[ExI] AI & education

[spike]

> ...On Behalf Of Adrian Tymes
> Subject: [ExI] AI & education
> 
> --- On Thu, 4/1/10, spike <spike66 at att.net> wrote:
> > We have worked our way into a
> > situation where all
> > the classroom material must be broken down into testable 
> skills, so it 
> > encourages memorization of algorithms, while functionally 
> discouraging 
> > actual thought.
> 
> A thought: for decades (possibly over a century by now), 
> people have redefined the boundaries of what counts as 
> "artificial intelligence", primarily based on what computers 
> were able to do...

Adrian, good to see you posting here again bud, we missed you for a long
time.

Ja, I see I should probably have used instead of the term "thought" rather
"creation of insight."  I agree with your examples: we are accustomed to the
idea that thought is some necessarily mysterious thing that occurs in the
brain.  If it can be reduced to a closed form solution, a system of
equations or an algorithm, we should be able to generate and catalog every
possible thought, at which time there would be no new insight and all
problems would be theoretically solved.

> ...
> 
> That point aside - without those tests, how do we know how 
> effective a teacher (or school) is? ...

Dunno.  I am surely not the first one to realize that in some cases the way
things are being taught is doing more harm that good.  That being said, I
was presumptuous as hell to comment about the 95%ile confidence criterion.
My own statistics education is now tragically 30 years old, so it is likely
that this problem has been fixed that by now.

Please have we any college students here, or recent-ish grads who can tell
us, do the statistics profs and texts still give the 95%ile criterion for
statistical significance?

Here's a parting shot for your entertainment.  A colleague noticed a
correllation between two variables and started raising alarms that they were
cause and effect related.  Reasoning: the correllation coefficient between
them was about 96%.  It wasn't a statistical significance of 96%, but rather
the correllation coefficient.  Since that was greater than 95%, surely this
signal meant something important.  (!)

I ended up pointing out there are *plenty* of measurands that have
correllation coefficients greater than 95%, plenty, and we already know they
are not cause and effect related.  A textbook example would be whiskey sales
and the number of Baptist ministers in the state of Taxifornia.  Those
correllate better than 99%, but that doesn't mean that the ministers are
buying all that whiskey or that the whiskey is somehow causing the ministers
to pop into existence.  Another example is the one from my own misspent
youth, the 99% correllation coefficient between the number of telephone
poles and lung cancer cases in Florida.  This caused the theory to arise
that the poles, possibly the creosote in the poles, was causing lung cancer.
Eventually it was shown that both are an effect of a common cause: the
steadily increaing number of lungs in that state were driving both the lung
cancer cases and the number of telephone poles.

I eventually showed that the 96% correllation in my calleague's observation
were two effects of a common cause, but not until after much money and time
had been wasted for nothing, all because she convinced a lot of people that
a correllation coefficent of 96% is important.

spike