The FRDCSA is a project which works towards the creation of AI through practical methods.  AI is the theory of building intelligently behaving artificial systems.  Strong AI (SAI) holds that human intelligence is Turing equivalent and therefore computer programs which fully model human intelligence are possible.  Weak AI (WAI), on the other hand, asserts that humans are not Turing equivalent and that therefore we may be concerned with writing increasingly intelligent programs, but these must fall short of human intelligence. 

We  take the  reasonable view  that both  SAI and  WAI  contain an unproven  assertion, namely,  that which  decides the  question of human/computer  equivalence.  Therefore,  we consider  a resticted theory of AI  we designate AI-, the intersection  of the axioms of SAI and WAI.

Reasoning  in  this  manner,  we  attempt  to  create  intelligent programs.   The  most obvious  approach  to  bootstrap  AI is  the concept  of seed  AI.  This  is the  idea that  the  AI researcher requires  only to  write  a  simple program  which,  by virtue  of unprecidented  historical   circumstances  and  the   authors  own superior design, is able to grow to be increasingly intelligent.

While   in  and  of   itself  possible,   it  relies   on  certain environmental  assumptions  which  are  not known  to  exist,  and therefore does not succeed with certainty.  To see the flaw in the crude argument for seed AI, it is best to consider the restatement "I am going to write a program that writes a smarter program."

When this idea becomes formalized,  it is equivalent to this: I am going  to write a  program A  that writes  another program  B that solves problems (aka. proves theorems) that A cannot.

This  however is what  is called  a transitive  closure violation. Suppose A cannot prove Phi, yet A writes B and B proves Phi.  This is equivalent to A proves Phi (transitively).  Contradiction.

While this  proof sketch is  not convincing outside of  the proper setting of  recursion theory, hopefully  it conveys the  idea.  In the  proper setting,  it is  evident that  B is  only part  of the intermediate execution of A. 

Most  new AI  researchers, who  understand  the goal  but not  the means, will attempt to continue to justify the concept.  They will resort to having an oracle  O which provides information to A, and thus A  is not affected  in this way.   But this idea is  simply a relativization of the same fundamental error.  In other words, one has  only to  bound  the size  of  the oracle  input  and then  to demonstrate that  the totality  of any additional  problems solved must be provided by the environment.