I believe that the first fully automated (pilotless) flight was done in 1968.
Of course it was preprogrammed with the course and exactly what it should do to
take off and land and so on. However, it was not completely preprogrammed
because it had to compensate for flight conditions.
Now the question is whether or not the helicopter that used machine learning to
learn for itself how to do certain stunts like flying upside down through a
small window was an example of holistic AGI or an example of preprogrammed
compensatory learning or something in between. It was obviously something more
than an example of a program that could compensate for local conditions but at
the same time it was not true AGI which could go on to do some different kind
of thing. The fact that flight maneuvers and flight controls can be adequately
described using numerical methods does not speak of a true general AGI
algorithm. (Of course it might have learned some new kinds of things which
went unnoticed.) However, the question is not one about the fringes of the
capabilities of a program it is one about designing mechanisms of intelligence
which are capable of enough generality and adequate learning to demonstrate
that it can continue to use what it has learned in different ways and in
different conceptual contexts. Because computing power is an issue this kind
of demonstration would be limited in our day. But a demonstration of adequate
generalization learning must go beyond a few tricks.
Now if Watson was combined with a method of holistic learning would it be a
super AGI program? Maybe it is possible but I think that there has to be
something more to it than just combing two different kinds of statistical
machine learning methods. I want to know about these other kinds of methods.
I feel that it is absolutely necessary for an AGI program to have a meta
awareness of what it is doing to demonstrate true general learning that would
be capable of some incremental scalability. So I am not saying that numerical
methods are absolutely inadequate but that without better designed methods to
incorporate meta awareness and conceptual integration contemporary machine
learning is inadequate as a basis for true AGI.
Jim Bromer
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
