Hans Moravec again

[norman]

I have infact already made a topic on the subject, some time ago. I find the subject interesting, however, and it would be interesting to hear people's about it. Quoting Hans Moravec, a futurist/transhumanist, from his article Dualism Through Reductionism:

You are in an operating room. A robot brain surgeon is in attendance. By your side is a potentially human equivalent computer, dormant for lack of a program to run. Your skull, but not your brain, is anesthetized. You are fully conscious. The surgeon opens your brain case and peers inside. Its attention is directed at a small clump of about 100 neurons somewhere near the surface. It determines the three dimensional structure and chemical makeup of that clump nondestructively with high resolution 3D NMR holography, phased array radio encephalography, and ultrasonic radar. It writes a program that models the behavior of the clump, and starts it running on a small portion of the computer next to you. Fine connections are run from the edges of the neuron assembly to the computer, providing the simulation with the same inputs as the neurons. You and the surgeon check the accuracy of the simulation. After you are satisfied, tiny relays are inserted between the edges of the clump and the rest of the brain. Initially these leave the brain unchanged, but on command they can connect the simulation in place of the clump. A button which activates the relays when pressed is placed in your hand. You press it, release it and press it again. There should be no difference. As soon as you are satisfied, the simulation connection is established firmly, and the now unconnected clump of neurons is removed. The process is repeated over and over for adjoining clumps, until the entire brain has been dealt with. Occasionally several clump simulations are combined into a single equivalent but more efficient program. Though you have not lost consciousness, or even your train of thought, your mind (some would say soul) has been removed from the brain and transferred to a machine. In a final step your old body is disconnected. The computer is installed in a shiny new one, in the style, color and material of your choice. You are no longer a cyborg halfbreed, your metamorphosis is complete.

Body identity assumes that a person is defined by the stuff of which a human body is made. Only by maintaining continuity of body stuff can we preserve an individual person. Pattern identity, on the other hand, defines the essence of a person, say myself, as the pattern and the process going on in my head and body, not the machinery supporting that process. If the process is preserved, I am preserved. The rest is mere jelly.

Consider the message "I am not jelly". As I type it , it goes from my brain, into the keyboard of my computer, through myriads of electronic circuits and over great amounts of wire, and after countless adventures shows up in bunches of books like the one you're holding. How many messages were there? I claim it is most useful to think there is only one, despite its massive replication. If I repeat it here: "I am not jelly", there is still only one message. Only if I change it in a significant manner: "I am not peanut butter" do we have a second message. And the message is not destroyed until the last written version is lost, and until it fades sufficiently in everybody's memory to be unreconstructable. The message is the information conveyed, not the particular encoding. The "pattern and process" that I claim is the real me has the same properties as the message above.

Keeping in mind that I do not share his opinion, nor am I a reductionist, what flaw do you see in his reasoning - if you see any?
Now, since the "pattern", or ego-identity, is a phenomenon, whether or not it exists is beside the point. Why? Because its only existence is as conceived, i.e., it doesn't exist from its own side (as we all know). It's a concept. The point is, since the argument itself is based on samsaric principles, we need not be concerned with its actual "truthfulness". We only need to be concerned with whether it holds up as an argument, in the same way that we are concerned about a book of fiction or a movie, and how they hold up logically. In this sense Truth is beside the point.

I decided to not bump the old thread, for the simple reason that it had already died a natural cause, and the interest had reached zero.

[ronnewmexico]

Ah....is this spiney Norman?
I find no overall relevence to the statement. That aside...


No individual may be preserved as individual, as it seems no individual is ever found. Process going on..not machinery of support. Well it seems that the machinery of support cannnot be seperated from that which it supports. If we can find such a clear distinction then seperate objects may be found in such a place.....then how do we know of such objects? Exchange only perceives object,so no individual nor seperate individual may ever be found. Show me how we may perceive without exchange and I will agree to what is proposed, the conception of a seperate individual.

If such distinction not be present the support being essential part and parcel of that which is described as pattern identity....then on what may we make such claim of distinction or differentiation?

Is support and pattern then differing? If such is differing present exactly the differing characteristics which make such subject and object of different nature.....I fail to see it. So then both identified of similiar nature where exactly is this distinction being made?
Is then one of causual nature and another not.... One but serving the other. Exactly how then does this transpire? One the server and the other the being served.

Seems not. The wall I see the water I put my hand into...it is not other I am perceiving it but myself it seems in contrast to what is considered other.
So for such to be so such must be shown....this world where things perceived are actually not exchanged with.
Then perhaps we can entertain singular idenity and pattern identity in this and not that.

The patterns I seem to find have no such distinction.

The essential flaw as I see it is in the summary conclusion what is thought is seperated from the context of how what is thought presents.
The thought but one more product of cause, causes another effect from which another cause arises. So no thought ever presents in singularity regardless of how profound or how base. This flaw to my opinion is a characteristic of a mundane way of approximation of things and we cannot discount this as being relevent as it is the crux of the assumptions being stated.

