Neil Nelson wrote:

> With my firewall-router the port number on the connection identifies

> which computer in the local network is sent the packet. Unless the

> access method is somewhat direct via an IP and port number, the

> communication will not go through, which of course is the purpose

> of the firewall.  Perhaps later code can be placed at the firewall to

> retransmit ICI packets meeting security requirements, but that is much

> farther off.



As you know (Neil), during the debuging of my Console, I had to target

your firewall for the information to arrive to the computer inside it. I

am unaware of what is the trick here, but I thought since you took it

over, you would like to know.



> >"Supercomputer" would probably not be a good name for

> >what *we* are building, since the main positive point of our system is

> >that it's working with cooperating individuals, not one big collective.

> >Our system is a parallel processor + you know, each machine is good for

> >something else, one big collective does not cover all the abbilities as

> >well as a parallel processor can.

>  

> Perhaps a different name would be better:  `Internet Collective Computer',

> `Internet Cluster Computer', `Internet Parallel Computer'.  We need some

> descriptive name that would also make sense for promotion and marketing



If we're talking about something sellable, then I think "Internet

Collective Computer" and "Internet Parallel Computer" are dramatic

enough, tho only "Internet Parallel Computer" matches the actual thing.



> The reason I mention this particular application much like CYC is that it

> would be a very fundamental ability useful in a large number of other

> areas.  Being able to commnicate with a computer in English, a computer

> that can draw upon a comprehensive knowledge base is a kind of Holy

> Grail of AI (Turing Test).  It would provide a very general interface

> method that could be connected to many other knowledge areas.  That is

> e.g., technical works such as various areas of science and mathematics

> utilize English as the set-up, primary, or most outer description language

> such that if you know English, you can start learning any of these other

> disciplines.  It would be a kind of universal language gateway.



With the fact that ICI functions can be spread over multiple computers

in place, I could say that we do not need to be building one huge

database anywhere. 



The information is surely located all over the world in huge specialized

databases, with efficient search engines in place, all we have to do, to

make this data CYC-ic would be to make the previously mentioned search

engines ICI compatible, then we can call them from a function that will

do the job of merging I/O of multiple such databases.



This simply removes the need for us to copy and reformat any existing

information.



Anyway, this is all completely possible to make via the ICI network.

With AI as a part of our original goal, I think there is little point in

not trying to make it.



When you mention English, I start wondering what's wrong with the

languages the computer knows so far, but ok, let's say this is to be the

core component of an AI. Keep in mind the typical design of ICI: If

you're missing a component that you can't make, simply put a blank

function in place and continue devolopment.



> >And a phone-based search engine, eh? =] This might take a lot of work if

> >we dig into it. =]

> 

> It would take a bit of work, but the primary work is pattern matching or

> similar to statistical learning that would be ideal for a large number of

> connectected computers.  There should be some basic learning algorithms

> that could be applied to both language parsing and voice.  The idea would

> be to identify and code these basic algorithms and then let them do most

> of the remainder via the large computation resource.  Some specific

> algorithm assistance could be provided to streamline the learning in the

> different areas, but it seems to me that a good portion can be left to the

> computation resource.



I spent some time analizing sounds and voice using a spectrograph and I

can tell it won't be exactly as simple. The tricks of the voice are 1.

Multple 1 frequency sounds with frequencies equaly distanced, 2. Noises

and 3. Short zones of silence.



It would go fine to be simple repeating pattern matching, but the

problem is that #1 changes from voice to voice (see attached pics),

while #2 and #3 remain the same.



> The specific problem is in the case where we want to put a new program

> on another computer and have that computer run the new program.  This

> would not be as important for fixed, distributive applications that may

> be most of the typical customer's interest but would be somewhat

> common in a cluster application where you were submitting portions of

> arbitrary programs to the different computers for parallel computation.

> This would be the typical for a supercomputer kind of arrangement.

> 

> Actually getting that specific program receive and run problem solved

> should take a very simple program that could be part of your ICI or

> ICSA.  It may be better placed in the ICI since I see the ICI as a kind

> of resource manager, while the ICSA is merely a communication

> tool.



Indeed. With an appropriate "Run" or "Add to function server" function

with the Console, the problem is immediately solved.



> >Since the procedure is to be so common, you might think of intigrating

> >the procedure in the very ICSA itself. The job would be exactly the

> >same, just with the Internet transfer aspect left out; a file or file's

> >meaningfull content would be transfered from one path to another

> >according to the given name of the target computer.

> >

> >It just happened to pop to mind, that my Console could realy utilize a

> >command to ICSA to spread the transmition over all computers in the

> >cluster.

> 

> Harold was also looking for a broadcast ability.  I am thinking that a

> computer in the system may want to broadcast to a certain subset of

> the computers in a kind of local ICI group or an ICI group for a

> particular purpose, but to just broadcast to every ICI connected

> computer, particularly when there may be a very large number, seems

> a bit unusual.



I happens every single time, you would want to execute an ICI function

and you don't know on what computer it is located. Not as unusual then,

but probably a bit unpractical.



> If we are broadcasting to a selected group then perhaps the ICSA can

> have a kind of group table that lists groups and their members such

> that you could send to a group name and the ICSA would then send

> the same message to each member of that group.



I supposed such a group table would only be in use at the central ICI

server (in the AIMetaSearch server), where the users would connect

first. Elsewhere I do not see much particualr use of such a thing, or

maybe, hm it's your choice if you want to implent it.



> And then if you wanted to send to every computer in the ICI system,

> you would want an ability to identify all those computers, put them

> in a group, and then send to that group in the previous manner.  The

> problem arises in that a local computer should be able to determine

> the particular computers it wishes to receive messages from.  There

> could be possibly be some sufficiently important ICI system messages

> that every computer should accept, but I am not sure what those

> messages would be at the moment.



Cluster formation and synchronization requires such, to mention just

one. The Console relies on a lot of crossposting to every member, those

that don't understand or cannot execute the sent, ignore it, the

remaining reply.



> My point was that we should begin working toward those applications

> in the short term that could use the large Internet connected resource we

> will soon be building up.  I have a reason to begin the CYC-like project

> after these preliminaries are out of the way, but use of a number of

> personal computers requires some agreement as to what that usage will

> be.  That is, I could not just think that my project will automatically

> utilize the resource.  An agreement will likely require an identification

> of several projects, their benefits, their implementation state, and some

> manner of resource scheduling.



I understand this aspect. But to my oppinion, we must first now too try

and utilize what has alread been done. Saying that we are begining work

on this, to me means: 

1. Indentifying the information we will asimilate, 

2. Locating the holders of the information, 

3. Contacting them, 

4. Disscussing the option of ICI integration with them, 

5. Making the linkup software and 

6. Adding their service to the ICI network.



Then once we would have enough databases linked up, creating upper-level

functions.



--



Don't feel bad about asking/telling me anything, I will always gladly

reply.



Digging for info? Try AI Meta Search:

Http://WWW.AIMetaSearch.Com



MesonAI -- If nobody else wants to do it, why shouldn't we?(TM)

Http://WWW.MesonAI.Com

