🚀 go-pugleaf

zarnuk@caen.UUCP (Paul Steven Mccarthy) writes:
>I argue that we can make machines perform "more intelligently" than
>they currently do.  Furthermore, I believe that the fruits of this
>pursuit have value for the human species regardless of whether or not 
>we can emulate human thinking/behavior.  Finally, I posit that we will 
>make faster progress in this field if we shy away from emotionally 
>(philosophically) charged terms like "conciousness" and "understanding", 
>in favor of focusing on specific, identifyable behaviors that we decide 
>are "more intelligent" than the way that machines currently operate.

The practical viewpoint.  Generally speaking, I feel this way as well, but am
also interested in whether or not we can emulate human thinking.  But, in 
terms of the rest of this posting, that is a side issue.

[Note:  I am rearranging parts of this posting so as to focus my discussion.]
>An "intelligent" machine must be able to:
>
>        - Perceive its environment.  (Inputs)
>        - Actively acquire new information, through both 
>          external sources -- asking questions, consulting
>          references -- and through constructing, performing 
>          and evaluating physical or intellectual experiments.
>	  (Query, Experiment)
>        - Passively acquire new information through observation
>	  of its environment. (Observe) 
>        - Exchange information with other "intelligent" machines.
>	  (Communicate)
>	- Alter its environment.  (Outputs)
>        - Recognize the affects of its actions on its
>	  environment.  (Feedback)

This is what you might call the I/O, or sensory package.  An interesting point
is that this is contrary to the Greek mode of reasoning in a vacumn; that is,
all that was necessary was a mind which could observe and postulate, and no
experimentation was deemed necessary.  There is also the issue of what is 
internal and what is external for a computer.  For example, assume that the
sum total of human knowledge is encodable in, say, an encyclopedia.  (This
assumption, for historical reasons, is called the metaphysical assumption.
More on this later.)  Then the computer has no need to consult outside sources;
it can experiment and reason about the results, add these to its data storage,
and continue.  Communication is not really necessary either; it only serves to
prove to the satisfaction of other intelligent entities that the computer is
intelligent.  A monk with a vow of silence, in isolation, is no less smart or
intelligent by not being able to communicate.

Feedback is, of course, key to the learning process, and this is one of my
main tests of intelligence.

>        - Recognize itself as a distinguishable element
>          of its environment.  (Self awareness)

I would argue that this is not necessarily the case.  There are religions which
preach that every human is part of some cosmic force, and that we are all one
(in various senses); does this mean that adherents to those religions are
not intelligent?  In fact, not recognizing that oneself is part and parcel of
the environment is unintelligent; man does this by refusing to see the damage
oil spills, strip mining, etc does to ecology.  Not attaching a special value
to oneself as opposed to everything else is rare in a human, but not unknown.

>	- Use existing information and patterns of information  
>	  to form opinions about propositions that cannot be, or
>	  have not been, empirically resolved.  (Swag)

This is just forming hypotheses, and can be lumped in with experimentation and
queries.

>        - Recognize patterns in information to synthesize new 
>          information.  (Inductive reasoning).
>        - Recognize similarities in patterns of information 
>	  from different domains.  (Draw analogies).
>	- Recognize differences in patterns of information.
>	  (Make distinctions).

Here's the kicker.  I call this the word jumble problem.  If you give a human
a bunch of letters in random order, and ask him to form words from that set
of letters, he will NOT arrange them in every possible configuration; he will
tend to only look at arrangements which "look" right.  Simulating this type of
behavior in a computer is a very difficult problem, since computers tend not
to attach meaning to one random group of symbols over another.  Douglas 
Hofstater has actually been working on this problem in the past; I don't know
what the current state of such work is.


>        - Occasionally abandon formal reasoning methods to 
>	  simply explore patterns in the information at its
>	  disposal.  (Dreams? Creativity?)

Creativity can be modeled as the act of forming hypothesis and applying the 
various reasoning methods to them.  I think what you are getting at here is
doing so to no particular purpose i.e. "for fun".  I would argue that's not
necessary for intelligence either.  Consider a hypothetical professor who 
does research 24 hours a day, and only stops to eat.  Is he/she non-intelligent
or is there something assumed about his/her behavior?

>	- Weigh probabilities of various outcomes and their
>	  implications to choose a course of action.  (Guess)

I have a problem with labeling this guessing.  What you discuss falls under the
heading of "If you don't know, take the best choice."  If one course of action
has a better chance to succeed, but is not certain, take that course.  This
is merely reasoning with probabilities, which most people tend to do at a
subconcious level i.e. if I run across the street, the chances are good I won't
get hit by this car.  Reasoning with probabilities is actually not very
different from reasoning about certainties.

>(Just your average, opinionated American S.O.B.)
>---Paul...


-- 
Subrata K. Sircar, Prophet & Charter Member of SPAMIT(tm)
sksircar@phoenix.princeton.edu       SKSIRCAR@PUCC.BITNET
"I don't want the world. I just want your half." -
They Might Be Giants (Ana Ng)