

In replication there are no models. There is the physics of cognition. That's it.

Also, there seems to be a misunderstanding about the role of replication in science. Replication helps you understand the physics. It is PRIOR to simulation. We do not know the full physics of cogbnition because we have been simulating assumed models to the complete exclusion of replication.

You do not do simulation to discover physics. You replicate, discover the physics, then simulate. What has been missed by the complete absence of replication in this area? We do not know because we have not done it yet!

also......

When we replicate flight it's obvious what is going on. Stuff flies around!

When we replicate brain tissue it is had to see what is going on. But it is, and replication of brain tissue is no less like the tissue than artificial flight is flight.

This requires a change in mindset.

Like I said somewhere else here: Physicists know they can't simulate their way to the physics: they replicate it to get the physics and then simulate to explore possible models.

Remember: The article is merely drawing attention to an operational anomaly in this particular specialised science. I do not intend to get dragged into arguments about what is or isn't consciousnes, thinking or perception etc.etc.... this has nothing to do with the main thrust of the article.

Also, I am not saying that computer-based AGI should be abandoned or that simulation is irrelevant!

I am saying that 100% confinement to computers for AGI is anomalous in science, and that simulation cannot uncover actual physics, and that we do not know the full scope of the physics because we've forgotten to use one of the most powerful science tools for 60+ years!

cheers
colin

Let's explore the Wright brothers analogy some more. I'm not familiar with the historic details, but suppose they have conducted experiments with objects in a wind tunnel *in order to figure out the physics* of aerodynamic properties of simple objects. Then they had to take these simple objects, wings, and put them onto a cabin, so that flight could be combined with carrying people.

Is _replication_, in your usage, the activity of isolating parts of the system in a way that allows them to be characterised? In neuroscience: voltage clamp, patch clamp experiments can be used for this kind of isolation. You can study a complex system by keeping everything fixed and varying one parameter at a time.

A lot is known about action potentials, but getting from there to intelligence is not as easy as putting a few wings on a cabin. Maybe a good way of putting it is to say that an action potential is not a simple unit of intelligence, whereas a wing *is* a simple unit of flight.

So we need to find out what a simple unit of intelligence is... Or cognition, that may be better terminology. That is hard, because if you take any brain area in isolation, it's not intelligent, or it doesn't do cognition. Another approach to try is to pare things away until we have a simple unit of intelligence/cognition. Locally anaesthetise parts of a brain until it stops exhibiting intelligent behaviour? We will find that a particular subset of areas is needed for cognition, none of which could be taken away. What then? We haven't really made things simpler that way. Go to simpler and simpler animals, but then we get to C. elegans, and conclude that it's just a simple machine, and it doesn't have any cognition of interest. So we went too far towards simplicity. Let's backtrack a bit. Ants? They are a lot more complicated, but their cognition is still fairly modest. It's somewhere between zebrafish and mice that the mysteries of visual cognition begin. And scientists can replicate them rather cheaply, and have been doing so for a long time. So I think they are on the right track or not far off.

1) I agree, except to the word "blocking". I think that most people in the field have understood the need for understanding sub-symbolic and distributed processing to capture cognition. Symbolic processing does exist (for instance in planning and grammatical language), but only as a special case.

2) I do not agree, hardware is not the problem. The majority opinion on the computational complexity yielded by biological neurons and their translation into neural abstractions is quite well supported by deep learning neural networks, such as the cat concept forming network demonstrated by Google/Cornell last week. Current super computers already operate roughly within the computational magnitude of biological brains. Yes, cheaper and faster would be nicer, but our main problem is not speed or neural abstraction, but architecture.

If Colin wants to make an argument about hardware that convinces computer scientists or funding agencies, he needs to stop using misguided metaphors and non-sequitur arguments. His idea that hardware beats computation is fundamentally flawed.

Instead, he needs to tackle the mathematics of his approach (i.e. formalize the topological effects of his spatially aligned action potentials), and demonstrate (= prove mathematically or experimentally) either:
a) His approach enables the (practical!) utilization of emergent computational effects that exceeds ordinary parallel computation by magnitudes,
b) His approach makes an important class of algorithms (applicable for learning, categorization, retrieval...) efficiently computable, which would not be efficiently computable by ordinary parallel computers.

Evgeni,

'Laws of nature' (LoN, statements of regularity) have multiple origins. A physics model is a LoN. In settling on a LoN we have:

1) Observation of the natural phenomenon as it is. This can support a particular model.

