Mark: In an AGI system, the same collection of "active knowledge" has
to simultaneously and continually be modified by a large number of
different cognitive processes. These processes all have to play
together nicely as they are acting on the same data at the same time,
and need to benefit from each others' intelligence in order to make
cognition happen. Therefore, the parameters of all the cognitive
processes need to be "tuned together" -- a tuning that will work for
one process in isolation will not necessarily work for that process
when it acts in the context of other processes...
This issue existed in Webmind, and exists in Novamente, but in
Novamente we have taken specific steps to palliate it and keep the
issue under control...
Ben
On 6/14/06, Mark Waser <[EMAIL PROTECTED]> wrote:
> My big issue is that the system depends on laborious experimentation to
> find stable configurations of local parameters that will get all these
> processes to happen at once.
> the problem is doing that
> whilst simultaneously getting the same mechanisms to handle 30 or 40 other
> cognitive processes.
I'm confused . . . . Why not have multiple independent instances of the same
mechanisms with different local parameters for different processes? Once
you uncouple the local parameters from instance to instance, making all the
processes happen at once should be no more complicated than making them
happen individually in isolation. Or am I misunderstanding you or missing
something really dumb?
----- Original Message -----
From: <"[EMAIL PROTECTED]"@pop.lightlink.com>
To: <[email protected]>
Sent: Wednesday, June 14, 2006 9:05 AM
Subject: Re: [agi] How the Brain Represents Abstract Knowledge
> Russell Wallace wrote:
>> On 6/14/06, "[EMAIL PROTECTED]
>> <mailto:[EMAIL PROTECTED]>"@pop.lightlink.com <http://pop.lightlink.com>
>> ** <" [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>"@pop.lightlink.com
>> <http://pop.lightlink.com>> wrote:
>>
>> Russell Wallace wrote:
>> > Has anyone yet made an artificial NN or anything like one handle
>> syntax?
>>
>> Uhhh: did you read my first post on this thread?
>>
>>
>> Yes; you appear to be saying that as far as you know nobody has yet made
>> NNs or similar do syntax, but that's because they went off into the dead
>> end of back propagation, and you believe it should be possible to create
>> something like NNs that do syntax and other such things, but you haven't
>> yet implemented any such. Do I understand you correctly?
>
> Not correct. What I said was that I have not written it up yet. I have
> implemented enough of it to show (to my satisfaction) that it can handle
> syntax. More importantly, I have developed a formalism (more precisely a
> theoretical framework) that shows how to use this NN-like system to handle
> a very large array of other cognitive processes, not just syntax.
>
> My big issue is that the system depends on laborious experimentation to
> find stable configurations of local parameters that will get all these
> processes to happen at once. I believe that this has to be done
> empirically, so I am constructing a development environment in order to
> facilitate that empirical process.
>
> I have hesitated to publish the formalism ahead of time because (among
> other things) it depends on a methodological and philosophical approach
> that is so much at odds with the status quo that I have no desire (or at
> least, not much desire... see my contribution to Ben's AGIRI workshop
> ;-) )
> to fight religious battles without the experimental data to back it up.
>
> I say all this because it would be extremely misleading to say that I
> simply "believe it should be possible to create something like NNs that do
> syntax and other such things". That makes it sound like I have nothing
> but
> a vague feeling that it ought to work. Getting a molecular system, like
> the one I have developed, to handle syntax and yet also have the benefits
> of an NN design is relatively straightforward: the problem is doing that
> whilst simultaneously getting the same mechanisms to handle 30 or 40 other
> cognitive processes.
>
> Richard
> Loosemore.
>
> ---------------------------------------------
> This message was sent using Endymion MailMan.
> http://www.endymion.com/products/mailman/
>
>
> -------
> To unsubscribe, change your address, or temporarily deactivate your
> subscription,
> please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
>
-------
To unsubscribe, change your address, or temporarily deactivate your
subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
