Substance Use & Misuse, 33(2), 439–461, 1998 439 Copyright © Semeion 1998 MetaNet * : The Theory of Independent Judges Massimo Buscema, Dr. Semeion Research Center of Sciences of Communication, Viale di Val Fiorita 88, Rome, Italy THE THEORY OF INDEPENDENT JUDGES It is known that different ANNs provided with different learning equations and qualitatively and quantitatively different internal architectures can compute similar functions (Kosko, 1992). This property can have important consequences in instances of solution of classification problems. Not all ANNs codify all k classes of the problem with the same efficacy, nor are their performances’ hierarchies always linear. Whereas ANN 1 can codify class A in a more efficacious way, ANN 2 can better codify class B and so on. Under such condition it is appropriate to select a MiniMax type of analysis: choosing the ANN with less costly errors and whose successes produce more. But this is a “monarchical” logic and the ANN, selected as the “less worse”, is an imperfect “monarch”. We propose an opposing logic inspired by democratic consideration. It consists, as it were, of democratic committees constituted by laws and statute by simple, imperfect subjects, but inclined in their plurality and globalness to compensate for the weaknesses of their single participants. We have defined the scientific consequences of this hypothesis: Theory of Independent Judges (TIJ) (Buscema, 1994). In the TIJ each ANN is an expert judge of the problem which he has had to face. Its credibility is defined by its performances in the Testing and/or Validation phase. * This Network was created by Massimo Buscema in 1995.