Jim, my thoughts are not easy to convey. I've tried not to ramble, and
this response is shorter than I might have liked to make it.
First, nothing is easy. To understand my view on why complexity is not
a problem initially, consider that the AGI has a simple program to perform:
Loop
pick the best thing to do of those things you can do.
execute for a few cycles, what was found to be best at the last check
go to loop
That's a simple program, but implementation has it's surprises. Notice
that if the program doesn't have many options for "doing" (not much
coded) then picking the best is pretty simple. As the unit has more in
it's kit of stuff to do, then it starts having to make a choice. This
making of a choice is where it gets complicated, but maybe not so much...
Imagine a robot that has only a few moving parts. What is it that it
can do initially? Roll from point A to point B; tilt it's video
camera; move it's robotic arm; rotate the "hand"; Squeeze or release
the simple gripper... You get the picture. At a given moment the
robot picks what it considers it's best movement. How does it pick?
That's the interesting part. And, we know that it will involve rules -
even if it is only following a sequence (which is rule following with an
implicit rule that the next step rules.) I hear the objection that the
robot can't pick a behavior - but that is the difference between a dumb
robot and an intelligent one.
The robot mentioned above is the "simple" start, but it is not adequate
to start adopting new rules. Nothing in its set of behaviors has the
power to formulate a new "unit" of rules to be adopted. This would take
a whole complex set of rules to give the robot the means to start
recognizing what is worth adopting and then knowing where and how to
store these new adopted rules (insights, behaviors, triggers...)
Jim, you asked
"Do you think that finding and incorporating new information which
would allow the program to know what it should do with the rules it had
collected would be easy? "
and my answer is that it is NOT EASY, but that is the "heart" of
building the AGI. What I don't get is why we do not spend more time
discussing an overall architecture that accomplishes these needed
functions?
I would consider it progress if we even discussed a small part of this
architecture. For example, one technique of the AGI might be to "find"
through the process of submitting requests through a mailing list. e.g.
"what meta-data will help me figure out what to do with a rule?"
It becomes obvious pretty quickly why the bootstrap of the AGI is so
daunting.
Stan
On 12/03/2015 04:02 PM, Jim Bromer wrote:
I have not really gotten past the coding problem either. If I could
work 10 hour days then that would not be a problem. However, I am not
sure why you would disagree with the theory that AGI is a complexity
problem. Yes the rules could come from anywhere but you mention that
as if the integration of a rule would be easy to recognize and easy to
integrate in previously acquired knowledge. If that were so then we
would not have a coding problem to overcome - we could readily
demonstrate how easy it is with limited problems and making our code
more sophisticated would be more like an ordinary programming project.
We start off with something simple and then just try to make it
better. Do you disagree with this? Do you think that finding and
incorporating new information which would allow the program to know
what it should do with the rules it had collected would be easy? I
don't understand that point of view.
Jim Bromer
On Thu, Dec 3, 2015 at 11:17 AM, Stanley Nilsen <[email protected]> wrote:
On 12/02/2015 08:45 PM, Jim Bromer wrote:
...
Once you
begin analyzing a statement that is important to you the
discontinuities of the implied expectations can really jump out. That
is an example of logic but it is not an example of Boolean Logic but
of applied logic, the application of logic to a situation in which the
possibilities belong to categories that can stand out in stark relief.
Even if the possibilities (of the expectations) are not logically
rigid, logical systems can still be formed to explore the
possibilities.
Jim Bromer
Sounds like the description of applied logic is very much like the "frames"
approach that Minsky discusses in his book Society of the Mind. In essence
you have expectations / rules that are appropriate for the situation and use
those rules to help with discernment. The "story lines" act more like
triggers that are matched up to the frame knowledge that you already have.
As I understand it, character recognition has improved because the "frame"
for where words belong has improved - there are rules that increase the
likelihood that a character will be one of ...
I've not looked at the code, or the technology, but seems like this has been
discussed before on this list. Would have been interesting work to be
involved in OCR over the years.
Frames brings up the point that I wanted to make about "adoption." One
invests time and uses resources to build frames - become knowledgeable. It
isn't that you have to invent the rules, you just need to recognize them and
incorporate them. Rules could come from several sources including a
question and answer session with a knowledgeable person. I believe that the
intelligence is not in what you know, but that you know enough to seek out a
source, inquire in the right way and then know what to do with the rules
that you collect. One issue with AGI is finding the best way to format the
collection so it can be used in an interesting way. Another issue is getting
started in the right direction - going for frames that can help build other
frames.
The show stopper for me is that there is a bunch of coding to be done and
the progress will be slow.
Stan
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/24379807-653794b5
Modify Your Subscription:
https://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/9320387-ea529a81
Modify Your Subscription: https://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
