Re: AI project & concepts
- From: "JGCASEY" <jgkjcasey@xxxxxxxxxxxx>
- Date: 3 Nov 2005 22:10:56 -0800
humiguel@xxxxxxx wrote:
....
> As I understand it, concepts can form
> hierarchies where complex concepts are
> built from simpler ones. This is the
> reason why you cannot teach Quantum
> Mechanics to a small child. The number
> of levels of the hierarchy associated
> with a concept determines its complexity.
That indeed makes sense to me. Of course
in science and mathematics you have to
start with a set of undefined terms that
correspond with a concept from which all
other terms are defined to avoid circular
definitions.
> In my view, intelligence is a tool
> devised by evolution to help living
> organisms to cope with the complexity
> (as in there's a huge amount of variables
> out there) of their natural environment.
> In all situations where the number of
> input variables is limited and you can
> write a program specification stating
> what the program should do for every
> combination of their values, you can and
> should use the GOFAI model because it is
> very accurate and efficient in terms of
> computer resources.
But GOFAI programs can use concepts?
For example a blob that is a alphanumeric
character is reduced to a set of properties
(measurements) that is then used to match
with a particular class of "objects".
> When you need to deal with the natural
> environment (or a substantial subset
> thereof like for instance language) you
> need something else.
Verbal concepts I assume would sit on a
nonverbal framework. Infants and pigeons
can give unique responses to classes of
objects such as trees, cats and so on
without language. For that reason I would
avoid starting with language in trying
to understand intelligence or concept
formation.
Concept formation involves discovering
attributes and the rules as to how they
are combined. I suspect that we are born
with the ability to extract useful primary
attributes.
It seems to me that a key issue in AI and
the natural environment is how to extract
the primary attributes from the sensory
inputs as the "axioms" on which to build.
Behaviorists saw concept learning as best
understood as an instance of discrimination
learning. Cognitive psychology relied more
on introspection evaluated against experimental
data. Cognitive psychologists see concept
learning as a process of testing hypotheses
for what the classification rules may be.
As far as I know a child starts with broad
concepts so a horse might be classified as
a big dog, or, if a horse was the first object
seen and labeled, then a dog might be
classified as a little horse :)
It doesn't take much effort to recognize
different languages even if you can't pick
out the words let alone understand them.
I don't speak French, German, Arabic or
any Asian language but I can recognize
them as belonging to one of those groups
when I hear them.
So I think concept formation starts from
the general to the specific.
Is the above how you understand it?
John Casey
.
- Follow-Ups:
- Re: AI project & concepts
- From: humiguel
- Re: AI project & concepts
- References:
- Re: AI project & concepts
- From: humiguel
- Re: AI project & concepts
- Prev by Date: Re: Same Old, same Old
- Next by Date: Re: Temporal Learning
- Previous by thread: Re: AI project & concepts
- Next by thread: Re: AI project & concepts
- Index(es):
Relevant Pages
|