Todor:

What are the definitive criteria for halting in these real world problemsolving 
activities?  
The "definitive criteria" for a given agent* are quite simple...
Er, there are none. 
I don't think so, they are rather obvious...
5."Shop on Amazon for some good new books on AI" 
It'd include opening a browser

Todor,

The important thing in understanding how intelligence works is to look at how 
you really solve actual problems - the whole problem.

Here's just one aspect of the actual problem of "shopping on Amazon etc.."

http://www.amazon.co.uk/s/ref=nb_sb_ss_c_1_23?url=search-alias%3Dstripbooks&field-keywords=artificial+intelligence&sprefix=artificial+intelligence%2Caps%2C256

Books ›  
"artificial intelligence"

Related Searches: machine learning.
      Showing 1 - 12 of 23,822 Results

      Please explain the definitive criteria/halting criteria for examining 
23,822 results - and for pursuing further results, as in "machine learning" and 
the many other terms you might create both for Amazon and searching on Google.  

      I would argue that there are actually a potential infinity of alternative 
searching methods - please explain why there are not, why we can all be 
satisfied with your single method (wh. actually doesn't specify halting 
criteria). 


Would you argue there Is one best method of searching any space/frame of 
options in computer programs?

From: Todor Arnaudov 
Sent: Saturday, June 30, 2012 2:16 PM
To: AGI 
Subject: Re: [agi] Real World Reasoning


Hi Mike,


>What are the definitive criteria for halting in these real world 
>problemsolving activities? 


The "definitive criteria" for a given agent* are quite simple. //The agent and 
his mind and experience should be defined, of course. 


>Er, there are none. 


I don't think so, they are rather obvious.


The complete details are in each one's head (and the details of the location, 
like the specific street, which also guide agent's activities, such as if 
there's a dead-end, of course the agent can't go on).
They depend on his or her entire brains and experience - that's right.


Also - brain has several tangled subsystem, which are confusing each other and 
has different priorities, different "memories, different representations, and 
different "rationality", which is converged to "one"  by the integrity of the 
body. Neocortex doesn't understand precisely what's going on below, and 
different parts of neocortex don't understand precisely why they receive 
signals that they recive.


The "formula" for definining each of this is big and has many "latent" 
variables which are known only for the internals of the control-causality unit.


I have discussed this in the references I posted in the previous message.


There you are an example for one of your cases (would take too much typing for 
all):


>5."Shop on Amazon for some good new books on AI" 


It'd include opening a browser, seeing the cursor pointing an icon, feeling a 
click, seeing the windows was open;
If the browser was already running - locate the browser on the taskbar, or 
point the cursor on the address bar or search bar.


In those steps, "locating" the browser is reducible down to mere matching for 
memories and current experience.


Then typing a search query or tha address or a bookmark (again, criteria are 
simple: have an intent to type the address or invoke particular web site, then 
see the sensory input matching the intent).


Then type the query "AI" or "AI books", see the list. Browse the list. Click on 
the top pages. Click on "most popular pages".


In this part one of the "halting criteria" are again:


"Typing" means intending to type a word/key. "Intentions" is a set of planned 
matches of inputs and experience.


Typing a key means to match and to coordinate/juxtapose current sensory-motor 
experience of moves/touch/vison with past experience (intentions - records), 
which confirm the successfulness of execution of each other.


Criticism: "Nooo... We all the time do something new, it's all different"


Answer: Not really. All humans do is adjusting the coordinates of the body with 
vectors of motion. Grasping, rotating hand, stretching an arm, back;rolling 
eyes, open a mouth, move the tongue so and so, which is coupled with speed and 
sequencing all the moves, and adjusting the moves depending on sensory 
experience. That's all we do, and all we can do, 

Back to the keyboard sequence:


If one tries to hit a key there's no touch sensation and the hand "dives"down 
and doesn't meet something to stop it, that's an indication that's the move was 
wrongly conducted.
The one knows that it's wrong very easily - this experience doesn't match the 
intended one, in previous times hitting a key comes together with feeling a 
key, particular sound, particular tactile sensation, particular proprioceptive 
sensation (the finger stops moving, even tough there's force applied on it to 
press down), the image on the screen is changing as intented (an intended 
symbol appears - I won't dive here into the symbols).


If the one is a touch typer, she would turn her sight down to the keyboard to 
see what's wrong in the coordinates of her hand, and the relative coordinates 
of her hand to the keyboard.
Then when getting the location of the keyboard and the hand, would reissue 
coordinated motion with intention to hit a particular button, and also may 
include visual monitoring in this case to assure the trajectory of finger tip 
reaches to plane of the key. The "click" sound, the appearance of a symbol on 
the screen etc.


If there's a list with books - if one's searching for "AI books", she's 
supposed to know what an "AI Book" or a "Book" looks like on Amazon, such as - 
a picture of the cover, a title (text which has larger font than the rest of 
the text in the section; a text which is next to the picture of the cover; a 
picture of a cover is a picture with a title, which matches the title which is 
aside in the closest region;...  etc. - there are many details.)


Then clicking on some of the titles - that means, locating the cursor on the 
title, the graphical image of the cursor to change its coordinates in the 
visual field so that its arrow/active point touches the pixels of the target 
title.


Then clicking, which means pressing a particular mouse button, which generally 
means to apply force to that button which is perpendicular to the plane on 
which the mouse lays.


Then the web page on the browser should change - if it doesn't change, that 
means the click was not completed or there was something wrong.


There should be so and so image, contents, etc...


(...)


There's one brain-induced criteria for halting such searching queries -- the 
time that your conscious can focus on this activity.
Also, the moment when mind decides that "that's not searching anymore, now I'm 
reading the book".


