You are correct. AI must be able to learn the problem, not just the solution. 
It has to be able to start out with minimal knowledge, & discover what it 
needs to know from scratch. That is not in dispute. So why is it that you make 
these accusations that we collectively disregard that fact? On what basis do 
you make the claim that we are unaware of it? And why do you lump current 
attempts at AI in with "classical AI"? Who said anything about having full 
knolwedge ahead of time? You're fighting against something that doesn't even 
exist.

Putting that aside for the momentm one thing you said made no sense to me: How 
does having partial knowledge make complexity drop to zero? I'm not sure what 
you mean in the first place, but if you mean that the representation of the 
system's environment has zero complexity just because it's incomplete, I would 
point to the part of the representation that *is* complete and ask, what 
happened to it's complexity?



On Aug 26, 2012 2:57 PM, Mike Tintner <[email protected]> wrote: 






Jim,

 

A crucial part of the mistake you’re making – which is 
standard here – is looking at conceptualisation (and indeed all AGI) from a 
position of “omniscience”  - FULL AND (damn near) PERFECT KNOWLEDGE – from 
the POV of s.o. who has already got to know everything there is about, say,
patterns or indeed any other category and is now merely getting his thoughts in 
order to solve the odd further problem.

 

This is how narrow AI operates -  s.o. has already worked 
out all the aspects of solving a rational problem, (like chess),  and the 
programmer has to order and organize them.

 

That’s not AGI – that’s not real life for any real world 
agent.

 

An AGI works from MINIMAL - EXTREMELY  PARTIAL, LIMITED 
KNOWLEDGE of any given subject - is s.o. who is ***getting to know the world*** 
-  who has to form concepts about the world from v. limited information – 
and “find his way around the world” as he goes along.

 

You think an AGI is merely a more complex narrow AI – an agent 
who a la T.S.P. has a complete map of the town or area – and merely has to work 
out the most efficient routes.

 

An AGI is instead s.o. like you who has to get to know every 
town or territory as he goes along – and has to be able to find his way around 
without a map.

 

An AGI is s.o. who can do a CREATIVE task about wh. he knows 
v. little to begin with – s.o.  I can ask,  : “go into that room 
you’ve never been in before, and clear it up...” You don’t start with a full 
knowledge of all the factors involved. 

 

You think AGI is no different from narrow AI – you give your 
robot beforehand a complete map of that room and every piece of furniture – 
and, 
like existing programs, he merely works out the most efficient ways of 
navigating and stacking the furniture.

 

Hey, we’ve got narrow AI’s that can do 
that.

 

AGI is about our sending robots into uncharted territory and 
having them sort things out as they go along – without our having to program 
their every step beforehand. We want robots that like living creatures can
*explore* the world from a v. limited brief and a few at most 
instructions

 

Only with FULL KNOWLEDGE of the complete set of factors 
involved in a problem, can there be complexity.  AGI always proceeds from 
MINIMAL KNOWLEDGE about a problem – and nothing like a definitive set of 
factors,  and there is no complexity.

 

As would-be AGI systembuilders here, we are all proceeding 
from v. partial knowledge of what an AGI machine will entail – zero complexity. 
Building a mechanical AGI is merely a more difficult form of walking around a 
new room or field or town, or handling a new object.

 

Rational/narrow ai problems -  full knowledge (of the set 
of factors)  & complexity. Creative/AGI problems – minimal knowledge 
(& no set of factors)  & no complexity 

 

P.S. “A fundamental assumption made by classical AI planners 
is that there is no uncertainty in the world; the planner has full knowledge of 
the conditions under which the plan will be executed.”

http://arxiv.org/pdf/cs.AI/9605106.pdf 


 

https://www.google.co.uk/search?q=A.I.+%22full+knowledge%22+%22limited+knowledge%22&sugexp=chrome,mod=4&sourceid=chrome&ie=UTF-8

 

About 11,500 results 

 


  
    
      
      AGI | Archives

 | Modify
 Your Subscription


      
    
  







-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com

Reply via email to