2) There is replication of the natural phenomenon. This can lead to novel observations (1), augment thereby augment a LoN, making it more general.

3) There is computer simulation of models. This can be suggestive of new (1) or changes to a LoN.

To find geniuinely novel phenomena, or if a model has holes in it that cannot obviously be plugged by simulation, then replication can be used to explore. The Wright Bros found artificial shapes that are not birds and yet flew. Genuine novelty.

The CERN supercollider is an example of (2). The physicists know they cannot simulate their way out of an epistemic cul-de-sac. So they replicate.

Replication is just a tool.

The point is, that in the case of neuroscience (2) is 100% untried!

And since the confirmation of EM coupling in tissue, the lack of replication is even more pronounced. Extremely complex coupling and esoteric wave mechanics could be right under our noses and we'll never know because we keep thinking we already have the needed models. We do not.

I keep saying this, and I hope that it sinks in.

I am not claiming that replicaiton is everything! I am saying that it is _missing_ from neuroscience, and that maybe there's a panopoly of undiscovered country that will highly inform the (1) and (3) activities in neuroscience (like EEG and patch clamp experiments).

I also claim that this situation is anomalous in the history of science. Only in the last 50 years could we have ever had this situation happen, because only in the last 50 years have we had ubiquitous computing, making (3) seductive, causing entire generations of folk that have no real grounding in the history of empirical science method. Therein lies systemic, trained-in confusion, for the first time in history, of the map and the territory.

Once again I defer to item one in my original posted comment above (the top): I find myself having to justify my position, when historically the reverse should be the case.

For 10 years I have been doing this, and it's getting old (me too!).

Can I just get on with replication instead?

cheers
Colin

Joscha,

"If Colin wants to make an argument about hardware that convinces computer scientists or funding agencies, he needs to stop using misguided metaphors and non-sequitur arguments. His idea that hardware beats computation is fundamentally flawed. "

1) My metaphors are merely a depiction of 2000 years of science. Please go read about it. It is the way it is.

2) "His idea that hardware beats computation is fundamentally flawed". I have not said this at all. Please read the dialog and the original article. I say that replication is a tried method that has been completely unexplored in brain tissue. How much has been missed? Nobody knows! That's the point.

Also: See the most recent comment to Evgeni.

Once again I find myself deferring the item 1 of my very first comment above. I should not have to be justifying this approach. How much evidence of this 'phlogistonised thinking' do I have to provide?

Nice thinking :)

You could also consider the way the brain interpret a flavor from geometrical formations aka molecules and up, The better they 'fit' their receptors the stronger the taste.

So where does the flavor we taste becomes created? In the molecules? Or is it a layer of 'theoretical definitions' our brains present us with, the 'conscious' me, interpreting the signals it get from the receptors??

So you have the receptors answering to a geometry identifying it. You have the 'brain' interpreting the same into a flavor. Then yourself, consciously 'tasting' that flavor.

Plasticity and some other way of synergy that I don't find the word for. And then the question. What of this do you think best represent 'intelligence and consciousness'?

The last 'layer' perhaps?
And that 'fog/cloud', if so, is something more that the sum of its constituents, as i see it.

I think in the next 100 years science might figure out ways to visualise very well what lights up when smells and tastes occur. But I fear that nothing interesting could be said about the "feel" of subjective experience, although a great deal has been said (cf. the field of philosophy).

It seems that justify you must. Historically, people's minds have been controlled by certain memes and not others for no better reason than they were the loudest or the clearest or stuck by some other mechanism. What seems logical to one person is neglected by another.

Replication. Do you think any cognition at all is important? Personally I'm mostly interested in the vision and motor capabilities of mammals and birds.
Is there anything to learn by replicating C. elegans to start off with? Does it have cognition that we don't understand? Does it do anything that cannot be simulated by our model of it?

In the article, you point out: "If there was no difference between a computed model of flight and flight, the computer should fly." -- This is a flawed metaphor, in my view. See, if our discourse would be the result of a computed model of thinking instead of thinking, it would still be the very same discourse.

The general idea that computation can be likened to phlogiston is wrong on multiple levels (but I like the narrative approach it engenders). You suggest that physics (i.e. hardware) can do more and fundamentally different things, which is wrong for epistemological reasons. All you can ever know and experience is (by definition) information, and all regularity in information can be accounted for by computation. Hence the question is not: can computation account for intelligence, but: can this particular hardware enable the necessary computations in an efficient manner?