That requires the conscious, the high levels of causality-control to set a 
variable saying "OK, I'm ready now, let's go reading this great book I found!"


Yes, this is partially arbitrary, but it's so because it's not essential - the 
definition of "reading", "searching", etc. are high level and can have 
arbitrary or non-essential components.
The labels of the activities don't matter. What matters is the content, like 
what I explained above with the sensory-motor matching - it goes for any kind 
of activities and tasks, just have to change the sensations, modalities, 
particular records, and the sequences of those.




>You can and will set up criteria - but they will be arbitrary (if probably 
>reasonable), they will be imprecise, they will conflict with other criteria, 
>and you will change or modify them tomorrow - 
>as you have your whole life. (And there are no halting criteria for arriving 
>at new criteria).


Yes, they may be changing, and there's nothing wrong with it. The essential 
(sensory-motor matching) is not changing.
And yes, some of the criteria are "?motional", depend on current feelings, 
current mood, current or recent social interaction etc. etc.


There are halting criteria for new criteria, when current fail/halt too much 
(for a given limit - it's defined in the internals of the evaluator), or when 
they bore you, or when they work too well, etc.

>Real world activities are a matter of creative, artistic judgment, not 
>rational, scientific judgment. 


In the articles/works I linked last time there are discussions on "Rational", 
also "simple" or "intentional". "Rational" is not just something dull, 
extremely simple written in a few letters or a short formula, and it's not 
something that's obvious all the times, that doesn't require context etc.


The "irrational" human behavior is a behavior that the observant doesn't 
understand. For example phobias. A little child had a stressful experience 
about something, and then she fears it when something reminds her of it. 
There's NOTHING irrational here, for the brain subsystem that conducts it this 
is a correct behavior, aimed to force the child to avoid the threat. It's 
neocortex and brain's fault that neocortex fails to understand it (and may need 
medical help) - it receives signals for fear, which means "danger", so it makes 
body to run away.


It's "irrational" namely if somebody looks at the problem more abstractly that 
the problem is posed.


I also don't agree with the common opinion that "rational"and "scientific" 
means  not from the real world.


Science is based on the real world, it's derived from it, it's justyfied by 
experience and aimed to predict the real world and to model it as intended 
(predicted) - it does pretty well and scalable the job.


>We can go through the vast list of human real world activities, one by one, 
>and you will find 
>no "formulae", no [step-by-step] rules anywhere in the culture/society. People 
>may tell you they have the "formula" for success in these activities, but it 
>won't be what any AI-er would 
>recognize as a formula or algorithm  - just vague, general prescriptive 
>principles.

I don't think so, the core principles are straight: to match intended to 
perceived (that's the criteria for success). 
"Intended" itself are records of perceived, that's part of the real world 
grounding.


 >Real world reasoning and real world activities are of a whole different 
 >REALM/WORLD to the rational problemsolving of narrow AI, logic, maths etc.


Science and technology are exactly real world problem solving, and they have 
systematically developed and applied methodologies.




-- Todor "Tosh" Arnaudov
http://research.twenkid.com
http://artificial-mind.blogspot.com





  a.. From: "Mike Tintner" <[email protected]> 
  b.. To: <[email protected]> 
  c.. Subject: Re: [agi] Real World Reasoning 
  d.. Date: Tue, 19 Jun 2012 11:11:00 +0100 


Todor:There are always criteria for halting

Problem Activities:

1."Walk down your local high street."

2."Let's have a conversation about AGI".

3."Have sex with your girlfriend."

4. Discuss: "Boris has delusions of grandeur."

5."Shop on Amazon for some good new books on AI"

6."Write a post in reply to this one."

7."Read this morning's newspaper."

What are the definitive criteria for halting in these real world problemsolving 
activities?

Er, there are none.

You can and will set up criteria - but they will be arbitrary (if probably 
reasonable), they will be imprecise, they will conflict with other criteria, 
and you will change or modify them tomorrow - as you have your whole life. (And 
there are no halting criteria for arriving at new criteria).

Real world activities are a matter of creative, artistic judgment, not 
rational, scientific judgment. We can go through the vast list of human real 
world activities, one by one, and you will find no "formulae", no 
[step-by-step] rules anywhere in the culture/society. People may tell you they 
have the "formula" for success in these activities, but it won't be what any 
AI-er would recognize as a formula or algorithm  - just vague, general 
prescriptive principles.

Real world reasoning and real world activities are of a whole different 
REALM/WORLD to the rational problemsolving of narrow AI, logic, maths etc.

The latter is deterministic, structured activity and problemsolving in 
artificial environments, the real world is free, unstructured activity and 
problemsolving in real, unstructured environments (and is v. commonly described 
in such terms).

The latter is where s.o. decided "*this* is the way we are going to 
count/calculate/spell/build a Lego house/locomote through a warehouse or a rail 
network/think about a problem... - and that's the 'right' and only way to do 
it"  The real world is where there are multiple ways to count/calculate or do 
anything else - and potentially an infinity of ways, incl. new and better ways.

Welcome to the real world.

P.S. Patchworks are a formalisation of real world, creative activities and 
products -  look again at those patchworks, and you will see there are no 
criteria for determining what parts, structure or form your next patchwork in 
the series should take (unlike your next variation on an existing pattern). 
Patchworks are free - free form, free structure, free parts - within loose 
constraints or boundaries.  Each new patchwork in a collection can have 
multiple and potentially infinite new forms,structures and parts. There are no 
criteria for halting in producing them - or indeed in writing a program or any 
work of art.




-- 

-- Todor "Tosh" Arnaudov
http://research.twenkid.com
http://artificial-mind.blogspot.com


      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