🚀 go-pugleaf

In article <1993Jan21.175245.28036@ensmp.fr> you write:
>For the last time, I'm willing to say something on the subject,
>but I swear, it's the last time (oups, should I really ?)
>
>In article <9301201813.AA05004@TIS.COM>, richard@TIS.COM (Richard Clark) writes:
>
>[stuff deleted]
>
>|> 	AI is the concept of allowing the program to infer its own set of
>|> conclusions based on both input and output data along with its current
>|> set of "rules" which it may then decide to modify, allowing it to then
>|> reach a different conclusion, which then may cause further modification
>|> of its "rules", etc, etc, etc... In other words; self-modifying code.
>
>[stuff deleted]
>
>|> 	There is *MUCH* more to it... and I've left out 99.9999% of it
>|> for simplicity's sake... but I hope you get the general idea... 
>
>Sorry, I can't say I find AI really different from Expert System. Well, if you
>want, it's rather working on itself than on outer data. But then ? It cannot
>really do things it wasn't meant to do: you have to grogram it to change itself
>according to data. Sorry, but I don't feel I can get too enthousiastic about it,
>really. Or it would sound just like an old Star Trek movie, really too old to
>run for any modern movie award.

	Not quite. The program is given a *goal*. The goal usually being
positive outside reinforcement. (Or positive *internal* reinforcement, with
satisfaction of that reinforcement acheiveable only through outside stimuli)
The program works towards that goal by
modifying *itself*. One DOES NOT program it with ANY specific data/behaviour
in mind. In fact even the goal acheivement programming is left quite vague.
Working with a neural network, self-modification becomes trivial, since the
behaviours of each cell change as their data changes, automaticaly. And while
each cell may have a specific goal, and change its behaviour according to
that specific goal in relation to itself, the overall mass of cells may
behave quite dramaticaly different.

	There is a *HUGE* difference in simply "learning" and "learning to
learn". The true AI program is able to "learn to learn" by its own self
modification. In other words, it becomes more intelligent by itself as it
learns learning techniques and creates within itself the ability to *think*.

>
>Besides, many modern Expert Systems want to learn (fix their own rules first)
>before practise. It's not let to evolve all the time because programmers think
>their might, were they always encounter the same errors, end up being biased
>towards errors, and not acomplish their jobs properly. Much like an old dog
>recovering from its training and attacking its masters.
>
>To change one's rules is to change one altogether.
>

	YES! EXACTLY! IT CHANGES ITSELF! THIS IS AN ABSOLUTE NECESSITY!

	There is also another *huge* difference when outputs are wrapped
around and fed back in as inputs. This allows the program to see the 
consequences of its actions and modify itself accordingly in order to
acheive its goal.

>All right, this was my contribution everybody. I quit now.
>
>-- 
>
>	Philippe Dobias
>
>	Ecole Nationale Supe'rieure des Mines de Paris.
>	60, Boulevard St Michel
>	75006 Paris, France.

	Like I said... You are confusing "Expert Systems" with true AI.
In an Expert system, one is attempting to create an "adult", complete with
all the knowledge insights and abilities of an "adult". In AI, the objective
is to create an "infant" that will grow...


				Richard "wish I had the time" Clark.



-----------------------Relativity Schmelativity-----------------------------
 Richard H. Clark				My opinions are my own, and
 LUNATIK - watch for me on the road...		ought to be yours, but under
 It's not my fault... I voted PEROT!		no circumstances are they
 richard@tis.com				those of my company...
-----------------------------------------------------------------------------