Thoughts have no seperate identity...such is certainly implied as I read it. The machines and all that...circumstances of presentation and causual representation.....essentially having no real consequence to this thing of thought. It changes not thoughts character as empty. Though we tend to worship thought as this seems to be tending. It can't be discounted for what it is or stamped out.... but to make it above what it is.... holy inherantly existant or some such, to my opinion is a error. Causual it certainly is as well as is everything.
.

[catmoon]

The only flaw I see is the assumption that the process would work. It might be that although one notices little difference at first, after proceding say halfway through one might notice problems with one's mind developing. Of course by that time half your brain would be in the garbage can.

[kirtu]

Keeping in mind that I do not share his opinion, nor am I a reductionist, what flaw do you see in his reasoning - if you see any?

Dr. Moravec is a fine professor but - his article that you quoted is pure speculation. In fact it is very difficult to train artificial neural networks to do useful work - for small problems it is quite easy now, for anything other than a small scale classifier it is quite difficult - and then to coordinate complex activity between multiple such classifiers other than in a non-emergent pattern has yet to be demonstrated. It is much easier to simulate intelligence using rule based systems and there too we haven't made sufficient progress (for one thing people are afraid to tackle real problems with large rule based systems). Progress in artificial intelligence, the very basis of Dr. Moravec's, has been quite uneven and many people wish to pronounce AI's numerous winter's or death's. Nonetheless, AI will be radically succesful eventually (this is the basis for Bill Joy's pesimistic warning "Why the Future Doesn't Need Us". Nontheless, mental patterns by themselves are not a mind in Buddhism.

So anyway, Dr. Moravec has yet to demonstrate this level on complexity. We can demonstrate that systems including learning systems can solve relatively simple problems but nothing close to the level of human or higher animal problem solving (although constrained search alone can be used to solve problems that humans cannot solve).

Kirt

[catmoon]

AI? Well let's see... we are trying to build it, but we don't really know what intelligence is, we don't know if it has parts or not, we don't know what it is made of, and we're pretty sure that in the human case, it runs on very sloppy noisy hardware while our current machines run almost noise-free. (Consider what a one part per million noise signal would do to the data on your hard drive.)

I don't think our current efforts are even in the right ballpark. Gigantic networks of billions of logic gates have a hard time matching the data processing capabilities of a cricket, which can orient itself and move in the real world without bashing into things, avoid most dangers, find food, and even reproduce itself.

[kirtu]

AI? Well let's see... we are trying to build it, but we don't really know what intelligence is, we don't know if it has parts or not, we don't know what it is made of,

Intelligence is glossed as reasonable behavior situated in a particular environment and circumstance. So it is a behavioral observation. Minksy generally says that an intelligent application/machine is something that does something that if a human did it, we would say that the human needed some intelligence or special knowledge to accomplish that task. The same definition can be applied to all other animals.

and we're pretty sure that in the human case, it runs on very sloppy noisy hardware while our current machines run almost noise-free. (Consider what a one part per million noise signal would do to the data on your hard drive.)

In the context of approximate reasoning we want what people might regard as noise. Animal information processing capability is exteremely complex.

We have algorithms that can actually minic and even predict human behavior in a situated context. For people like me who are big rule advocates - there is a group at Carnegie-Mellon that runs studies, compiles data on people's reactions and composes rules to replicate their behavior (there are lots of such groups - this is the core activity of cognitive psychology). So rules systems have been created to model various learning problems and these results are now being applied to tutoring systems with some success (although not enough success - the kids at the hs I taught at did not follow the program as they should have - they have multiple problems but often think that the deck is aggressively stacked against them and so they put forth minimal effort). This particular system is based on John Anderson's Act-R theory.

I have written an expert system (a rule based system encapsulating expert knowledge) that analyses computer log files at a much higher level that most systems administrators are capable of. Particular applications are capable of outperforming almost all humans on earth but of course only in a particualr context.

I don't think our current efforts are even in the right ballpark. Gigantic networks of billions of logic gates have a hard time matching the data processing capabilities of a cricket, which can orient itself and move in the real world without bashing into things, avoid most dangers, find food, and even reproduce itself.

We have a long way to go but could get there quickly. Both Moravec and his critics are agreed that the primary factor that can predict future AI success is hardware innovation, in particualr Moore's Law. However that is not the whole story. We can really make intelligent applications although we usually choose not to but the intelligence is quite limited. The coming century will be mostly a slow increase in the overall intelligence of our systems + useful robotic systems. The downside is that Bill Joy and Cameron might be right after all and we trigger a form of "Terminator" - please see Joy's essay "Why the Future Doesn't Need Us" - I tried to get my hs students to read and analyse this but they were not up to the task and we had too many fires to put out daily for me to lead them through it. But then again I was a bad teacher. However it is essential reading for adults and has sparked an ongoing debate between Joy and Kurzweil.

Kirt

[catmoon]

I was trying to say that the noisy environment of the brain and it's fault tolerant "algorithms" are the way to go. Hardware running this way consumes far less power than current approaches. I don't think Moore's Law will ever get us to AI - I think that component density and cooling problems will eventually stall semiconductor systems before they reach real intelligence. On rereading I can see it looked like I was saying the opposite!

All the same, the progress you mentioned is impressive. In a digital sort of way.