Common Misconceptions Refuted
I was at first greatly startled to realize that the pronunciation
and liguistic mistakes that nonnative speakers of a language make
are pretty much identical. A similar thing happens with the
language of the FRDCSA, those who are new to it wind up making
certain very predictable mistakes. In fact, in exactly the same
way as mathematics students do, because our assertions are
completely mathematical. Therefore, we assert some things, with
proof, where by proof we mean a mathematical proof. It is with
this formal sense of the word that we use when we say that we are
going to refute certain mistakes that the student makes.
The program that writes a better program.
We anticipate that when many people, particularly programmers,
first start out investigating AI, they hit on the idea of writing
a program that would write more capable programs. They clearly
see the positive reinforcement of this method. But people who
have this idea are probably operating under a vague definition of
program which does not have the same properties as a the formal
definition of a program. Formally speaking, a program is simply a
total recursive function, although there are other equivalent
definitions which are useful in different theoretical fields.
What the person probably means is that the program would obtain
new information from a source, say, online information, and that
changes the picture because now our program is assumed to have
access to an oracle. But the properties of an individual program
are fixed, and if they somehow changed during execution, would
create a transitive closure violation.