Re: The skill of learning skills.

On Mon, 02 Jan 2006 21:56:35 GMT, Risujin <risujin@xxxxxxxxxxx> in
comp.ai.philosophy wrote:

>Lester Zick wrote:
>>>>On Mon, 02 Jan 2006 01:08:09 GMT, Risujin <risujin@xxxxxxxxxxx> in
>>>>comp.ai.philosophy wrote:
>>>>>If we define intelligence as the skill of learning skills, we arrive at
>>>>>a gradual definition for intelligence. It is easy to see how a system
>>>>>may be capable of learning some kinds of skills but not others. The
>>>>>breadth/variety of learnable skills and their relevance to achieving the
>>>>>program's goal determines the degree of intelligence.
>>>>>
>>>>>Now we need to define a skill, skill variety, and what exactly we mean
>>>>>by learning a skill. If we think of a skill as a state operator...:
>>>>>
>>>>>* A skill is a procedure for inducing something to transition from one
>>>>>state to another. In other words, a set of instructions which the system
>>>>>is capable of following that predictably produces certain changes in the
>>>>>environment.
>>>>>
>>>>>* A skill can be of any level of coarseness. We don't necessarily need
>>>>>to specify every action in detail. We can imagine abstract skills such
>>>>>as "drive a car" and reason with them without referring to sub-skills
>>>>>("start the car").
>>>>>
>>>>>* The definition of having learned a skill is trivial in this
>>>>>case--either the system can or cannot carry out the procedure. We don't
>>>>>necessarily need to constrain how the system carries out or acquires
>>>>>procedures.
>>>>>
>>>>>* The power of a system can be measured in terms of the range of goals
>>>>>it is capable of achieving compared with the theoretical range of goals
>>>>>a system with the same inputs/outputs could possibly achieve.
>>>>>
>>>>>* The intelligence of a system can be measured in terms of the range of
>>>>>skills it is capable of acquiring compared to the range of skills
>>>>>theoretically possible for it to perform. The intelligence of a system
>>>>>is its capacity for increasing its power over the environment.
>>>>>
>>>>>Curiously, with this definition a static skill set would imply that the
>>>>>intelligence of a system decreases as its power increases. Only learning
>>>>>skills which expand the potential skill set of a system increase its
>>>>>intelligence.
>>>>>
>>>>>An intelligent system would only require processing power to match human
>>>>>intelligence for time-constrained tasks. If we limit the domain of
>>>>>problems to non-timed goals, there is no reason at least a marginally
>>>>>intelligent system could not be built on a home PC today.
>
>>>Maybe I should've made it more explicit but I wasn't actually going for
>>>a mechanical or any other kind of solution to the general intelligence
>>>problem. If you read carefully, the skills system I proposed falls far
>>>short of defining a mechanism. What I was suggesting is instead a
>>>possible measure of intelligence as an alternative to "we'll know it
>>>when we see it".
>>
>> Okay. But how is your possible measure of intelligence superior to
>> "we'll know it when we see it"? All the different schools of thought
>> seem to have their own measure of what they're looking for in smarts.
>
>The "we'll know it when we see it" definition is as bad as it gets. You
>cannot derive anything from such a definition except a yes/no answer
>with no reasoning.

Well I agree except your resolution of the paradox seems to be "we'll
know it when it gets skillful". Traveler seems to think "we'll know it
when it gets timing". Glen's seems to be "we'll know it when it gets
trainable".Dan's appears to be "we'll know it when it's neruological".
Darwin's was "we'll know it when it gets evolutionary". These all take
the resolution of the paradox a step further without solving the
paradox or even being able to say exactly what the paradox is. Besides
there are already systems with timing, trainability neurology, evolve,
and are skillful without evidence of intelligence.

>My proposed measure is not unique. It may not be the best around either,
>I'm not claiming it is. It does have some predictive power though. I
>noted some interesting consequences after the definition. It is also
>more concrete than many other definitions. I was hoping we could tighten
>it up further, or come up with something else that is concrete.

The problem here is how to tighten it up in the direction of
intelligence. We could combine all the factors cited above and
still not know what intelligence is.

>>>A mechanism for producing/conforming to the behavior I describe above is
>>>the ultimate goal here but I'm not claiming to have anything remotely
>>>close to it now.
>>
>> But there are plenty of people who do make exactly that claim. You
>> seem to disagree with them. So how does your suggestion contradict
>> what they claim?
>
>Doubtless there are system which conform to the skills measure. It
>defines and measures intelligence. If there are intelligent systems they
>will rank somehow in this measure.

Sure except the question remains as to whether and how these systems
are intelligent.

>I don't see why we are looking for contradiction here.

I thought you were against "we'll know it when we see it" approaches
to intelligence. We don't resolve that problem by replacing the term
"see" with "timing" "trainability" "neurology" "evolve" or "skillful".
The problem is the "we'll know it when it . . ." part of the equation
and not the ". . . see, train,, skillful, etc." part. Those terms are
just descriptive references to various aspect of intelligence.

In other words it's a methodological or epistemological problem and
not simply a matter of what is methodologized. Empirical epistemology
proceeds as follows: I make a claim that such and such is intelligent
and has significant predictability according to some characteristic
then defy anyone to show an example of intelligence which contradicts
my claim and that doesn't have that characteristic.

Hell I could combine all of the above characteristics into an olio of
intelligence predictability and still not have any understanding of
the thing in itself. Even Turing's "can't tell the difference" is just
an extension of "we'll know it when we see it" criterion because the
problem isn't technical; it's a problem with the empirical method.

~v~~

.

Relevant Pages

  • Re: Smullyans Quiz Problem
    ... >> The paradox here is literally that they contradict our intuition. ... "Wilde"-like paradox in this; you know, the "The only thing worse than ... as in fact the point is that you cannot communicate ...
    (sci.math)
  • Re: intrinsic energy
    ... I think what is so frustrating about you, Sam, is ... that you contradict yourself every time you open your mouth. ... (for paradox read QM) ... It's a wonderful and fascinating world, John! ...
    (sci.physics)
  • Re: chaos <=> paradox. Prove me wrong. A challenge.
    ... > SENSE AND EXPECTATIONS. ... conclusions and it would no longer "contradict our->expectations<-, ... it no longer contradicts your expectation. ... If peolpe grew accustomed to the liars paradox, ...
    (sci.math)
  • Re: set theory : the blunder
    ... they replied with (well the most sensible replies ... it seems a paradox then claiming x = ... Twentieth-Century Ideas". ...
    (sci.math)
  • Re: set theory : the blunder
    ... they replied with (well the most sensible replies) ... it seems a paradox then claiming x = ... Twentieth-Century Ideas". ...
    (sci.math)