This is the argument you should make (that your suggested implementation would be vastly more efficient), instead of loosing yourself in misunderstandings about flight and phlogiston.

If you say that nobody had tried to replicate brain tissue (functionally), many researchers will disagree (see, for instance, the Blue Brain project etc.). Thus, your claim should be more qualified: "I believe that my EM field approach captures an important effect that has so far been ignored". That is an interesting direction for a hypothesis.

Instead, you make two additional, overreaching claims:
1. Without the EM field approach, neural network models cannot work.
2. With the EM field approach, we can nail the Strong AI problem.

1) is ridiclous, because learning and self-organization have been demonstrated in ANNs.
2) is unconvincing, because most animals are not generally intelligent, and yet would implement your approach in their brains.

I do not wish to detract you from your interesting work, and I do not believe that I will be able to detract you from the unfortunate arguments that you needlessly invested yourself in. However, I suggest that you write up your arguments and submit them to high-level journals, to get qualified feedback.

Maybe, but the consciousness is a subjective experience as I see it. And as such it's what I call 'theoretical', although more real to us than what creates it :)

It seems as if most of the things we call real, just are interpretations we do.

Joscha,

It is a brute fact that NOBODY has replicated and explored brain tissue physics. If so, please deliver evidence of it. I fear you don’t know what replication actually is. You mistake the blue brain for replication. It’s not replication.

An information metaphor ( = map) is not the actual physics of tissue (the territory). Look at a brain. Where’s the ‘information’? No such thing. There’s atoms and cells etc.

And so forth.

Look. Leave to the literature trail. When my replication results come out, then we can have a decent discussion based on solid experimentation.

Colin, it is a brute fact that nobody has replicated wing tissue physics, and hence, no artificial system can fly. Look at a bird. Where are 'aerodynamic principles'? No such thing. There's atoms and cells etc.

Do you see how your argument falls on its face?

Please, let me summarize the discussion:

AI researcher:
1. Cognition is information processing.
2. Neurons facilitate information processing. We can abstract from that, and functionally replicate what neurons do.
3. We have demonstrated the viability by creating autonomous learning systems that way; we have replicated stuff like motor control, associative learning, concept formation etc.
4. Our systems are far from complete, though, because brains are not just lots of neurons any more than computers are just lots of transistors: the architecture matters, and is only sketchily understood.
5. Humans differ from animals not in the way their neurons work, but in the overall architecture.
5. Once we get the architecture right, we can model human cognition.

Colin:
1. Cognition is something else than information processing. Nobody knows what, so we must build the particular hardware I suggest.
2. Neurons facilitate the exchange of action potentials within a hypothetical complex EM field interaction. Your functional replications show that you do not know what replication is: you must recreate exactly that hypothetical EM field interaction in hardware.
3. I do not care that your approach works.
4. Your systems are not incomplete, they are wrong. Once we build my hardware, you will see that the EM field interaction was missing.
5. I do not care that animals with the same kinds of biological neurons are not thinking like humans. (Let us worry about that until my system is built?)
6. Once we build my hardware, we can model human cognition.

Does this sum it up?

:) I insist on "blocking". In the sense that people in the field are well aware of the problem, but ... eh ... there is still work to be done there :)

That cat concept forming network sounds interesting, but I can't find it on the net. Is it in some video talk?

No.

No matter how may times I do this and how concisely I put it, you keep getting it wrong.

6. Once we build my hardware, we can model human cognition.

I do not want to _MODEL_ anything. I want to replicate it.

Sorry I can't do this any more. Life is too short.

In a corporate environment you will often find people who say "the company would want A and B", or "A is in the interest of the company, while B is not". This surprised me at first because it seemed like personifying the company. Was is a euphemism for the president or the chairperson? No. Then how do we find out what the company wants? This is a set of memes that are shared by the management (more or less). So this way a virtual thing like the company becomes conscious: it's made up of a bunch of conscious entities (each of which is replaceable) all of whom share the view that they *are* a unified whole: the company. Personal consciousness probably works just like that: each part of your brain is independent, but it thinks in terms of you, the union of all of them.

The above paragraph is an example of the usual thing you hear about subjective conscious experience. One thing to note is that it's quite useless discourse. If only we could come up with a test to see if a system had subjective conscious experience! Now that would be interesting. In the meantime, the most practical thing about subjective conscious experience is probably to try and expand your own, for example by attaching an input device like a bat's sonar and see for yourself what that would feel like. Except where to plug it in? Into the thalamus?